# Akholi > Akholi publishes primary research on outsourcing, service engagements, global capability centers, and agentic AI. Its 2026 study by Phil Hatch examines 150 agentic AI outsourcing engagements at Tier 1 financial institutions through evidence from 850 vendor delivery-team respondents. The study comprises an Introduction, Parts 1–6, Appendices, 216 companion analyses, and nine public PDF downloads. Findings describe this nonrepresentative sample and do not establish causation or legal breach. ## Complete resource directory - [Akholi](): - [Agentic AI Outsourcing at Tier 1 Financial Institutions](): Research on 150 agentic AI outsourcing engagements: outcomes, governance, manual alteration, workforce conditions, and proposed controls. Phil Hatch, Akholi. - [Academy](): - [Practice](): - [The Practitioner's Toolkit](): - [Contact](): - [Privacy Policy](): - [Terms](): - [Outsourcing Performance Tuning](): - [Outsourcing Preparation](): - [Global Capability Center Planning](): - [Global Capability Center Performance Tuning](): - [Agentic-AI Governance, Oversight & Risk](): - [Crisis Response](): - [Outsourcing Lifecycle Management Foundation](): - [Governance, Oversight & Risk Master Class](): - [Outsourcing Performance Optimization Master Class](): - [Book](): - [About Akholi](): Akholi publishes primary research and advises on outsourcing, agentic AI, service engagements, and global capability centers. Phil Hatch is its CEO. - [Retained Advisory](): - [Accessibility](): - [The Executive Action Plan](): Proposed institutional controls and a staged first-year implementation schedule. A volume of the 2026 Akholi study. - [Governance, Oversight, and Control](): How delivery conditions affect governance, oversight, and control. A volume of the 2026 Akholi study. - [The Study and How to Read It](): Study design, evidence, definitions, and reading conventions. A volume of the 2026 Akholi study. - [Engagement Outcomes and Their Correlates](): Engagement maturity, performance, cancellation, and associated structural conditions. A volume of the 2026 Akholi study. - [The Delivery Workforce](): Experience, management, satisfaction, trust, and knowledge at risk of departure. A volume of the 2026 Akholi study. - [Manual Alteration by the Delivery Team](): The work altered by delivery teams, its disclosure, and its visibility. A volume of the 2026 Akholi study. - [Definitions, Indicators, Methods, and Disclosures](): Definitions, indicator and exhibit inventories, methods, companion analyses, and disclosures. A volume of the 2026 Akholi study. - [The Five Failure Categories](): Separate indicator analysis across the five failure categories. A volume of the 2026 Akholi study. - [Study Data — The Underlying Data](): Search 216 companion analyses supporting the 2026 agentic AI outsourcing study, including indicator definitions, distributions, segments, and outcome associations. - [About Phil Hatch](): Phil Hatch is Akholi’s CEO and the author of the 2026 study of agentic AI outsourcing at Tier 1 financial institutions. - [Reported return on the engagement: ROI %](): Respondents reported a return on 17 of the 30 engagements in production. Among the engagements with a reported return, the return on the engagement had a mean close to its median \(mean 24.7, n = 17 engagements, median 23.0\). The rule printed in Analysis Results recorded all 10 canceled engagements in production at zero return. Active engagements in production \(n = 20 engagements\) had a mean return \(mean 21.0\), and 3 of them were recorded at zero. - [The three remaining measures among the engagements not canceled: 94 active engagements](): The three remaining measures moved together on the 94 active engagements. Higher customer satisfaction came with a higher engagement indicator mean \(r = 0.84, 95% CI 0.77 to 0.89\). Customer satisfaction fell as intent to leave rose \(r = −0.73, 95% CI −0.81 to −0.62\). A higher engagement indicator mean came with lower intent to leave \(r = −0.69, 95% CI −0.78 to −0.57\). Cancellation likelihood, as respondents rated it, rose as customer satisfaction fell \(r = −0.66, 95% CI −0.76 to −0.52\). Of the 94 rated active engagements, 19 were at high and 39 at moderate cancellation likelihood. - [Shared structure across the four study measures: cancellation, customer satisfaction, engagement indicator mean, intent to leave](): The four study measures moved together across the engagements studied \(n = 150 engagements\). The first principal component carried most of their shared variance \(first component share 0.88\). Cancellation and customer satisfaction formed the strongest pair \(r = −0.94, 95% CI −0.96 to −0.92\), and cancellation and the engagement indicator mean formed the weakest \(r = −0.72, 95% CI −0.79 to −0.63\). Cancellation rose with intent to leave \(r = 0.80, 95% CI 0.74 to 0.85\), and customer satisfaction fell as intent to leave rose \(r = −0.90, 95% CI −0.92 to −0.86\). - [The four study measures by engagement indicator mean quintile: 150 engagements in five bands of 30](): Every study measure moved in order across the five engagement indicator mean quintiles set by the rule printed in Analysis Results. Cancellation fell across the quintiles \(rho −0.73, p < 0.001, n = 150 engagements\), from the lowest quintile \(93.33% of 30\) to the top quintile \(0.00% of 30\). Customer satisfaction rose from the lowest quintile \(13.0, n = 30 engagements\) to the top quintile \(75.8, n = 30 engagements\), and the quintiles differed under test \(F 84.65, p < 0.001\). - [How many months has your employer been serving this customer across all work, not just this engagement?](): Respondents reported that their employer had served the financial institution for several years on average \(mean 32.6, median 33.0, n = 354 respondents\). The association with cancellation was equivalent to zero \(TOST p < 0.001\). Customer satisfaction on active engagements showed a positive association that fell short of the test \(r = 0.20, 95% CI 0.00 to 0.39, p 0.055, n = 94 active engagements\) and could not be shown equivalent to zero inside the bound. The association with the engagement indicator mean was equivalent to zero \(TOST p < 0.001\). - [The engagement is not achieving stated KPIs](): Distressed answers on whether the engagement was achieving its stated KPIs came from a minority of answers \(241 of 782 answers, 30.82%\) and a minority of engagements \(32 of 150 engagements\). On this item a higher figure marks an engagement closer to its stated KPIs \(each figure is 11 minus the raw answer\). Engagements rated further from their stated KPIs were canceled more often \(r = −0.59, 95% CI −0.69 to −0.48, p < 0.001, n = 150 engagements\). - [The customer is happy with the performance of the engagement at this stage](): Distressed ratings on whether the financial institution was satisfied with the engagement's performance at this stage came from a minority of respondents \(89 of 379 answers, 23.48%\). Engagements whose respondents reported a more satisfied financial institution were canceled less often \(r = −0.57, 95% CI −0.67 to −0.45, p < 0.001, n = 150 engagements\). Higher ratings came with higher customer satisfaction on active engagements \(r = 0.53, 95% CI 0.36 to 0.66, p < 0.001, n = 94 active engagements\). - [How long ago did this specific engagement start, in months?](): Respondents reported a mean engagement age in months since the engagement started \(mean 5.6, median 5.0, n = 804 respondents\). The association with cancellation was equivalent to zero \(TOST p 0.003\). Customer satisfaction on active engagements showed the same tested null \(TOST p 0.019\). The association with the engagement indicator mean was also equivalent to zero \(TOST p 0.011\). Intent to leave among respondents returned a tested null as well \(TOST p 0.005\). - [The engagement is meeting delivery dates successfully](): Distressed ratings on meeting delivery dates came from a minority of answers \(234 of 798 answers, 29.32%\) and a minority of engagement means \(30 of 150 engagements\). Engagements whose respondents rated delivery lower were canceled more often \(r = −0.60, 95% CI −0.70 to −0.49, p < 0.001, n = 150 engagements\). Higher delivery ratings came with higher customer satisfaction on active engagements \(r = 0.59, 95% CI 0.44 to 0.71, p < 0.001, n = 94 active engagements\). Delivery ratings rose with the engagement indicator mean \(r = 0.78, 95% CI 0.71 to 0.84, p < 0.001, n = 150 engagements\). - [The customer is increasing the number of agentic AI outsourcing engagements](): Agreement that the financial institution was increasing its number of agentic AI outsourcing engagements \(mean 5.3, median 5.0, n = 247 respondents\) showed tested nulls against all four study measures. The association with cancellation was equivalent to zero \(TOST p 0.001\). Customer satisfaction on active engagements showed the same tested null \(TOST p 0.019\). The association with the engagement indicator mean was also equivalent to zero \(TOST p 0.005\). Intent to leave returned a tested null as well \(TOST p < 0.001\). - [How many agentic AI engagements does your employer run for banks today?](): Respondents reported how many agentic AI engagements their employer ran for financial institutions, with the mean close to the median \(mean 24.6, median 25.0, n = 328 respondents\). The association with cancellation was equivalent to zero \(TOST p < 0.001\). The association with customer satisfaction on active engagements was equivalent to zero \(TOST p 0.018\). The association with the engagement indicator mean was equivalent to zero \(TOST p < 0.001\). The association with intent to leave was equivalent to zero \(TOST p 0.001\). - [What type of a bank is the customer?](): Wealth management and private banking was the institution type respondents named most often \(140 of 848 respondents, 16.51%\), followed by Asset management \(132 of 848 respondents, 15.57%\). Private equity and private capital was named least often \(39 of 848 respondents, 4.60%\). Cancellation ranged from Payments and merchant acquiring at the low end \(23.08% canceled, n = 13 engagements\) to Asset management at the high end \(54.17% canceled, n = 24 engagements\), with no difference under test \(chi-square 5.71, p 0.692, Cramer's V 0.20, n = 141 engagements\). - [How many agentic AI outsourcing engagements does your employer run concurrently with this customer today?](): The number of agentic AI outsourcing engagements respondents reported running concurrently with the financial institution \(mean 17.2, median 18.0, n = 254 respondents\) showed no association with any of the four study measures. The association with cancellation was equivalent to zero \(TOST p 0.009\). Customer satisfaction on active engagements was also equivalent to zero \(TOST p 0.006\). The engagement indicator mean showed the same tested null \(TOST p < 0.001\). Intent to leave was equivalent to zero as well \(TOST p 0.002\). - [Corporate Functional or LOB Operational](): Engagements split almost evenly between LOB Operational \(76 of 150 engagements, 50.67%\) and Corporate Functional \(74 of 150 engagements, 49.33%\) work. Cancellation was higher on LOB Operational engagements \(39.47% canceled, n = 76 engagements\) than on Corporate Functional engagements \(35.14% canceled, n = 74 engagements\), with no difference under test \(chi-square 0.30, p 0.583, Cramer's V 0.04, n = 150 engagements\). - [Is the outsourcing engagement Corporate Functional, or Line of Business Operational?](): Respondents split close to evenly between LOB Operational \(450 of 849 respondents, 53.00%\) and Corporate Functional engagements \(399 of 849 respondents, 47.00%\). Cancellation did not differ between the two classes of work \(chi-square 0.30, p 0.583, Cramer's V 0.04, n = 150 engagements\). Customer satisfaction on active engagements did not differ between them \(Welch F 0.05, p 0.828, omega squared −0.010, n = 94 engagements\). The engagement indicator mean did not differ between them \(Welch F 0.07, p 0.795, omega squared −0.006, n = 150 engagements\). - [What is the customer's approximate total asset size, in USD?](): Respondents most often placed the financial institution's total assets in the second-highest asset band \(90 of 242 respondents, 37.19%\) and least often in the highest band \(35 of 242 respondents, 14.46%\). Cancellation could not be compared across asset bands \(fewer than two answer levels with at least 10 engagements\). Customer satisfaction could not be compared either \(fewer than two answer levels with at least 10 engagements\). The engagement indicator mean comparison was also not computable \(fewer than two answer levels with at least 10 engagements\). - [What stage is the engagement in?](): Engagement Canceled was the most common stage answer \(309 of 810 respondents, 38.15%\), followed by Development / buildout \(131 of 810 respondents, 16.17%\). Cancellation was an identity with the Engagement Canceled answer \(by construction\). Customer satisfaction on active engagements did not differ across the stage answers tested \(Welch F 0.69, p 0.603, omega squared −0.013, n = 85 engagements\), with Engagement Canceled, Team Buildout, and On Hold left out of the test \(n = 0, n = 7, n = 2 engagements\). - [Headcount](): Engagement headcount had a mean close to its median across the study set \(mean 8.4, median 8.0, n = 150 engagements\). The association with cancellation was equivalent to zero \(TOST p 0.010\). The association with customer satisfaction on active engagements was equivalent to zero \(TOST p 0.003\). The association with the engagement indicator mean was equivalent to zero \(TOST p 0.005\). The association with intent to leave was equivalent to zero \(TOST p < 0.001\). - [How many people are on the vendor team for this engagement today?](): Vendor team size \(mean 8.8, median 9.0, n = 789 respondents\) showed no association with any of the four study measures. The association with cancellation was equivalent to zero \(TOST p 0.010\). Customer satisfaction on active engagements was also equivalent to zero \(TOST p 0.003\). The engagement indicator mean showed the same tested null \(TOST p 0.005\). Intent to leave was equivalent to zero as well \(TOST p 0.002\). Team size differed after Holm control by process family \(Welch F 4.28, Holm p 0.034, omega squared 0.061\) and by class of work \(Welch F 12.32, Holm p 0.004, omega squared 0.070\). - [The engagement is within budget](): Distressed ratings on staying within budget came from a minority of answers \(151 of 554 answers, 27.26%\) and a minority of engagement means \(31 of 150 engagements\). Engagements whose respondents rated budget performance lower were canceled more often \(r = −0.54, 95% CI −0.64 to −0.42, p < 0.001, n = 150 engagements\). Higher within-budget ratings came with higher customer satisfaction on active engagements \(r = 0.62, 95% CI 0.48 to 0.73, p < 0.001, n = 94 active engagements\). - [Per the current plan, how many people will be on the vendor team 12 months from now?](): The vendor team size respondents reported under the current plan for a year from now \(mean 5.6, median 6.0, n = 339 respondents\) was associated with all four study measures. Engagements with smaller planned teams were canceled more often \(r = −0.86, 95% CI −0.89 to −0.81, p < 0.001, n = 150 engagements\). Larger planned teams came with higher customer satisfaction on active engagements \(r = 0.47, 95% CI 0.30 to 0.62, p < 0.001, n = 94 active engagements\). Planned team size rose with the engagement indicator mean \(r = 0.73, 95% CI 0.65 to 0.80, p < 0.001, n = 150 engagements\). - [What is the contract term of this engagement, in months?](): Respondents reported multi-year contract terms at the median \(mean 30.9, median 36.0, n = 339 respondents\). The association with cancellation was equivalent to zero \(TOST p 0.006\). The association with customer satisfaction on active engagements was equivalent to zero \(TOST p 0.004\). The association with the engagement indicator mean was equivalent to zero \(TOST p 0.003\). The association with intent to leave was equivalent to zero \(TOST p < 0.001\). Contract term differed after Holm control by vendor type \(Welch F 4.06, Holm p 0.029, omega squared 0.030\). - [Customer Satisfaction %](): The study set mean for customer satisfaction was below its median \(mean 43.9, median 53.5, n = 150 engagements\). Customer satisfaction was lower on canceled engagements \(r = −0.94, 95% CI −0.96 to −0.92, p < 0.001, n = 150 engagements\). Its cell against itself was an identity by construction \(r = 1.00\). Higher customer satisfaction came with a higher engagement indicator mean \(r = 0.85, 95% CI 0.80 to 0.89, p < 0.001, n = 150 engagements\). - [What is the annual contract value of this engagement, in USD?](): Respondents most often placed annual contract value in the second value band \(106 of 248 respondents, 42.74%\) or the third \(102 of 248 respondents, 41.13%\), and the lowest band was the least common answer \(13 of 248 respondents, 5.24%\). Cancellation was lower in the third band \(18.75% canceled, n = 16 engagements\) than in the second band \(33.33% canceled, n = 15 engagements\), with no difference in a test of those two bands \(chi-square 0.86, p 0.433, Cramer's V 0.17, n = 31 engagements\). - [Engagement indicator mean](): The engagement indicator mean across the study set was just above its median \(mean 4.0, median 3.8, n = 150 engagements\). Canceled engagements carried a lower engagement indicator mean \(r = −0.72, 95% CI −0.79 to −0.63, p < 0.001, n = 150 engagements\). A higher engagement indicator mean came with higher customer satisfaction on active engagements \(r = 0.84, 95% CI 0.77 to 0.89, p < 0.001, n = 94 active engagements\). This block is the engagement indicator mean itself. Its own cell was an identity by construction \(r = 1.00\). - [What type of company is your employer?](): IT services major was the most common employer type among respondents \(316 of 793 respondents, 39.85%\), and Pure-play agentic AI firm was the least common \(75 of 793 respondents, 9.46%\). Cancellation was lowest for IT services major \(32.20% canceled, n = 59 engagements\) and highest for BPO specialist \(45.95% canceled, n = 37 engagements\), with no difference under test \(chi-square 2.02, p 0.731, Cramer's V 0.12, n = 150 engagements\). - [Does the customer operate its own global capability centers?](): A majority of respondents reported that the financial institution operated its own global capability centers \(190 of 349 respondents, 54.44%\), and the rest reported that it did not \(159 of 349 respondents, 45.56%\). Cancellation did not differ between Yes and No answers \(chi-square 0.60, p 0.439, Cramer's V 0.09, n = 78 engagements\), with I do not know left out of the test \(n = 40 engagements\). Customer satisfaction on active engagements did not differ between Yes and No answers \(Welch F 2.15, p 0.148, omega squared 0.020, n = 53 engagements\), with I do not know left out \(n = 22 engagements\). - [I know the KPIs the customer uses to measure success](): Distressed ratings on knowing the KPIs the financial institution uses to measure success came from a minority of answers \(226 of 787 answers, 28.72%\). Engagements whose respondents reported weaker knowledge of those KPIs were canceled more often \(r = −0.64, 95% CI −0.72 to −0.53, p < 0.001, n = 150 engagements\). Stronger KPI knowledge came with higher customer satisfaction on active engagements \(r = 0.65, 95% CI 0.51 to 0.75, p < 0.001, n = 94 active engagements\). Ratings rose with the engagement indicator mean \(r = 0.81, 95% CI 0.75 to 0.86, p < 0.001, n = 150 engagements\). - [Use Case](): Corporate and back-office functions was the most common process family \(47 of 150 engagements, 31.33%\), ahead of Technology engineering and operations \(35 of 150 engagements, 23.33%\). Cancellation ranged from Payments, cards and disputes at the low end \(16.67% canceled, n = 12 engagements\) to Client servicing and communications at the high end \(56.25% canceled, n = 16 engagements\), with no difference under test \(chi-square 7.48, p 0.113, Cramer's V 0.24, n = 125 engagements\). - [Per the current plan, how many months from today until the engagement reaches Production/Delivery?](): Respondents reported a mean expected time to Production/Delivery in months on the current plan \(mean 3.9, median 4.0, n = 545 respondents\). Engagements with a longer expected time to Production/Delivery were canceled less often \(r = −0.37, 95% CI −0.53 to −0.19, p < 0.001, n = 104 engagements\). Expected time and customer satisfaction on active engagements showed a tested null \(TOST p 0.006\). A longer expected time came with a higher engagement indicator mean \(r = 0.20, 95% CI 0.01 to 0.38, p 0.038, n = 104 engagements\). - [If the engagement has not been canceled, how likely is the customer to cancel the engagement?](): Distressed answers on the financial institution's likelihood to cancel came from a minority of answers \(76 of 328 answers, 23.17%\) and a minority of engagements \(26 of 150 engagements\). On this item a higher figure marks a financial institution less likely to cancel \(each figure is 11 minus the raw answer\). Engagements rated more likely to cancel were canceled more often \(r = −0.54, 95% CI −0.65 to −0.42, p < 0.001, n = 150 engagements\). - [The customer is shifting more work to their global capability centers](): Higher ratings that the financial institution was shifting more work to its global capability centers came with weaker results on all four study measures. Engagements whose respondents rated the shift higher were canceled more often \(r = 0.54, 95% CI 0.42 to 0.65, p < 0.001, n = 150 engagements\). Higher ratings came with lower customer satisfaction on active engagements \(r = −0.66, 95% CI −0.76 to −0.53, p < 0.001, n = 94 active engagements\). Ratings fell as the engagement indicator mean rose \(r = −0.76, 95% CI −0.82 to −0.68, p < 0.001, n = 150 engagements\). - [Age of Engagement \(Months\)](): Engagement age, recorded in months for every engagement \(mean 5.7, median 5.0, n = 150 engagements\), held no association with any of the four study measures. The association with cancellation was close to zero \(r = −0.08, 95% CI −0.24 to 0.08, p 0.327, n = 150 engagements\) and tested as equivalent to zero \(TOST p 0.003\). Customer satisfaction on active engagements showed a near-zero association \(r = 0.09, 95% CI −0.11 to 0.29, p 0.374, n = 94 active engagements\) that tested as equivalent to zero \(TOST p 0.019\). - [Approximately how many employees does your employer have worldwide?](): The worldwide employee count respondents reported for their employer \(mean 110,954.9, median 90,854.0, n = 538 respondents\) showed no association with any of the four study measures. The association with cancellation was close to zero \(r = −0.06, 95% CI −0.22 to 0.10, p 0.471, n = 150 engagements\) and tested as equivalent to zero \(TOST p 0.001\). Customer satisfaction on active engagements showed a near-zero association \(r = −0.07, 95% CI −0.27 to 0.13, p 0.500, n = 94 active engagements\) that tested as equivalent to zero \(TOST p 0.011\). - [What was the peak vendor team headcount on this engagement to date?](): Peak vendor team headcount \(mean 10.5, median 10.0, n = 787 respondents\) was associated with every study measure except customer satisfaction, which returned a tested null. Engagements with larger peak teams were canceled more often \(r = 0.24, 95% CI 0.09 to 0.39, p 0.003, n = 150 engagements\). Customer satisfaction on active engagements returned a tested null \(TOST p 0.025\). Larger peak teams came with a lower engagement indicator mean \(r = −0.25, 95% CI −0.39 to −0.09, p 0.002, n = 150 engagements\). - [How many AI agents are running in production on this engagement today?](): At least half of respondents reported no AI agents running in production on their engagement \(mean 1.2, median 0.0, n = 787 respondents\). The association with cancellation was equivalent to zero \(TOST p < 0.001\). Customer satisfaction on active engagements showed the same tested null \(TOST p 0.008\). The association with the engagement indicator mean was also equivalent to zero \(TOST p < 0.001\). Intent to leave returned a tested null as well \(TOST p < 0.001\). The agent count differed after Holm control by class of work only \(Welch F 10.30, Holm p 0.007, omega squared 0.057\). - [Engagement Phase](): Build-Out \(60 of 150 engagements, 40.00%\) and Transition/Cut-Over \(60 of 150 engagements, 40.00%\) were the most common engagement phases, ahead of Production \(30 of 150 engagements, 20.00%\). Cancellation was highest in Build-Out \(48.33% canceled, n = 60 engagements\) and lowest in Transition/Cut-Over \(28.33% canceled, n = 60 engagements\), with no difference under test \(chi-square 5.39, p 0.068, Cramer's V 0.19, n = 150 engagements\). - [The customer is scaling their agentic AI outsourcing engagements](): Ratings of whether the financial institution was scaling its agentic AI outsourcing engagements \(mean 5.5, median 5.0, n = 364 respondents\) showed no association with any of the four study measures. The association with cancellation was equivalent to zero \(TOST p < 0.001\). Customer satisfaction on active engagements was also equivalent to zero \(TOST p 0.003\). The engagement indicator mean showed the same tested null \(TOST p 0.001\). Intent to leave was equivalent to zero as well \(TOST p < 0.001\). - [What is the pricing model for this engagement?](): FTE-based pricing was the most common answer \(140 of 347 respondents, 40.35%\), ahead of Fixed price \(61 of 347 respondents, 17.58%\) and Transaction or outcome based pricing \(58 of 347 respondents, 16.71%\). Cancellation was highest for Consumption based pricing \(60.00% canceled, n = 10 engagements\), with no difference across the tested answer levels \(chi-square 2.97, p 0.412, Cramer's V 0.20, n = 71 engagements\). Customer satisfaction could not be compared across answer levels \(fewer than two answer levels with at least 10 engagements\). - [Bank Type](): Asset management was the most common institution type among the engagements \(24 of 150 engagements, 16.00%\), with Wealth management and private banking close behind \(23 of 150 engagements, 15.33%\). Cancellation ranged from Payments and merchant acquiring at the low end \(23.08% canceled, n = 13 engagements\) to Asset management at the high end \(54.17% canceled, n = 24 engagements\), with no difference under test \(chi-square 5.71, p 0.693, Cramer's V 0.20, n = 141 engagements\). - [How many years has your employer been delivering agentic AI outsourcing?](): Respondents reported the years their employer had delivered agentic AI outsourcing \(mean 1.0, median 1.0, n = 571 respondents\). The association with cancellation was close to zero \(r = −0.12, 95% CI −0.28 to 0.04, p 0.136, n = 150 engagements\) and tested as equivalent to zero \(TOST p 0.012\). Customer satisfaction on active engagements showed a near-zero association \(r = −0.09, 95% CI −0.28 to 0.12, p 0.407, n = 94 active engagements\) that tested as equivalent to zero \(TOST p 0.017\). - [Indicator associations net of the engagement indicator mean: 100 component indicators against three measures](): The analysis tested each component indicator against cancellation, customer satisfaction, and intent to leave, net of the leave-one-out engagement indicator mean, in 300 tests. Each test was a partial correlation computed at engagement level. Benjamini-Hochberg control applied across all tests at once, as the Rule row in Analysis Results states. Tests with p below the nominal threshold before correction numbered 33 of 300. Tests surviving Benjamini-Hochberg control numbered 0 of 300. - [Outsourcing Type](): The engagements split across the four outsourcing types, from CXM \(42 of 150 engagements, 28.00%\) to KPO \(31 of 150 engagements, 20.67%\). Cancellation was close across types, from CXM \(35.71% canceled, n = 42 engagements\) to KPO \(38.71% canceled, n = 31 engagements\), with no difference under test \(chi-square 0.08, p 0.994, Cramer's V 0.02, n = 150 engagements\). - [What is the outsourcing type?](): Respondents named ITO most often as the outsourcing type \(244 of 849 respondents, 28.74%\) and KPO least often \(165 of 849 respondents, 19.43%\). Cancellation ranged from CXM at the low end \(35.71% canceled, n = 42 engagements\) to KPO at the high end \(38.71% canceled, n = 31 engagements\), with no difference under test \(chi-square 0.08, p 0.994, Cramer's V 0.02, n = 150 engagements\). - [Is Canceled?](): The cancellation flag recorded a study set mean \(mean 0.4, median 0.0, n = 150 engagements\). This block is cancellation itself. Its cancellation cell was an identity by construction \(r = 1.00\). The clean estimate for customer satisfaction could not be computed, because it ran inside active engagements and no active engagement was canceled. Canceled engagements carried a lower engagement indicator mean \(r = −0.72, 95% CI −0.79 to −0.63, p < 0.001, n = 150 engagements\). - [How many months has the engagement been in its current stage?](): Respondents reported a mean time in the current stage in months \(mean 3.1, median 2.0, n = 794 respondents\). The association with cancellation was equivalent to zero \(TOST p 0.005\). The associations with customer satisfaction on active engagements \(TOST p 0.012\) and with the engagement indicator mean \(TOST p 0.027\) were also equivalent to zero. The association with intent to leave was equivalent to zero as well \(TOST p 0.009\). - [Rate how accessible the customer's data is to this engagement when and where the work requires it](): Ratings of how accessible the financial institution's data was to the engagement when and where the work required it \(mean 3.5, median 3.0, n = 798 respondents\) were associated with all four study measures. Engagements with higher ratings were canceled less often \(r = −0.57, 95% CI −0.67 to −0.45, p < 0.001, n = 150 engagements\). Higher ratings came with higher customer satisfaction on active engagements \(r = 0.67, 95% CI 0.54 to 0.77, p < 0.001, n = 94 active engagements\). - [Rate the maturity of the customer's data governance over the data this engagement uses](): Ratings of the maturity of the financial institution's data governance over the data the engagement used \(mean 3.6, median 3.0, n = 797 respondents\) were associated with all four study measures. Engagements with higher ratings were canceled less often \(r = −0.56, 95% CI −0.66 to −0.44, p < 0.001, n = 150 engagements\). Higher ratings came with higher customer satisfaction on active engagements \(r = 0.67, 95% CI 0.54 to 0.77, p < 0.001, n = 94 active engagements\). - [Rate the readiness of the customer's technical infrastructure to support the agentic AI this engagement deploys](): Ratings of the readiness of the financial institution's technical infrastructure for the agentic AI the engagement deployed \(mean 3.5, median 3.0, n = 803 respondents\) were associated with all four study measures. Engagements with higher ratings were canceled less often \(r = −0.60, 95% CI −0.69 to −0.49, p < 0.001, n = 150 engagements\). Higher ratings came with higher customer satisfaction on active engagements \(r = 0.74, 95% CI 0.63 to 0.82, p < 0.001, n = 94 active engagements\). - [Rate the maturity of the customer's processes in scope, meaning how standardized and repeatable they are](): Ratings of the maturity of the financial institution's processes in scope \(mean 3.5, median 3.0, n = 789 respondents\) were associated with all four study measures. Engagements with higher ratings were canceled less often \(r = −0.57, 95% CI −0.67 to −0.45, p < 0.001, n = 150 engagements\). Higher ratings came with higher customer satisfaction on active engagements \(r = 0.72, 95% CI 0.61 to 0.81, p < 0.001, n = 94 active engagements\). Ratings rose with the engagement indicator mean, computed with this item left out \(r = 0.80, 95% CI 0.74 to 0.85, p < 0.001, n = 150 engagements\). - [Compared to 5 years ago, is my employer more or less prepared to manage outsourcing engagements?](): Ratings of how prepared the employer was to manage outsourcing engagements compared with five years earlier \(mean 3.7, median 4.0, n = 786 respondents\) were associated with all four study measures. Engagements with higher ratings were canceled less often \(r = −0.56, 95% CI −0.66 to −0.44, p < 0.001, n = 150 engagements\). Higher ratings came with higher customer satisfaction on active engagements \(r = 0.66, 95% CI 0.52 to 0.76, p < 0.001, n = 94 active engagements\). Ratings rose with the engagement indicator mean \(r = 0.85, 95% CI 0.80 to 0.89, p < 0.001, n = 150 engagements\). - [Rate how well your employer is transforming its commercial model as agentic AI reduces the billable human effort it sells today](): Ratings of how well the employer was transforming its commercial model as agentic AI reduced the billable human effort it sold \(mean 3.7, median 4.0, n = 812 respondents\) were associated with all four study measures. Engagements with higher ratings were canceled less often \(r = −0.60, 95% CI −0.69 to −0.48, p < 0.001, n = 150 engagements\). Higher ratings came with higher customer satisfaction on active engagements \(r = 0.68, 95% CI 0.55 to 0.77, p < 0.001, n = 94 active engagements\). - [Rate the customer's willingness to redesign processes for agentic AI rather than automate them as they are](): Ratings of the financial institution's willingness to redesign processes for agentic AI rather than automate them as they were \(mean 3.5, median 3.0, n = 804 respondents\) were associated with all four study measures. Engagements with higher ratings were canceled less often \(r = −0.56, 95% CI −0.66 to −0.44, p < 0.001, n = 150 engagements\). Higher ratings came with higher customer satisfaction on active engagements \(r = 0.70, 95% CI 0.58 to 0.79, p < 0.001, n = 94 active engagements\). - [Rate the customer organization's readiness to absorb the changes to daily work that agentic AI introduces](): Ratings of the financial institution's readiness to absorb the changes to daily work that agentic AI introduced \(mean 3.5, median 4.0, n = 794 respondents\) were associated with all four study measures. Engagements with higher ratings were canceled less often \(r = −0.59, 95% CI −0.69 to −0.48, p < 0.001, n = 150 engagements\). Higher ratings came with higher customer satisfaction on active engagements \(r = 0.73, 95% CI 0.61 to 0.81, p < 0.001, n = 94 active engagements\). - [Rate the customer's capacity to govern and oversee the risks of AI agents doing this work](): Ratings of the financial institution's capacity to govern and oversee the risks of AI agents doing the work \(mean 3.5, median 3.0, n = 797 respondents\) were associated with all four study measures. Engagements with higher ratings were canceled less often \(r = −0.57, 95% CI −0.67 to −0.45, p < 0.001, n = 150 engagements\). Higher ratings came with higher customer satisfaction on active engagements \(r = 0.75, 95% CI 0.64 to 0.83, p < 0.001, n = 94 active engagements\). - [Rate the effectiveness of your employer's process and leadership development programs](): Ratings of the effectiveness of the employer's process and leadership development programs \(mean 3.7, median 4.0, n = 803 respondents\) were associated with all four study measures. Engagements with higher ratings were canceled less often \(r = −0.59, 95% CI −0.69 to −0.48, p < 0.001, n = 150 engagements\). Higher ratings came with higher customer satisfaction on active engagements \(r = 0.69, 95% CI 0.57 to 0.78, p < 0.001, n = 94 active engagements\). Ratings rose with the engagement indicator mean, computed with this item left out \(r = 0.85, 95% CI 0.80 to 0.89, p < 0.001, n = 150 engagements\). - [Rate the effectiveness of the customer's management of its side of this engagement](): Ratings of how effectively the financial institution managed its side of the engagement \(mean 3.6, median 4.0, n = 811 respondents\) were associated with all four study measures. Engagements with higher ratings were canceled less often \(r = −0.56, 95% CI −0.66 to −0.44, p < 0.001, n = 150 engagements\). Higher ratings came with higher customer satisfaction on active engagements \(r = 0.72, 95% CI 0.60 to 0.80, p < 0.001, n = 94 active engagements\). Ratings rose with the engagement indicator mean, computed with this item left out \(r = 0.82, 95% CI 0.76 to 0.87, p < 0.001, n = 150 engagements\). - [Rate the stability of the customer's business case and ROI definition for this engagement](): Ratings of the stability of the financial institution's business case and definition of return for the engagement \(mean 3.5, median 3.0, n = 794 respondents\) were associated with all four study measures. Engagements with higher ratings were canceled less often \(r = −0.57, 95% CI −0.67 to −0.46, p < 0.001, n = 150 engagements\). Higher ratings came with higher customer satisfaction on active engagements \(r = 0.72, 95% CI 0.60 to 0.80, p < 0.001, n = 94 active engagements\). - [Rate the quality of the customer's data across the systems this engagement depends on](): Ratings of the quality of the financial institution's data across the systems the engagement depended on \(mean 3.6, median 4.0, n = 795 respondents\) were associated with all four study measures. Engagements with higher ratings were canceled less often \(r = −0.60, 95% CI −0.70 to −0.49, p < 0.001, n = 150 engagements\). Higher ratings came with higher customer satisfaction on active engagements \(r = 0.68, 95% CI 0.55 to 0.77, p < 0.001, n = 94 active engagements\). - [Rate how consistently the customer performs the same processes across its business units and regions](): Ratings of how consistently the financial institution performed the same processes across its business units and regions \(mean 3.6, median 3.0, n = 801 respondents\) were associated with all four study measures. Engagements with higher ratings were canceled less often \(r = −0.58, 95% CI −0.68 to −0.47, p < 0.001, n = 150 engagements\). Higher ratings came with higher customer satisfaction on active engagements \(r = 0.74, 95% CI 0.63 to 0.82, p < 0.001, n = 94 active engagements\). - [Rate the overall effectiveness of your employer's top leadership team](): Ratings of the overall effectiveness of the employer's top leadership team \(mean 3.6, median 4.0, n = 792 respondents\) were associated with all four study measures. Engagements with higher ratings were canceled less often \(r = −0.58, 95% CI −0.68 to −0.46, p < 0.001, n = 150 engagements\). Higher ratings came with higher customer satisfaction on active engagements \(r = 0.71, 95% CI 0.59 to 0.79, p < 0.001, n = 94 active engagements\). Ratings rose with the engagement indicator mean, computed with this item left out \(r = 0.86, 95% CI 0.81 to 0.89, p < 0.001, n = 150 engagements\). - [Rate how well the customer's data is normalized, with consistent formats and definitions across source systems](): Ratings of how well the financial institution's data was normalized across source systems \(mean 3.5, median 3.0, n = 809 respondents\) were associated with all four study measures. Engagements with higher ratings were canceled less often \(r = −0.60, 95% CI −0.69 to −0.49, p < 0.001, n = 150 engagements\). Higher ratings came with higher customer satisfaction on active engagements \(r = 0.68, 95% CI 0.56 to 0.78, p < 0.001, n = 94 active engagements\). Ratings rose with the engagement indicator mean, computed with this item left out \(r = 0.81, 95% CI 0.75 to 0.86, p < 0.001, n = 150 engagements\). - [Rate how well your employer's internal management processes have been redesigned for agentic AI delivery rather than traditional staffed delivery](): Ratings of how well the employer's internal management processes had been redesigned for agentic AI delivery \(mean 3.7, median 4.0, n = 799 respondents\) were associated with all four study measures. Engagements with higher ratings were canceled less often \(r = −0.57, 95% CI −0.67 to −0.45, p < 0.001, n = 150 engagements\). Higher ratings came with higher customer satisfaction on active engagements \(r = 0.68, 95% CI 0.55 to 0.78, p < 0.001, n = 94 active engagements\). - [Rate the availability of the customer's subject matter experts when this engagement needs them](): Ratings of the availability of the financial institution's subject matter experts when the engagement needed them \(mean 3.5, median 3.0, n = 793 respondents\) were associated with all four study measures. Engagements with higher ratings were canceled less often \(r = −0.59, 95% CI −0.68 to −0.47, p < 0.001, n = 150 engagements\). Higher ratings came with higher customer satisfaction on active engagements \(r = 0.70, 95% CI 0.58 to 0.79, p < 0.001, n = 94 active engagements\). - [Rate how quickly the customer responds when this engagement raises issues that need customer action](): Ratings of how quickly the financial institution responded when the engagement raised issues that needed its action \(mean 3.5, median 3.0, n = 789 respondents\) were associated with all four study measures. Engagements with higher ratings were canceled less often \(r = −0.60, 95% CI −0.69 to −0.48, p < 0.001, n = 150 engagements\). Higher ratings came with higher customer satisfaction on active engagements \(r = 0.74, 95% CI 0.63 to 0.82, p < 0.001, n = 94 active engagements\). - [Rate the agentic AI knowledge and skills of the customer staff working with this engagement](): Ratings of the agentic AI knowledge and skills of the financial institution staff working with the engagement \(mean 3.5, median 3.0, n = 793 respondents\) were associated with all four study measures. Engagements with higher ratings were canceled less often \(r = −0.59, 95% CI −0.69 to −0.48, p < 0.001, n = 150 engagements\). Higher ratings came with higher customer satisfaction on active engagements \(r = 0.70, 95% CI 0.58 to 0.79, p < 0.001, n = 94 active engagements\). - [Rate how clearly ownership and accountability for AI are assigned inside the customer organization](): Ratings of how clearly ownership and accountability for AI were assigned inside the financial institution \(mean 3.5, median 3.0, n = 802 respondents\) were associated with all four study measures. Engagements with higher ratings were canceled less often \(r = −0.57, 95% CI −0.67 to −0.45, p < 0.001, n = 150 engagements\). Higher ratings came with higher customer satisfaction on active engagements \(r = 0.70, 95% CI 0.59 to 0.79, p < 0.001, n = 94 active engagements\). - [Rate how well the customer's legacy systems support the integration this engagement requires](): Ratings of how well the financial institution's legacy systems supported the integration the engagement required \(mean 3.5, median 3.0, n = 801 respondents\) were associated with all four study measures. Engagements with higher ratings were canceled less often \(r = −0.58, 95% CI −0.67 to −0.46, p < 0.001, n = 150 engagements\). Higher ratings came with higher customer satisfaction on active engagements \(r = 0.72, 95% CI 0.60 to 0.80, p < 0.001, n = 94 active engagements\). - [Rate the customer's ability to provision the system access and identities that AI agents require](): Ratings of the financial institution's ability to provision the system access and identities that AI agents required \(mean 3.5, median 3.0, n = 801 respondents\) were associated with all four study measures. Engagements with higher ratings were canceled less often \(r = −0.58, 95% CI −0.67 to −0.46, p < 0.001, n = 150 engagements\). Higher ratings came with higher customer satisfaction on active engagements \(r = 0.69, 95% CI 0.57 to 0.79, p < 0.001, n = 94 active engagements\). - [Rate how completely the customer's processes are mapped and documented to the depth agentic AI requires](): Ratings of how completely the financial institution's processes were mapped and documented to the depth agentic AI required \(mean 3.5, median 3.0, n = 795 respondents\) were associated with all four study measures. Engagements with higher ratings were canceled less often \(r = −0.56, 95% CI −0.66 to −0.44, p < 0.001, n = 150 engagements\). Higher ratings came with higher customer satisfaction on active engagements \(r = 0.70, 95% CI 0.57 to 0.79, p < 0.001, n = 94 active engagements\). - [Rate how well the customer's change control process handles the changes this engagement requires](): Ratings of how well the financial institution's change control process handled the changes the engagement required \(mean 3.6, median 3.0, n = 800 respondents\) were associated with all four study measures. Engagements with higher ratings were canceled less often \(r = −0.59, 95% CI −0.69 to −0.48, p < 0.001, n = 150 engagements\). Higher ratings came with higher customer satisfaction on active engagements \(r = 0.75, 95% CI 0.64 to 0.83, p < 0.001, n = 94 active engagements\). - [Rate your employer's ability to manage agentic AI outsourcing engagements end to end](): The mean rating was 3.7 and the median 4.0 \(n = 787 respondents\). Ratings of 3 or below, the distressed range, made up 46.25% of the answers. The rating was associated with all four study measures. - [Rate the depth of your employer's knowledge of the financial industry](): The mean rating was 3.6 and the median 4.0 \(n = 802 respondents\). Ratings of 3 or below, the distressed range, made up 48.88% of the answers. The rating was associated with all four study measures. - [Rate your employer's ability to manage agentic AI outsourcing engagements for global banks in particular](): The mean rating was 3.6 and the median 4.0 \(n = 779 respondents\). Ratings of 3 or below, the distressed range, made up 48.91% of the answers. The rating was associated with all four study measures. - [Rate your employer's current supply of qualified agentic AI talent against what its engagements need](): The mean rating was 3.7 and the median 4.0 \(n = 802 respondents\). Ratings of 3 or below, the distressed range, made up 47.63% of the answers. The rating was associated with all four study measures. - [Rate the adequacy of your employer's investment in training its workforce](): The mean rating was 3.7 and the median 4.0 \(n = 789 respondents\). Ratings of 3 or below, the distressed range, made up 49.18% of the answers. The rating was associated with all four study measures. - [Rate how well your employer's training programs build practical agentic AI skills](): The mean rating was 3.7 and the median 4.0 \(n = 804 respondents\). Ratings of 3 or below, the distressed range, made up 47.26% of the answers. The rating was associated with all four study measures. - [Rate how well your employer's training programs build financial industry knowledge](): The mean rating was 3.7 and the median 4.0 \(n = 788 respondents\). Ratings of 3 or below, the distressed range, made up 48.22% of the answers. The rating was associated with all four study measures. - [Rate the maturity of your employer's playbook for taking agentic use cases from concept to governed production](): The mean rating was 3.7 and the median 4.0 \(n = 807 respondents\). Ratings of 3 or below, the distressed range, made up 48.57% of the answers. The rating was associated with all four study measures. - [Rate your employer's quality assurance capability for agents whose outputs are not identical every run](): The mean rating was 3.7 and the median 4.0 \(n = 794 respondents\). Ratings of 3 or below, the distressed range, made up 47.48% of the answers. The rating was associated with all four study measures. - [Rate your employer's ability to price agentic work profitably when billable hours no longer measure the value delivered](): The mean rating was 3.6 and the median 4.0 \(n = 795 respondents\). Ratings of 3 or below, the distressed range, made up 48.18% of the answers. The rating was associated with all four study measures. - [Rate how well the technical platforms your employer builds or licenses are optimized for agentic AI outsourcing](): The mean rating was 3.7 and the median 4.0 \(n = 794 respondents\). Ratings of 3 or below, the distressed range, made up 48.36% of the answers. The rating was associated with all four study measures. - [Rate your employer's readiness to take accountability when an agent it operates makes a costly error](): The mean rating was 3.7 and the median 4.0 \(n = 794 respondents\). Ratings of 3 or below, the distressed range, made up 46.98% of the answers. The rating was associated with all four study measures. - [Rate the depth of your employer's knowledge of agentic AI](): The mean rating was 3.6 and the median 3.0 \(n = 807 respondents\). Ratings of 3 or below, the distressed range, made up 51.55% of the answers. The rating was associated with all four study measures. - [Rate the depth of your employer's experience running agentic AI systems in live production, beyond pilots](): The mean rating was 3.6 and the median 4.0 \(n = 794 respondents\). Ratings of 3 or below, the distressed range, made up 48.24% of the answers. The rating was associated with all four study measures. - [Rate your employer's ability to operate large numbers of agents in production, including monitoring and exception handling](): The mean rating was 3.7 and the median 4.0 \(n = 798 respondents\). Ratings of 3 or below, the distressed range, made up 47.49% of the answers. The rating was associated with all four study measures. - [Rate your employer's overall qualification to deliver elite agentic AI outsourcing services to banks](): The mean rating was 3.6 and the median 4.0 \(n = 803 respondents\). Ratings of 3 or below, the distressed range, made up 47.95% of the answers. The rating was associated with all four study measures. - [Rate how effectively the customer and vendor jointly manage this engagement now that agents perform part of the work](): The mean rating was 3.7 and the median 4.0 \(n = 789 respondents\). Ratings of 3 or below, the distressed range, made up 48.80% of the answers. The rating was associated with all four study measures. - [Rate how clearly the two firms have settled who decides what agents may do without human approval](): The mean rating was 3.6 and the median 3.0 \(n = 808 respondents\). Ratings of 3 or below, the distressed range, made up 50.25% of the answers. The rating was associated with all four study measures. - [Rate how clearly ownership of each workflow step is assigned when agents, vendor staff, and customer staff work in the same process](): The mean rating was 3.7 and the median 4.0 \(n = 808 respondents\). Ratings of 3 or below, the distressed range, made up 48.02% of the answers. The rating was associated with all four study measures. - [Rate how clearly accountability lands on one firm when an agent's action causes a problem in this engagement](): The mean rating was 3.7 and the median 3.0 \(n = 799 respondents\). Ratings of 3 or below, the distressed range, made up 50.31% of the answers. The rating was associated with all four study measures. - [Rate how well the two firms maintain a single shared view of the risks that agent work introduces](): The mean rating was 3.6 and the median 4.0 \(n = 793 respondents\). Ratings of 3 or below, the distressed range, made up 48.30% of the answers. The rating was associated with all four study measures. - [Rate how effectively the two firms manage issues arising from agent work as one team rather than through two separate processes](): The mean rating was 3.7 and the median 4.0 \(n = 799 respondents\). Ratings of 3 or below, the distressed range, made up 48.94% of the answers. The rating was associated with all four study measures. - [Rate how well the speed of joint issue resolution matches the speed at which agent-caused problems can spread](): The mean rating was 3.7 and the median 4.0 \(n = 809 respondents\). Ratings of 3 or below, the distressed range, made up 49.44% of the answers. The rating was associated with all four study measures. - [Rate how quickly the two firms together can stop an agent that is behaving badly in live operations](): The mean rating was 3.6 and the median 3.0 \(n = 804 respondents\). Ratings of 3 or below, the distressed range, made up 50.50% of the answers. The rating was associated with all four study measures. - [Rate how well the joint change control process keeps pace with the speed of agent behavior changes](): The mean rating was 3.7 and the median 3.5 \(n = 798 respondents\). Ratings of 3 or below, the distressed range, made up 50.00% of the answers. The rating was associated with all four study measures. - [Rate how well agent updates made by one firm become visible to the other firm before they take effect](): The mean rating was 3.6 and the median 4.0 \(n = 803 respondents\). Ratings of 3 or below, the distressed range, made up 49.07% of the answers. The rating was associated with all four study measures. - [Rate how well the two firms jointly govern changes agents make to their own behavior between formal releases](): The mean rating was 3.7 and the median 4.0 \(n = 806 respondents\). Ratings of 3 or below, the distressed range, made up 48.14% of the answers. The rating was associated with all four study measures. - [Rate how well human review points are placed across the two firms at the moments where agent errors would matter most](): The mean rating was 3.6 and the median 4.0 \(n = 797 respondents\). Ratings of 3 or below, the distressed range, made up 49.31% of the answers. The rating was associated with all four study measures. - [Rate how well joint human review capacity keeps up with the volume of output the agents produce](): The mean rating was 3.7 and the median 4.0 \(n = 805 respondents\). Ratings of 3 or below, the distressed range, made up 47.33% of the answers. The rating was associated with all four study measures. - [Rate how much genuine scrutiny agent outputs receive at human review points rather than routine approval](): The mean rating was 3.6 and the median 4.0 \(n = 794 respondents\). The rating was associated with all four study measures. Engagements with higher ratings were canceled less often \(r = −0.64, 95% CI −0.73 to −0.53, p < 0.001, n = 150 engagements\). Higher ratings accompanied higher customer satisfaction among active engagements \(r = 0.62, 95% CI 0.48 to 0.73, p < 0.001, n = 94 engagements\) and a higher engagement indicator mean, computed without the indicator \(r = 0.85, 95% CI 0.80 to 0.89, p < 0.001, n = 150 engagements\). - [Rate how reliably the joint quality assurance loop catches agent errors before they reach live operations](): The mean rating was 3.7 and the median 4.0 \(n = 787 respondents\). Ratings of 3 or below, the distressed range, made up 47.90% of the answers. The rating was associated with all four study measures. - [Rate how well the quality measures both firms use capture the kinds of errors agents actually make](): The mean rating was 3.6 and the median 3.0 \(n = 792 respondents\). Ratings of 3 or below, the distressed range, made up 51.14% of the answers. The rating was associated with all four study measures. - [Rate how fully each firm can see what the other firm's agents actually did when questions arise](): The mean rating was 3.7 and the median 3.0 \(n = 780 respondents\). Ratings of 3 or below, the distressed range, made up 50.26% of the answers. The rating was associated with all four study measures. - [Rate how closely the speed of joint decision making matches the speed at which the agents act](): The mean rating was 3.7 and the median 4.0 \(n = 786 respondents\). Ratings of 3 or below, the distressed range, made up 48.47% of the answers. The rating was associated with all four study measures. - [Rate how well today's joint governance model fits work performed by agents rather than work performed by people](): The mean rating was 3.6 and the median 4.0 \(n = 803 respondents\). Ratings of 3 or below, the distressed range, made up 49.19% of the answers. The rating was associated with all four study measures. - [Rate how ready the joint management model is for the level of agent autonomy you expect in this engagement within two years](): The mean rating was 3.6 and the median 3.0 \(n = 808 respondents\). Ratings of 3 or below, the distressed range, made up 50.37% of the answers. The rating was associated with all four study measures. - [Rate the factual reliability of the AI tools' outputs on this engagement before any human review](): The mean rating was 4.6 and the median 5.0 \(n = 796 respondents\). Ratings of 3 or below, the distressed range, made up 32.41% of the answers. The rating was associated with all four study measures. - [Rate the accuracy of citations, references, and quotations the tools produced against the real sources](): The mean rating was 4.5 and the median 5.0 \(n = 791 respondents\). Ratings of 3 or below, the distressed range, made up 33.50% of the answers. The rating was associated with all four study measures. - [Rate the consistency of the tools' outputs when given the same input and instructions more than once](): The mean rating was 4.6 and the median 5.0 \(n = 800 respondents\). Ratings of 3 or below, the distressed range, made up 32.38% of the answers. The rating was associated with all four study measures. - [Rate the end-to-end reliability of the tools on multi-step tasks, from first step to finished output](): The mean rating was 4.5 and the median 5.0 \(n = 782 respondents\). Ratings of 3 or below, the distressed range, made up 32.48% of the answers. The rating was associated with all four study measures. - [Rate the predictability of the tools' token consumption for a given volume of work](): The mean rating was 4.6 and the median 4.0 \(n = 795 respondents\). Ratings of 3 or below, the distressed range, made up 31.32% of the answers. The rating was associated with all four study measures. - [Rate the efficiency of the tools' token use, considering retries, loops, and reprocessing needed to complete work](): The mean rating was 4.6 and the median 5.0 \(n = 788 respondents\). Ratings of 3 or below, the distressed range, made up 32.36% of the answers. The rating was associated with all four study measures. - [Rate how completely the tools kept each client's and each session's information isolated from every other client and session](): The mean rating was 4.6 and the median 5.0 \(n = 791 respondents\). Ratings of 3 or below, the distressed range, made up 30.85% of the answers. The rating was associated with all four study measures. - [Rate how reliably the tools retained instructions and constraints given earlier in long tasks and sessions](): The mean rating was 4.6 and the median 5.0 \(n = 796 respondents\). Ratings of 3 or below, the distressed range, made up 31.66% of the answers. The rating was associated with all four study measures. - [Rate how completely the tools captured material facts located anywhere in long documents, including mid-document content](): The mean rating was 4.6 and the median 5.0 \(n = 805 respondents\). Ratings of 3 or below, the distressed range, made up 30.81% of the answers. The rating was associated with all four study measures. - [Rate the integrity of the tools' persistent memory against contamination by incorrect or malicious content encountered during work](): The mean rating was 4.6 and the median 4.0 \(n = 806 respondents\). Ratings of 3 or below, the distressed range, made up 31.89% of the answers. The rating was associated with all four study measures. - [Rate the tools' resistance to hostile instructions embedded in the content they read and process](): The mean rating was 4.6 and the median 5.0 \(n = 812 respondents\). Ratings of 3 or below, the distressed range, made up 32.14% of the answers. The rating was associated with all four study measures. - [Rate how consistently the agents took only intended and authorized actions when operating connected tools and systems](): The mean rating was 4.6 and the median 5.0 \(n = 789 respondents\). Ratings of 3 or below, the distressed range, made up 32.07% of the answers. The rating was associated with all four study measures. - [Rate how safely the tools handled sensitive data, with no exposure through outputs, logs, or connected systems](): The mean rating was 4.5 and the median 4.0 \(n = 800 respondents\). Ratings of 3 or below, the distressed range, made up 32.38% of the answers. The rating was associated with all four study measures. - [Rate how reliably the tools' built-in guardrails and safety controls held during production use](): The mean rating was 4.7 and the median 5.0 \(n = 787 respondents\). Ratings of 3 or below, the distressed range, made up 32.02% of the answers. The rating was associated with all four study measures. - [Rate the overall production readiness of the AI tools for bank-grade work as experienced on this engagement](): The mean rating was 4.5 and the median 5.0 \(n = 802 respondents\). Ratings of 3 or below, the distressed range, made up 32.04% of the answers. The rating was associated with all four study measures. - [Rate how accurately the tools' own reports of success or failure matched the actual outcomes of their work](): The mean rating was 4.6 and the median 5.0 \(n = 786 respondents\). Ratings of 3 or below, the distressed range, made up 31.93% of the answers. The rating was associated with all four study measures. - [Rate the tools' ability to explain and evidence how each output was produced, at the depth an audit requires](): The mean rating was 4.6 and the median 5.0 \(n = 800 respondents\). Ratings of 3 or below, the distressed range, made up 32.75% of the answers. The rating was associated with all four study measures. - [Rate how well the tools supported reliable testing, meaning correct behavior could be proven before work went live](): The mean rating was 4.6 and the median 5.0 \(n = 792 respondents\). Ratings of 3 or below, the distressed range, made up 32.70% of the answers. The rating was associated with all four study measures. - [Rate the stability of the tools' behavior over the engagement when nothing changed on the engagement side](): The mean rating was 4.6 and the median 5.0 \(n = 797 respondents\). Ratings of 3 or below, the distressed range, made up 33.00% of the answers. The rating was associated with all four study measures. - [Rate the quality of the tools' logs, traces, and usage telemetry for detecting and diagnosing failures](): The mean rating was 4.6 and the median 5.0 \(n = 796 respondents\). Ratings of 3 or below, the distressed range, made up 32.04% of the answers. The rating was associated with all four study measures. - [Rate how well the agents on this engagement are protected from breaking silently when a third party changes something without warning](): The mean rating was 4.6 and the median 5.0 \(n = 781 respondents\). Ratings of 3 or below, the distressed range, made up 31.88% of the answers. The rating was associated with all four study measures. - [Rate the quality of the fallback the engagement has ready when a critical third-party feed, platform, or model stops working](): The mean rating was 4.6 and the median 5.0 \(n = 802 respondents\). Ratings of 3 or below, the distressed range, made up 32.04% of the answers. The rating was associated with all four study measures. - [Rate the security posture of the third-party tools, platforms, and sub-processors the engagement's agents connect to](): The mean rating was 4.6 and the median 5.0 \(n = 788 respondents\). Ratings of 3 or below, the distressed range, made up 31.85% of the answers. The rating was associated with all four study measures. - [Rate the quality of advance notice third-party providers give before changes that can affect the agents on this engagement](): The mean rating was 4.6 and the median 5.0 \(n = 801 respondents\). Ratings of 3 or below, the distressed range, made up 31.21% of the answers. The rating was associated with all four study measures. - [Rate how reliably word of an upcoming third-party change reaches the people who operate the affected agents in time to act](): The mean rating was 4.6 and the median 5.0 \(n = 798 respondents\). Ratings of 3 or below, the distressed range, made up 31.95% of the answers. The rating was associated with all four study measures. - [Rate the engagement's monitoring for unannounced changes in the third-party APIs, data feeds, and models its agents depend on](): The mean rating was 4.6 and the median 5.0 \(n = 794 respondents\). Ratings of 3 or below, the distressed range, made up 31.61% of the answers. The rating was associated with all four study measures. - [Rate how well the contracts with third-party providers deal with provider changes that can break the agents on this engagement](): The mean rating was 4.7 and the median 5.0 \(n = 799 respondents\). Ratings of 3 or below, the distressed range, made up 32.04% of the answers. The rating was associated with all four study measures. - [Rate how well the engagement absorbs model version deprecations and retirements forced by model providers](): The mean rating was 4.6 and the median 5.0 \(n = 793 respondents\). Ratings of 3 or below, the distressed range, made up 32.16% of the answers. The rating was associated with all four study measures. - [Rate the availability of true machine interfaces, rather than screens built for humans, across the third-party systems the agents must use](): The mean rating was 4.6 and the median 5.0 \(n = 802 respondents\). Ratings of 3 or below, the distressed range, made up 29.80% of the answers. The rating was associated with all four study measures. - [Rate the stability of the third-party APIs and data structures the agents depend on, from one provider release to the next](): The mean rating was 4.7 and the median 5.0 \(n = 796 respondents\). Ratings of 3 or below, the distressed range, made up 30.53% of the answers. The rating was associated with all four study measures. - [Rate the quality of the data arriving from third-party feeds, judged by whether agents can use it without human cleanup](): The mean rating was 4.6 and the median 5.0 \(n = 795 respondents\). Ratings of 3 or below, the distressed range, made up 31.32% of the answers. The rating was associated with all four study measures. - [Rate how well third-party rate limits and quotas hold up under the volume the agents on this engagement generate](): The mean rating was 4.6 and the median 5.0 \(n = 805 respondents\). Ratings of 3 or below, the distressed range, made up 32.92% of the answers. The rating was associated with all four study measures. - [Rate the engagement's ability to still perform by hand the manual workarounds that once absorbed third-party quirks and breakage](): The mean rating was 4.7 and the median 5.0 \(n = 808 respondents\). Ratings of 3 or below, the distressed range, made up 30.82% of the answers. The rating was associated with all four study measures. - [Rate how completely the engagement knows where humans used to quietly fix third-party data and interface problems before agents took over](): The mean rating was 4.6 and the median 5.0 \(n = 792 respondents\). Ratings of 3 or below, the distressed range, made up 33.08% of the answers. The rating was associated with all four study measures. - [Rate how deliberately the quiet human fixes for third-party quirks were replaced with explicit checks when agents took over the work](): The mean rating was 4.6 and the median 5.0 \(n = 796 respondents\). Ratings of 3 or below, the distressed range, made up 33.54% of the answers. The rating was associated with all four study measures. - [Rate how well the engagement manages its dependence on a small number of model and cloud providers](): The mean rating was 4.6 and the median 5.0 \(n = 802 respondents\). Ratings of 3 or below, the distressed range, made up 32.04% of the answers. The rating was associated with all four study measures. - [Rate how well this engagement could keep operating if it had to switch its primary model or cloud provider](): The mean rating was 4.6 and the median 5.0 \(n = 811 respondents\). Ratings of 3 or below, the distressed range, made up 31.44% of the answers. The rating was associated with all four study measures. - [Rate how well third-party licenses on this engagement cover work done by agents rather than named human users](): The mean rating was 4.6 and the median 5.0 \(n = 799 respondents\). Ratings of 3 or below, the distressed range, made up 31.91% of the answers. The rating was associated with all four study measures. - [Rate how completely the engagement has mapped every third and fourth party its agents depend on, down the full chain](): The mean rating was 4.6 and the median 5.0 \(n = 791 respondents\). Ratings of 3 or below, the distressed range, made up 31.86% of the answers. The rating was associated with all four study measures. - [Rate how well the engagement manages agent dependence on gated external rails such as payment networks and regulatory submission systems](): The mean rating was 4.6 and the median 5.0 \(n = 794 respondents\). Ratings of 3 or below, the distressed range, made up 30.86% of the answers. The rating was associated with all four study measures. - [CME Respondent Count](): The Engagements sheet records, for each engagement, the number of its respondents who were customer management executives. The count was zero on 148 of the 150 engagements and one on two, a mean of 0.01 \(n = 150 engagements\). All of its variation therefore lies in two engagements. - [IC Respondent Count](): The Engagements sheet records, for each engagement, the number of its respondents who were individual contributors. The count ran from 2 to 8, with a mean of 4.4 \(95% CI 4.3 to 4.6\) and a median of 4.0 \(n = 150 engagements\). - [PD Respondent Count](): The Engagements sheet records, for each engagement, the number of its respondents who were portfolio directors. The count was one on 31 of the 150 engagements \(20.67%\) and zero on the other 119 \(n = 150 engagements\). No engagement had more than one portfolio director among its respondents. - [Total Respondents](): The Engagements sheet records, for each engagement, the number of its respondents. The total ran from 3 to 9, with a mean of 5.7 \(95% CI 5.5 to 5.9\) and a median of 6.0 \(n = 150 engagements\). - [What is your role on this engagement?](): All 850 respondents answered the question. Individual contributors numbered 667 \(78.47%\), engagement managers 150 \(17.65%\), portfolio directors 31 \(3.65%\) and customer management executives 2. Every engagement had exactly one engagement manager among its respondents and a majority of individual contributors. Individual Contributor was the assigned answer on every engagement. No test across answer levels was computed at engagement level. The table of principal results therefore carries no engagement-level test. - [How many years of experience do you have in your career?](): Respondents typed the number of years of experience in their career. The mean answer was 3.9 years and the median 2.0, from 0 to 25 \(n = 797 respondents\). An answer of zero, recorded as under one year, came from 61 respondents. - [How many years of experience do you have with agentic AI?](): Respondents typed the number of years of experience they had with agentic AI. Of the 793 who answered, 467 \(58.89%\) typed zero, recorded as under one year, and 326 \(41.11%\) typed one year. No respondent typed more than one year. The engagement mean of agentic AI experience was not associated with cancellation \(r = 0.00, 95% CI −0.16 to 0.16, p = 0.993, n = 150 engagements\), customer satisfaction among active engagements \(r = 0.13, 95% CI −0.08 to 0.32, p = 0.220, n = 94 engagements\) or the engagement indicator mean \(r = 0.02, 95% CI −0.14 to 0.18, p = 0.829, n = 150 engagements\). - [How many years have you been working in the outsourcing industry?](): Respondents typed the number of years they had worked in the outsourcing industry. The mean answer was 2.6 years and the median 1.0, from 0 to 17 \(n = 806 respondents\). An answer of zero, recorded as under one year, came from 112 respondents. - [How many years have you been working within the financial industry?](): Respondents typed the number of years they had worked within the financial industry. The mean answer was 2.1 years and the median 1.0, from 0 to 15 \(n = 789 respondents\). An answer of zero, recorded as under one year, came from 136 respondents. - [How many agentic AI outsourcing engagements have you worked on across your career, including this one?](): Respondents typed the number of agentic AI outsourcing engagements they had worked on across their careers, including the current one. The mean answer was 1.8 and the median 2.0, from 1 to 5 \(n = 799 respondents\). For 371 respondents \(46.43%\), the current engagement was the first. - [How well do you know the financial industry?](): The mean rating was 4.8 and the median 5.0 \(n = 796 respondents\). Ratings of 3 or below, the distressed range, made up 27.26% of the answers. The rating was associated with all four study measures. - [How well do you know agentic AI specific to the financial industry?](): The mean rating was 4.9 and the median 5.0 \(n = 797 respondents\). Ratings of 3 or below, the distressed range, made up 28.23% of the answers. The rating was associated with all four study measures. - [Rate your overall agentic AI expertise](): The mean rating was 4.9 and the median 5.0 \(n = 800 respondents\). Ratings of 3 or below, the distressed range, made up 27.75% of the answers. The rating was associated with all four study measures. - [What professional certifications do you have?](): Of the 850 respondents, 793 answered the question. None, a listed answer, was the most frequent \(236, 29.76%\), followed by Six Sigma Green Belt \(127\), ITIL Foundation \(126\) and AWS Certified Cloud Practitioner \(124\). Each answer was given by respondents of a single role: individual contributors gave None or one of those three certifications, engagement managers gave PMP alone or with ITIL or Scrum Master, or Six Sigma Black Belt, and portfolio directors gave PMP with an MBA or ITIL Expert, or PgMP. - [What AI certifications do you have?](): Of the 850 respondents, 788 answered the question. None, a listed answer, was given by 437 \(55.46%\). The other 351 named one of five AI certifications, each held by between 66 and 78 respondents: Databricks Generative AI Fundamentals, Google Cloud Generative AI, Azure AI Fundamentals, a DeepLearning.AI certificate and AWS Certified Machine Learning. The answers did not differ by role, vendor type or any segment family recorded at engagement level. - [Where did you gain your AI knowledge?](): Of the 850 respondents, 800 answered the question. The answers spread almost evenly across the six sources, from 125 to 140 respondents each: vendor and partner training, on the job, online courses, university coursework, self-taught and the employer's training program. The source did not differ by role, vendor type or any segment family recorded at engagement level. - [How happy are you with your current job?](): The mean rating was 4.2 and the median 4.0 \(n = 780 respondents\). Ratings of 3 or below, the distressed range, made up 46.79% of the answers. The rating was associated with all four study measures. - [How likely are you to leave your current job in the next 12 months?](): Every rating is read on the goodness view, the reversed scale on which a higher figure is the better outcome. The mean rating was 4.2 and the median 4.0 \(n = 798 respondents\). Ratings of 3 or below on the goodness view, the distressed range, made up 46.37% of the answers. The rating was associated with each of the other three study measures. The indicator is itself the measure of intent to leave. No association with intent to leave was computed. - [How happy is your extended team with the current engagement?](): The mean rating was 4.3 and the median 4.0 \(n = 788 respondents\). Ratings of 3 or below, the distressed range, made up 45.30% of the answers. The rating was associated with all four study measures. - [How likely are members of your current team to leave their jobs in the next 12 months?](): Every rating is read on the goodness view, the reversed scale on which a higher figure is the better outcome. The mean rating was 4.3 and the median 4.0 \(n = 799 respondents\). Ratings of 3 or below on the goodness view, the distressed range, made up 46.06% of the answers. The rating was associated with all four study measures. - [Compared to 5 years ago, I am happier with the work I am doing now](): The mean rating was 4.9 and the median 5.0 \(n = 804 respondents\). Ratings of 3 or below, the distressed range, made up 26.62% of the answers. The rating was associated with all four study measures. - [Compared to 5 years ago, I am more likely to seek a new job](): Every rating is read on the goodness view, the reversed scale on which a higher figure is the better outcome. The mean rating was 4.9 and the median 5.0 \(n = 801 respondents\). Ratings of 3 or below on the goodness view, the distressed range, made up 26.97% of the answers. The rating was associated with all four study measures. - [Compared to 5 years ago, the team I work with seems to be happier](): The mean rating was 4.9 and the median 5.0 \(n = 797 respondents\). Ratings of 3 or below, the distressed range, made up 30.36% of the answers. The rating was associated with all four study measures. - [Compared to 5 years ago, the team I work with is more likely to seek a new job](): Every rating is read on the goodness view, the reversed scale on which a higher figure is the better outcome. The mean rating was 4.9 and the median 5.0 \(n = 799 respondents\). Ratings of 3 or below on the goodness view, the distressed range, made up 27.66% of the answers. The rating was associated with all four study measures. - [I want to continue working on projects in the global financial industry](): The mean rating was 4.9 and the median 5.0 \(n = 784 respondents\). Ratings of 3 or below, the distressed range, made up 27.30% of the answers. The rating was associated with all four study measures. - [I would prefer to work directly for a bank in one of their global capability centers](): The statement is rated from 1 to 10 with no good or bad end. The mean rating was 6.0 and the median 6.0 \(n = 799 respondents\). The rating was associated with all four study measures. - [Working directly for a bank would be better than working for an outsourcing vendor](): The statement is rated from 1 to 10 with no good or bad end. The mean rating was 6.1 and the median 6.0 \(n = 802 respondents\). The rating was associated with all four study measures. - [I trust the customer](): The mean rating was 4.9 and the median 5.0 \(n = 799 respondents\). Ratings of 3 or below, the distressed range, made up 25.53% of the answers. The rating was associated with all four study measures. - [The customer knows and cares about me](): The mean rating was 4.9 and the median 5.0 \(n = 789 respondents\). Ratings of 3 or below, the distressed range, made up 27.38% of the answers. The rating was associated with all four study measures. - [I talk with peers inside the customer regularly](): The mean rating was 5.5 and the median 6.0 \(n = 809 respondents\). Ratings of 3 or below, the distressed range, made up 5.93% of the answers. - [I talk with engagement management inside the customer regularly](): The mean rating was 5.5 and the median 6.0 \(n = 790 respondents\). Ratings of 3 or below, the distressed range, made up 5.19% of the answers. - [I talk with the customer's leadership team regularly](): The mean rating was 5.6 and the median 6.0 \(n = 333 respondents\). Ratings of 3 or below, the distressed range, made up 6.01% of the answers. - [I know how the customer feels about my team](): The mean rating was 4.9 and the median 5.0 \(n = 798 respondents\). Ratings of 3 or below, the distressed range, made up 26.57% of the answers. The rating was associated with all four study measures. - [I know how the customer feels about the engagement I am on](): The mean rating was 4.9 and the median 5.0 \(n = 791 respondents\). Ratings of 3 or below, the distressed range, made up 27.69% of the answers. The rating was associated with all four study measures. - [My manager is effective](): The mean rating was 4.8 and the median 5.0 \(n = 802 respondents\). Ratings of 3 or below, the distressed range, made up 30.17% of the answers. The rating was associated with all four study measures. - [My manager knows the financial industry well](): The mean rating was 4.8 and the median 5.0 \(n = 800 respondents\). Ratings of 3 or below, the distressed range, made up 31.63% of the answers. The rating was associated with all four study measures. - [My manager knows agentic AI well](): The mean rating was 4.7 and the median 5.0 \(n = 803 respondents\). Ratings of 3 or below, the distressed range, made up 30.14% of the answers. The rating was associated with all four study measures. - [My manager cares about me](): The mean rating was 4.7 and the median 5.0 \(n = 804 respondents\). Ratings of 3 or below, the distressed range, made up 30.22% of the answers. The rating was associated with all four study measures. - [My manager cares about the larger team](): The mean rating was 4.7 and the median 5.0 \(n = 797 respondents\). Ratings of 3 or below, the distressed range, made up 30.99% of the answers. The rating was associated with all four study measures. - [My manager knows how to manage agentic AI outsourcing engagements](): The mean rating was 4.8 and the median 5.0 \(n = 800 respondents\). Ratings of 3 or below, the distressed range, made up 30.13% of the answers. The rating was associated with all four study measures. - [My manager knows how to manage agentic AI outsourcing engagements for global banks](): The mean rating was 4.8 and the median 5.0 \(n = 791 respondents\). Ratings of 3 or below, the distressed range, made up 30.59% of the answers. The rating was associated with all four study measures. - [My employer cares about me](): The mean rating was 5.0 and the median 5.0 \(n = 786 respondents\). Ratings of 3 or below, the distressed range, made up 24.43% of the answers. The rating was associated with all four study measures. - [My employer cares about the team](): The mean rating was 4.9 and the median 5.0 \(n = 802 respondents\). Ratings of 3 or below, the distressed range, made up 26.68% of the answers. The rating was associated with all four study measures. - [My employer's top leadership team cares about my wellbeing and development](): The mean rating was 4.9 and the median 5.0 \(n = 812 respondents\). Ratings of 3 or below, the distressed range, made up 26.35% of the answers. The rating was associated with all four study measures. - [I trust my employer](): The mean rating was 4.9 and the median 5.0 \(n = 799 respondents\). Ratings of 3 or below, the distressed range, made up 27.28% of the answers. The rating was associated with all four study measures. - [I trust my manager](): The mean rating was 4.7 and the median 5.0 \(n = 807 respondents\). Ratings of 3 or below, the distressed range, made up 30.36% of the answers. The rating was associated with all four study measures. - [I am being used effectively by my employer](): The mean rating was 4.9 and the median 5.0 \(n = 808 respondents\). Ratings of 3 or below, the distressed range, made up 26.24% of the answers. The rating was associated with all four study measures. - [Years of experience by reported intent to leave: C007 by C019](): Respondents likely to leave their job, those who answered C019 at seven or above, reported fewer years of career experience \(C007\) than respondents not likely to leave: a mean of 3.77 years \(n = 409 respondents\) against 4.44 years \(n = 337 respondents\). The difference, −0.66 years, reached significance by Welch's test \(t = −2.04, p = 0.041\) and with standard errors clustered by engagement \(t = −2.64, df = 149, p = 0.009, 150 engagements\). The effect is small \(Cohen's d = −0.15\). - [The dashboard and performance reports given to the customer are accurate](): The mean rating was 4.8 and the median 5.0 \(n = 794 respondents\). Ratings of 3 or below, the distressed range, made up 27.71% of the answers. The rating was associated with all four study measures. - [The vendor manipulates the dashboard and performance reports given to the customer](): Every rating is read on the goodness view, the reversed scale on which a higher figure is the better outcome. The mean rating was 4.7 and the median 5.0 \(n = 786 respondents\). Ratings of 3 or below on the goodness view, the distressed range, made up 29.90% of the answers. The rating was associated with all four study measures. - [Compared to 5 years ago, is manipulation of the dashboards and performance reports given to the customer more or less common now?](): Every rating is read on the goodness view, the reversed scale on which a higher figure is the better outcome. The mean rating was 4.8 and the median 5.0 \(n = 795 respondents\). Ratings of 3 or below on the goodness view, the distressed range, made up 31.07% of the answers. The rating was associated with all four study measures. - [I fully understand how these laws, regulations, and similar affect the outsourcing engagement](): The mean rating was 4.8 and the median 5.0 \(n = 800 respondents\). Ratings of 3 or below, the distressed range, made up 28.50% of the answers. The rating was associated with all four study measures. - [My management team fully understands these laws, regulations, and similar, and how they affect the engagement](): The mean rating was 4.9 and the median 5.0 \(n = 516 respondents\). Ratings of 3 or below, the distressed range, made up 28.88% of the answers. The rating was associated with all four study measures. - [The customer's team actively discusses these laws, regulations, and similar with our team](): The mean rating was 4.8 and the median 5.0 \(n = 801 respondents\). Ratings of 3 or below, the distressed range, made up 28.84% of the answers. The rating was associated with all four study measures. - [The training available to me effectively teaches me about these laws, regulations, and similar](): The mean rating was 3.6 and the median 4.0 \(n = 798 respondents\). Ratings of 3 or below, the distressed range, made up 48.25% of the answers. The rating was associated with all four study measures. - [I have altered client-provided data](): Of the 839 respondents who answered, 421 \(50.18%\) answered True and 418 \(49.82%\) answered False. At engagement level, the answer was associated with all four study measures. Engagements whose respondents most often said they had altered data provided by the financial institution were canceled more often \(61.84% of 76 engagements against 10.17% of 59\), and recorded lower customer satisfaction among active engagements, a lower engagement indicator mean and higher intent to leave. - [I discussed the data alteration with internal management](): Of the 401 respondents who answered, 292 \(72.82%\) answered True and 109 \(27.18%\) answered False. At engagement level, the answer was associated with cancellation, the engagement indicator mean and intent to leave. Engagements whose respondents most often said they had discussed the data alteration with internal management were canceled less often \(39.77% of 88 engagements against 64.00% of 25\), recorded a higher engagement indicator mean and had lower intent to leave among their respondents. - [I or my management discussed data alteration with the client](): Of the 366 respondents who answered, 3 \(0.82%\) answered True and 363 \(99.18%\) answered False. No association with the study measures was tested, because fewer than two answers were assigned to at least ten engagements. The share answering True did not differ by any segment family after Holm correction. Of the 109 written responses, 96 were negative. At least 53 of the 96 negative responses said the financial institution was not told, or that the alteration stayed within the vendor team. - [My greater team regularly alters client provided data](): The statement is rated from 1 to 10 with no good or bad end. The mean rating was 5.4 and the median 5.0 \(n = 773 respondents\). The rating was associated with all four study measures. - [I have altered client processes](): Of the 841 respondents who answered, 490 \(58.26%\) answered True and 351 \(41.74%\) answered False. At engagement level, the answer was associated with all four study measures. Engagements whose respondents most often said they had altered the financial institution's processes were canceled more often \(50.00% of 94 engagements against 15.22% of 46\), and recorded lower customer satisfaction among active engagements, a lower engagement indicator mean and higher intent to leave. - [I discussed the process alteration with internal management](): Of the 459 respondents who answered, 341 \(74.29%\) answered True and 118 \(25.71%\) answered False. At engagement level, the answer was associated with cancellation, the engagement indicator mean and intent to leave. Engagements whose respondents most often said they had discussed the process alteration with internal management were canceled less often \(35.00% of 100 engagements against 65.22% of 23\), recorded a higher engagement indicator mean and had lower intent to leave among their respondents. - [I or my management discussed process alteration with the client](): Of the 414 respondents who answered, 9 \(2.17%\) answered True and 405 \(97.83%\) answered False. No association with the study measures was tested, because fewer than two answers were assigned to at least ten engagements. The share answering True did not differ by any segment family after Holm correction. Of the 124 written responses, 99 were negative. At least 57 of the 99 negative responses said the financial institution was not told, or that the alteration stayed within the vendor team. - [My greater team regularly alters client processes](): The statement is rated from 1 to 10 with no good or bad end. The mean rating was 5.7 and the median 6.0 \(n = 779 respondents\). The rating was associated with all four study measures. - [I have altered AI outputs](): Of the 828 respondents who answered, 278 \(33.57%\) answered True and 550 \(66.43%\) answered False. At engagement level, the answer was associated with all four study measures. Engagements whose respondents most often said they had altered AI outputs were canceled more often \(76.19% of 42 engagements against 22.00% of 100\), and recorded lower customer satisfaction among active engagements, a lower engagement indicator mean and higher intent to leave. - [I discussed the AI output alteration with internal management](): Of the 253 respondents who answered, 150 \(59.29%\) answered True and 103 \(40.71%\) answered False. At engagement level, the answer was not associated with any study measure tested. The share answering True did not differ by any segment family after Holm correction. Of the 95 written responses, 67 were positive and 26 negative. The positive responses, nearly all under C216, said the respondent had told the team lead or manager, and some added that the financial institution was not told. Most negative responses came from respondents who had told no one, or no one above them. - [I or my management discussed AI output alteration with the client](): Of the 224 respondents who answered, 1 \(0.45%\) answered True and 223 \(99.55%\) answered False. No association with the study measures was tested, because fewer than two answers were assigned to at least ten engagements. The share answering True did not differ by any segment family after Holm correction. Of the 50 written responses, 43 were negative. At least 22 of the 43 negative responses said the financial institution was not told, or that the correction stayed within the vendor team. - [My greater team regularly alters AI outputs](): The statement is rated from 1 to 10 with no good or bad end. The mean rating was 4.4 and the median 4.0 \(n = 665 respondents\). The rating was associated with all four study measures. - [The three alteration questions by respondent role: C199, C205, C211 by C005](): Three Part 5 questions asked respondents whether they had altered data provided by the financial institution \(C199\), the financial institution's processes \(C205\) and AI outputs \(C211\). Saying True was almost entirely confined to individual contributors. Among engagement managers the share saying True was 6.08% or lower on each question, and no portfolio director said True to any of the three. The difference between roles reached significance on all three questions. - [The three alteration questions as reported by managers and directors: C199, C205, C211 by C005](): The analysis sets the answers of engagement managers and portfolio directors on the three alteration questions \(C199, C205 and C211\) beside those of individual contributors, together with the paired questions on whether the alteration was discussed with internal management \(C200, C206 and C212\) and with the financial institution \(C201, C207 and C213\). Answers are summed across the three questions in each group. A respondent can contribute up to three answers. Engagement managers gave 14 True answers across the three alteration questions, against 1,175 from individual contributors. - [Agentic AI Outsourcing at Tier 1 Financial Institutions: Introduction — The Study and How to Read It](): PDF. Research on 150 agentic AI outsourcing engagements serving Tier 1 financial institutions, based on 850 vendor delivery-team respondents in January–June 2026. The study examines outcomes, five failure categories, governance, the delivery workforce, manual alteration, and proposed executive controls. Its cross-sectional, nonrepresentative, delivery-side evidence is not independently validated by financial institutions and does not establish causation or legal breach. The Introduction establishes the study design, evidentiary limits, definitions, four core study measures, rating conventions, statistical methods, and organization of the report. - [Executive Summary](): Principal outcomes, manual alteration, governance, experience, and process redesign. Evidence from 150 agentic AI outsourcing engagements. - [Study Design, Evidentiary Basis, and Definitions](): Sample composition, evidentiary limits, definitions, and analytical conventions. Evidence from 150 agentic AI outsourcing engagements. - [The Immaturity of Agentic AI Outsourcing](): Lifecycle position, operating scale, and limited agentic AI delivery experience. Evidence from 150 agentic AI outsourcing engagements. - [Cancellation and the Risk of Cancellation](): Realized cancellation and delivery-team estimates of risk among active engagements. Evidence from 150 agentic AI outsourcing engagements. - [Customer Satisfaction, Delivery Performance, and Measured Return](): Reported satisfaction, dates, budgets, key performance indicators, and recorded return. Evidence from 150 agentic AI outsourcing engagements. - [The Engagement Indicator Mean and the Interdependence of the Four Measures](): Shared variation across cancellation, satisfaction, indicator ratings, and intent to leave. Evidence from 150 agentic AI outsourcing engagements. - [Structural Attributes and Outcome Associations](): Financial institutions, relationships, work, commercial terms, vendors, and team size. Evidence from 150 agentic AI outsourcing engagements. - [Comparative Standing of the Five Failure Categories](): Relative ratings, distressed answers, and shared associations across five categories. Evidence from 150 agentic AI outsourcing engagements. - [Financial Institution Readiness](): Data, systems, processes, skills, access, and institutional ownership. Evidence from 150 agentic AI outsourcing engagements. - [Co-Management](): Joint decisions, risk management, change control, human review, and agent management. Evidence from 150 agentic AI outsourcing engagements. - [Vendor Capability](): Leadership, management processes, knowledge, experience, talent, and delivery practices. Evidence from 150 agentic AI outsourcing engagements. - [AI Technology](): Reliability, memory, cost, information protection, authorized action, and observability. Evidence from 150 agentic AI outsourcing engagements. - [Third-Party Ecosystem](): Provider changes, interfaces, contracts, security, fallback capacity, and external dependencies. Evidence from 150 agentic AI outsourcing engagements. - [Reported Governance Weaknesses Across All 150 Engagements](): The governance score across engagements, segments, lifecycle stages, and study measures. Evidence from 150 agentic AI outsourcing engagements. - [Co-Management and Governance Across the Other Four Categories](): Joint decisions, risks, changes, review, and management of agent work. Evidence from 150 agentic AI outsourcing engagements. - [Financial Institution Readiness and the Governance of Agentic Work](): Data readiness, institutional ownership, agent governance, and authorized access. Evidence from 150 agentic AI outsourcing engagements. - [Vendor Preparation and Accountability in the Control Environment](): Vendor playbooks, quality assurance, and accountability for agent error. Evidence from 150 agentic AI outsourcing engagements. - [Technical Controls and Governance Across the Engagement](): Information protection, guardrails, testability, and observability. Evidence from 150 agentic AI outsourcing engagements. - [Third-Party Dependencies and the Limits of Internal Control](): Provider changes, contracts, concentration, fallback capacity, and external rails. Evidence from 150 agentic AI outsourcing engagements. - [Understanding Governing Obligations and the Control Environment](): Delivery-team understanding of the obligations governing the engagement. Evidence from 150 agentic AI outsourcing engagements. - [The Fit Between Governance Processes and the Pace of Agent Work](): Decision speed, review capacity, inherited processes, and the pace of agent work. Evidence from 150 agentic AI outsourcing engagements. - [Manual Alteration and the Operation and Visibility of Controls](): Reporting, escalation, disclosure, and the visibility of intervention. Evidence from 150 agentic AI outsourcing engagements. - [Practices Adopted in Place of Controls](): Workarounds adopted as routine practice and remedies proposed by respondents. Evidence from 150 agentic AI outsourcing engagements. - [Composition and Experience of the Delivery Workforce](): Role composition, career experience, agentic AI experience, and self-assessed knowledge. Evidence from 150 agentic AI outsourcing engagements. - [The Manager](): Reported manager effectiveness and financial-industry and agentic AI knowledge. Evidence from 150 agentic AI outsourcing engagements. - [Job Satisfaction](): Current and retrospective satisfaction, fatigue, challenges, and sources of support. Evidence from 150 agentic AI outsourcing engagements. - [Trust, Care, and Contact](): Trust, care, and contact across the financial institution, manager, and vendor. Evidence from 150 agentic AI outsourcing engagements. - [Intent to Leave](): Intent to leave, job seeking, financial-industry commitment, and attitudes to manual alteration. Evidence from 150 agentic AI outsourcing engagements. - [Knowledge at Risk of Departure](): Experience and undocumented manual practices held by staff likely to leave. Evidence from 150 agentic AI outsourcing engagements. - [The Five Manual Alteration Methods](): Intervention in data, processes, AI outputs, reporting, and third-party work. Evidence from 150 agentic AI outsourcing engagements. - [Manual Alteration by Cancellation Status and Engagement Indicator Mean](): Comparisons of canceled and active engagements and their indicator means. Evidence from 150 agentic AI outsourcing engagements. - [Distribution of Manual Alteration Across Roles](): Individual contributors, engagement managers, and portfolio directors. Evidence from 150 agentic AI outsourcing engagements. - [Reasons Given for Manual Alteration](): Delivery-team explanations for the five manual alteration methods. Evidence from 150 agentic AI outsourcing engagements. - [Disclosure of Manual Alteration](): Internal discussion and financial institution disclosure of manual alteration. Evidence from 150 agentic AI outsourcing engagements. - [Recording of Manual Alteration](): Systems of record and the visibility of manual work in reported performance. Evidence from 150 agentic AI outsourcing engagements. - [Regulatory and Legal Context Reported by Respondents](): Respondent-named obligations and the work affected by manual alteration. Evidence from 150 agentic AI outsourcing engagements. - [Regulatory and Legal Knowledge of the Delivery Team](): Reported knowledge, management understanding, institutional discussion, and training. Evidence from 150 agentic AI outsourcing engagements. - [Recommended Controls](): Seven proposed controls for readiness, co-management, reporting, oversight, and manual alteration. Evidence from 150 agentic AI outsourcing engagements. - [Proposed Implementation Schedule](): A staged audit and remediation program from Week 1 through the first year. Evidence from 150 agentic AI outsourcing engagements. - [Appendix A: Definitions](): Controlled terminology and principal definitions used across the study. Evidence from 150 agentic AI outsourcing engagements. - [Appendix B: Indicator Inventory](): Question wording, indicator codes, analytical classes, and report locations. Evidence from 150 agentic AI outsourcing engagements. - [Appendix C: Tables and the Indicators Behind Them](): Source indicators and answering bases supporting the numbered exhibits. Evidence from 150 agentic AI outsourcing engagements. - [Appendix D: Method](): Survey design, sampling, measures, statistical methods, coding, and limitations. Evidence from 150 agentic AI outsourcing engagements. - [Appendix E: Companion Indicator Statistical Analyses](): A register of the 216 detailed analyses underlying the report. Evidence from 150 agentic AI outsourcing engagements. - [Appendix F: Index](): A subject guide to the report’s topics, indicators, and methods. Evidence from 150 agentic AI outsourcing engagements. - [Appendix G: Disclosure of the Use of Artificial Intelligence](): AI tools in analysis, coding, editorial preparation, and verification. Evidence from 150 agentic AI outsourcing engagements. - [Agentic AI Outsourcing at Tier 1 Financial Institutions: Appendices — Definitions, Indicators, Methods, and Disclosures](): PDF. Research on 150 agentic AI outsourcing engagements serving Tier 1 financial institutions, based on 850 vendor delivery-team respondents in January–June 2026. The study examines outcomes, five failure categories, governance, the delivery workforce, manual alteration, and proposed executive controls. Its cross-sectional, nonrepresentative, delivery-side evidence is not independently validated by financial institutions and does not establish causation or legal breach. The Appendices contain study definitions, the indicator inventory, the exhibit inventory, methods and limitations, the register of 216 companion analyses, the index, and disclosure of artificial intelligence use. - [Agentic AI Outsourcing at Tier 1 Financial Institutions: Complete Study](): PDF. Agentic AI outsourcing at Tier 1 financial institutions depended on extensive human intervention within an immature delivery model. This study examines 150 engagements through responses from 850 members of vendors’ delivery teams collected between January and June 2026. It connects engagement outcomes with financial institution readiness, co-management, vendor capability, AI technology, the third-party ecosystem, governance, and workforce conditions. The central concern is whether reported automation performance adequately represents the human work and control arrangements required to sustain delivery. By the study date, 56 of 150 engagements had been canceled \(37.33%\). A further 58 of the 94 active engagements met the study’s threshold for at least moderate perceived cancellation risk. Vendor-reported customer satisfaction averaged 43.85 points on a 100-point scale across all 150 engagements. Positive recorded returns were reported for 17 of the 30 engagements in Production, with a mean of 24.71% among those 17. These figures were not independently verified financial results. Only 84 of 782 respondents \(10.74%\) rated their engagement as achieving its stated key performance indicators under the study’s rating criterion. Manual alteration was reported across all 150 engagements when the five methods were combined. Delivery teams altered financial-institution-provided data, financial institution processes, AI outputs, reporting, and third-party work. Respondents described interventions that sustained delivery but were often absent from formal records or reported performance. Discussion with the financial institution was reported in 13 of 1,004 valid answers to the structured disclosure questions \(1.29%\). This denominator counts answers to the disclosure items; it does not count distinct respondents or engagements. Organizational conditions received less favorable ratings than technology and third-party conditions. Financial institution readiness, co-management, and vendor capability had category means of 3.55, 3.65, and 3.67 on the 10-point scale. The corresponding means were 4.60 for AI technology and 4.64 for the third-party ecosystem. Written accounts described decision delays, limited review capacity, weak change control, and operating practices that had developed outside the formal control framework. Limited agentic AI experience and resistance to process redesign provided further context. The findings support an independent audit of live agentic AI outsourcing engagements, including their underlying manual work, reporting calculations, decision rights, and review capacity. The sample is not established as representative of the industry. Evidence comes from vendor delivery staff and study records; financial institutions and other parties did not independently validate it. Cross-sectional associations do not establish causation, and the study does not determine a legal or regulatory breach. The report provides a detailed basis for institutional examination rather than an industry-wide failure estimate. - [Agentic AI Outsourcing at Tier 1 Financial Institutions: Part 1 — Engagement Outcomes and Their Correlates](): PDF. Research on 150 agentic AI outsourcing engagements serving Tier 1 financial institutions, based on 850 vendor delivery-team respondents in January–June 2026. The study examines outcomes, five failure categories, governance, the delivery workforce, manual alteration, and proposed executive controls. Its cross-sectional, nonrepresentative, delivery-side evidence is not independently validated by financial institutions and does not establish causation or legal breach. Part 1 examines engagement maturity, cancellation, perceived cancellation risk, customer satisfaction, delivery performance, recorded return, and structural attributes associated with the four study measures. - [Agentic AI Outsourcing at Tier 1 Financial Institutions: Part 2 — The Five Failure Categories](): PDF. Research on 150 agentic AI outsourcing engagements serving Tier 1 financial institutions, based on 850 vendor delivery-team respondents in January–June 2026. The study examines outcomes, five failure categories, governance, the delivery workforce, manual alteration, and proposed executive controls. Its cross-sectional, nonrepresentative, delivery-side evidence is not independently validated by financial institutions and does not establish causation or legal breach. Part 2 provides separate, exhaustive indicator analysis across financial institution readiness, co-management, vendor capability, AI technology, and the third-party ecosystem. - [Agentic AI Outsourcing at Tier 1 Financial Institutions: Part 3 — Governance, Oversight, and Control](): PDF. Research on 150 agentic AI outsourcing engagements serving Tier 1 financial institutions, based on 850 vendor delivery-team respondents in January–June 2026. The study examines outcomes, five failure categories, governance, the delivery workforce, manual alteration, and proposed executive controls. Its cross-sectional, nonrepresentative, delivery-side evidence is not independently validated by financial institutions and does not establish causation or legal breach. Part 3 examines how governance, oversight, and control operate across organizational readiness, joint management, technical controls, third-party dependencies, obligations, reporting, and manual alteration. - [Agentic AI Outsourcing at Tier 1 Financial Institutions: Part 4 — The Delivery Workforce](): PDF. Research on 150 agentic AI outsourcing engagements serving Tier 1 financial institutions, based on 850 vendor delivery-team respondents in January–June 2026. The study examines outcomes, five failure categories, governance, the delivery workforce, manual alteration, and proposed executive controls. Its cross-sectional, nonrepresentative, delivery-side evidence is not independently validated by financial institutions and does not establish causation or legal breach. Part 4 examines delivery workforce composition, experience, managers, satisfaction, trust, contact, intent to leave, and the operational knowledge potentially lost through departure. - [Agentic AI Outsourcing at Tier 1 Financial Institutions: Part 5 — Manual Alteration by the Delivery Team](): PDF. Research on 150 agentic AI outsourcing engagements serving Tier 1 financial institutions, based on 850 vendor delivery-team respondents in January–June 2026. The study examines outcomes, five failure categories, governance, the delivery workforce, manual alteration, and proposed executive controls. Its cross-sectional, nonrepresentative, delivery-side evidence is not independently validated by financial institutions and does not establish causation or legal breach. Part 5 examines five manual alteration methods, their distribution and stated reasons, disclosure, recording, and the regulatory and legal knowledge reported by delivery staff. - [Agentic AI Outsourcing at Tier 1 Financial Institutions: Part 6 — The Executive Action Plan](): PDF. Research on 150 agentic AI outsourcing engagements serving Tier 1 financial institutions, based on 850 vendor delivery-team respondents in January–June 2026. The study examines outcomes, five failure categories, governance, the delivery workforce, manual alteration, and proposed executive controls. Its cross-sectional, nonrepresentative, delivery-side evidence is not independently validated by financial institutions and does not establish causation or legal breach. Part 6 proposes seven institutional control actions and a staged implementation schedule extending from the first week through the first year. ## Machine readable resources - [Full text corpus](https://akholi.com/llms-full.txt): Complete available text; unavailable exports are explicitly identified. - [Corpus manifest](https://akholi.com/akholi-manifest.json): Source URLs, versions, export availability, and checksums.