Composition and Experience of the Delivery Workforce

The respondent sample was dominated by individual contributors with limited career and agentic AI experience. It comprised 667 individual contributors (78.47%), 150 engagement managers (17.65%), 31 portfolio directors (3.65%), and two customer management executives. Each of the 150 engagements was represented by one engagement manager and a majority of individual contributors. These proportions describe the sample, rather than the composition of every delivery team.

Career experience, outsourcing-industry experience, and financial-industry experience are examined separately from agentic AI experience. Every recorded answer on years of agentic AI experience was zero or one. This restricted range limits the comparisons that can be made.

“I might not have been working as long as them but I have more AI experience than my entire management team combined.” -Individual Contributor, BPO, Payments, cards and disputes

Experience differed by role, and role was associated with respondents’ ratings and intent to leave. The unadjusted experience comparisons therefore do not isolate the contribution of experience from that of role. Engagement-level associations for career and industry experience met the study’s equivalence criteria. Equivalence tests assess whether an association lies within prespecified bounds, here −0.30 to +0.30 for the correlation coefficient, r. The coefficient measures the strength and direction of a linear relationship, from −1 to +1. Equivalence within these bounds does not establish a zero relationship or the absence of operational importance.

Got put on the account with no training on the agent tooling at all, I learned it off a guy who ran it on a diferent account. The course they did put me on was banking basics I already knew”

INDIVIDUAL CONTRIBUTOR · ITO, Technology engineering and operations

1 Role Composition of the Engagements

Individual contributor was the most frequent role in every engagement’s respondent sample. Assigning each engagement its most frequent role consequently produced the same classification for all 150 engagements. With no variation between engagements, this classification could not support an engagement-level comparison. Respondent-level comparisons nevertheless identified differences among roles.

1.1 Role Composition of the Respondent Sample

Indicator C005 · What is your role on this engagement? · Full statistical analysis ›

Individual contributors formed the majority of respondents, and reported intent to leave differed by role. All 850 respondents answered the role question. Individual contributors and engagement managers were represented on all 150 engagements. Portfolio directors were represented on 31 engagements (20.67%). A role’s absence from the respondent sample does not establish its absence from the delivery team.

Respondent-level tests used standard errors, estimates of uncertainty, clustered by engagement to account for similarities among answers from respondents on the same engagement. The p value is the probability, under the null hypothesis of no difference or association, of a result at least as extreme as the one observed. Differences by role were statistically significant for respondents’ own ratings and intent to leave (p < 0.001 for each). Ratings of intent to leave at 7 or above were reported by 342 of 616 individual contributors (55.52%), 75 of 150 engagement managers (50.00%), and 13 of 31 portfolio directors (41.94%). Comparisons of cancellation and customer satisfaction did not identify statistically significant differences by respondent role. These comparisons describe the engagements on which respondents worked and do not attribute their outcomes to a particular role.

Role was addressed in 22 written answers from 22 respondents on 21 engagements. Seven answers were coded as negative. Six of those seven described decisions that respondents considered beyond their role but made because neither of the two firms would decide. Written answers were coded by subject and sentiment. Sentiment scores took five values, −10, −5, 0, +5, and +10, with negative values denoting unfavorable accounts and positive values favorable accounts. These scores describe the written answers, rather than an independently observed outcome.

“I end up being the person who decide who owns what, and that is not really my job level. My team lead take it fine but I notice the newer guys ask me instead of asking the client side, so now im the middle man for surveillance rule questions.” -Individual Contributor, ITO, Regulatory, audit, tax and legal

“Same story every week, nobody wants to be the one who says yes so I decide. Its not my job to own that call but somebody has to” -Individual Contributor, CXM, Technology engineering and operations

“I dont say much on this one, but it cost me sleep some week, numbers move and I am the one who decide what go to client file” -Individual Contributor, BPO, Corporate and back-office functions

The problem is process readiness on their side is genuinely poor, and the reconciliation team re-keys counterparty static data most weeks just to keep the break queue moving. i do not touch it myself now, my role is managing how i frame it upward, as far as i have visibility, and kindly worded updates do not seem to change anything on the client side.”

ENGAGEMENT MANAGER · BPO, Trade, settlement and reconciliation

2 Career, Outsourcing, and Financial Industry Experience

Reported career, outsourcing-industry, and financial-industry experience were limited and closely related to role. Each was recorded in years. A recorded zero was classified as less than one year, rather than proof of no experience.

For all three experience questions, engagement-level associations with cancellation, customer satisfaction, and the engagement indicator mean met the study’s equivalence criteria. The small respondent-level associations with intent to leave also met those criteria. Because experience was closely related to role, these unadjusted results do not establish its independent contribution.

2.1 Career Experience of the Delivery Workforce

Indicator C007 · How many years of experience do you have in your career? · Full statistical analysis ›

Experience differed substantially by role. Reported career experience had a mean of 3.9 years and a median of 2.0 years, with recorded values from 0 to 25 (n = 797 respondents). Sixty-one respondents recorded zero, classified as less than one year. The mean was 1.9 years among individual contributors (n = 614 respondents), 10.0 among engagement managers (n = 150), and 15.1 among portfolio directors (n = 31).

Career experience was not statistically significantly associated at engagement level with cancellation (r = −0.11, p = 0.191, n = 150 engagements), customer satisfaction among active engagements (r = 0.03, p = 0.757, n = 94 engagements), or the engagement indicator mean (r = 0.10, p = 0.208, n = 150 engagements). All three met the equivalence criteria. Greater career experience was associated at respondent level with lower intent to leave (r = −0.09, cluster bootstrap 95% CI −0.15 to −0.04, p = 0.001, n = 746 respondents), also within the equivalence bounds. A confidence interval (CI) expresses uncertainty around an estimate. The cluster bootstrap resampled whole engagements to preserve the grouping of respondents.

Four written answers addressed career experience. Three were from respondents early in their careers. These accounts provide illustrations of reported experience and do not establish its prevalence.

“Im a fresher and know more about AI than anyone in our executive team.” -Individual Contributor, KPO, Corporate and back-office functions

2.2 Experience in the Outsourcing Industry

Indicator C009 · How many years have you been working in the outsourcing industry? · Full statistical analysis ›

Reported outsourcing-industry experience was limited, with a mean of 2.6 years and a median of 1.0 year, with recorded values from 0 to 17 (n = 806 respondents). For 112 respondents, the recorded zero represented less than one year of experience.

At engagement level, outsourcing-industry experience was not statistically significantly associated with cancellation (r = −0.04, p = 0.617, n = 150 engagements), customer satisfaction among active engagements (r = 0.06, p = 0.573, n = 94 engagements), or the engagement indicator mean (r = 0.07, p = 0.415, n = 150 engagements). All three associations met the equivalence criteria. At respondent level, greater outsourcing-industry experience was associated with lower intent to leave (r = −0.07, cluster bootstrap 95% CI −0.14 to −0.01, p = 0.025, n = 758 respondents). The association was within the equivalence bounds and may reflect the relationship between experience and role.

2.3 Experience in the Financial Industry

Indicator C010 · How many years have you been working within the financial industry? · Full statistical analysis ›

Reported financial-industry experience was limited, with a mean of 2.1 years and a median of 1.0 year, with recorded values from 0 to 15 (n = 789 respondents). A zero, classified as less than one year of experience, was recorded by 136 respondents.

At engagement level, financial-industry experience was not statistically significantly associated with cancellation (r = −0.03, p = 0.684, n = 150 engagements), customer satisfaction among active engagements (r = −0.08, p = 0.470, n = 94 engagements), or the engagement indicator mean (r = 0.00, p = 0.955, n = 150 engagements). All three associations met the equivalence criteria. At respondent level, the association with intent to leave was also within the equivalence bounds (r = −0.06, cluster bootstrap 95% CI −0.12 to 0.00, p = 0.033, n = 742 respondents). It did not meet the criteria in three robustness checks, which repeat an analysis using alternative assumptions or samples. The respondent-level association therefore requires particular caution.

One written answer addressed financial-industry experience. It described a delivery team without banking experience learning financial tasks from tickets. This account is illustrative.

“Send us on proper payroll training first… nobody in my team ever worked in bank before, we learn from tickets” -Individual Contributor, BPO, Corporate and back-office functions

3 Agentic AI Experience

Agentic AI experience was limited across roles, and nearly half of respondents were working on their first such engagement. The mean recorded response was 0.4 for individual contributors (n = 611), 0.5 for engagement managers (n = 150), and 0.3 for portfolio directors (n = 30). Because every answer was zero or one, these means summarize the recorded categories rather than precise elapsed durations.

The number of agentic AI outsourcing engagements worked on provides a separate measure of experience. It includes the current engagement and is examined independently of reported years of experience.

3.1 Reported Years of Agentic AI Experience

Indicator C008 · How many years of experience do you have with agentic AI? · Full statistical analysis ›

No respondent reported more than one year of agentic AI experience. Of the 793 respondents who reported years of agentic AI experience, 467 (58.89%) recorded zero, classified as less than one year, and 326 (41.11%) recorded one year. Within this restricted range, no statistically significant difference by role was identified.

At engagement level, reported agentic AI experience was not statistically significantly associated with cancellation (r = 0.00, p = 0.993, n = 150 engagements), customer satisfaction among active engagements (r = 0.13, p = 0.220, n = 94 engagements), or the engagement indicator mean (r = 0.02, p = 0.829, n = 150 engagements). Its respondent-level association with intent to leave was also not statistically significant (r = −0.04, p = 0.232, n = 746 respondents).

All four associations met the equivalence criteria, and the reported conclusions were retained under the applicable robustness checks. No statistically significant difference in reported agentic AI experience was identified across the segment families examined. The restricted range prevents these results from establishing the value of experience beyond the period represented in the study.

3.2 Agentic AI Outsourcing Engagements Worked On

Indicator C011 · How many agentic AI outsourcing engagements have you worked on across your career, including this one? · Full statistical analysis ›

For 371 of the 799 respondents (46.43%), the current engagement was their first agentic AI outsourcing engagement. The mean number of agentic AI outsourcing engagements worked on, including the current engagement, was 1.8. The median was 2.0, with recorded values from 1 to 5 (n = 799 respondents). The mean was 2.5 engagements for engagement managers (n = 150 respondents), 2.4 for portfolio directors (n = 31), and 1.7 for individual contributors (n = 617).

At engagement level, the number of agentic AI outsourcing engagements worked on was not statistically significantly associated with cancellation (r = −0.05, n = 150 engagements), customer satisfaction among active engagements (r = −0.02, n = 94 engagements), or the engagement indicator mean (r = 0.07, n = 150 engagements). The respondent-level association with intent to leave was also not statistically significant (r = −0.03, n = 751 respondents). All four associations met the equivalence criteria, and the conclusions were unchanged under the reported robustness checks.

Eight written answers addressed experience on previous engagements. They described respondents drawing comparisons with earlier engagements and provide illustrative accounts of that experience.

“I have seen one engagement only. What I can say is I got no sanctions training at all before I started, I learnt the lists from a colleague.” -Individual Contributor, KPO, Financial crime and client onboarding

“i have run four of these now. on the account before this one, a card operations desk, we had a standing call every tuesday with one person from their side who could say yes or no on the spot, and if she said no it was no and we moved on.” -Engagement Manager, CXM, Corporate and back-office functions

4 Self-Assessed Knowledge and Expertise

Respondents rated their financial-industry knowledge, knowledge of agentic AI in financial services, and overall agentic AI expertise below the scale’s midpoint. Higher ratings were associated with more favorable results on the four core study measures. These self-assessments describe perceived capability rather than independently tested knowledge. A distressed rating is at the unfavorable end of the question’s 10-point scale. It is 3 or below when lower ratings are unfavorable and 8 or above when higher ratings are unfavorable. At engagement level, the same thresholds apply to the mean rating.

Each self-assessment was associated with the four core study measures, with the conclusions retained under the applicable robustness checks. Correlations with the engagement indicator mean ranged from r = 0.78 to r = 0.80 (n = 150 engagements). None of the three questions is included in that mean. Shared perceptions of the engagement may contribute to the observed relationships, and the study does not distinguish their contribution from differences in knowledge.

Written answers described learning through colleagues, financial institution material, and experience during delivery. Some respondents valued that learning despite other difficulties on the engagement.

“None of us came from wealth management, so when the tool flags an advisor call for suitability we are reading the rule for the first time ourselves. Thus people stay late going through the clients policy pack on their own time. Nobody trained us for this, we taught each other” -Individual Contributor, CXM, Credit, underwriting and suitability

4.1 Self-Rated Knowledge of the Financial Industry

Indicator C012 · How well do you know the financial industry? · Full statistical analysis ›

Self-rated financial-industry knowledge received distressed ratings in 217 of 796 answers (27.26%). The mean rating was 4.8 and the median was 5.0 (n = 796 respondents). Respondents on canceled engagements reported a mean of 3.5 (n = 309), below the means in the three active lifecycle stages.

At engagement level, higher self-rated financial-industry knowledge was associated with a lower cancellation rate (r = −0.61, n = 150 engagements), higher customer satisfaction among active engagements (r = 0.77, n = 94 engagements), and a higher engagement indicator mean (r = 0.80, n = 150 engagements). At respondent level, it was associated with lower intent to leave (r = −0.76, n = 750 respondents). The conclusions were retained under the applicable robustness checks. The mean rating was distressed on 25 of 150 engagements.

Thirty written answers addressed financial-industry knowledge. Sixteen came from comments on vendor capability and 14 from the question about valued aspects of the job. Eighteen were positive and 10 negative. Comments on vendor capability described respondents acquiring financial-industry knowledge from colleagues during delivery. Answers about valued aspects of the job identified that knowledge as a benefit of the work.

“I come on this desk knowing nothing about settlement, they give me two e-learning module and a recording, and after that i am on live breaks in same week. My senior teach me, not company.” -Individual Contributor, BPO, Trade, settlement and reconciliation

“i had one week of a slide deck about banking and then i am on a private equity reporting job with no idea what a capital call is. laptop and access come quick, that part they did well, i give them that. everything else i learn from a colleague who is also new.” -Individual Contributor, KPO, Corporate and back-office functions

4.2 Self-Rated Knowledge of Agentic AI in the Financial Industry

Indicator C013 · How well do you know agentic AI specific to the financial industry? · Full statistical analysis ›

Self-rated knowledge of agentic AI in the financial industry received distressed ratings in 225 of 797 answers (28.23%). The mean rating was 4.9 and the median was 5.0 (n = 797 respondents). Respondents on canceled engagements reported a mean of 3.6 (n = 306).

Higher ratings were associated with a lower cancellation rate (r = −0.57, n = 150 engagements), higher customer satisfaction among active engagements (r = 0.76, n = 94 engagements), and a higher engagement indicator mean (r = 0.78, n = 150 engagements). At respondent level, higher ratings were associated with lower intent to leave (r = −0.75, n = 748 respondents). The conclusions for all four associations were retained under the applicable robustness checks. The mean rating was distressed on 22 of 150 engagements.

Seven written answers addressed knowledge of agentic AI in financial services, and all were positive. Each described learning during the engagement. Three also described cancellation or substantial delivery difficulties. These accounts illustrate the reported value of learning without establishing the effectiveness of the engagement.

“learned agentic payroll operations … team stayed steady through the shutdown” -Engagement Manager, BPO, Corporate and back-office functions

4.3 Self-Rated Agentic AI Expertise

Indicator C014 · Rate your overall agentic AI expertise · Full statistical analysis ›

Self-rated agentic AI expertise received distressed ratings in 222 of 800 answers (27.75%). The mean rating was 4.9 and the median was 5.0 (n = 800 respondents). Respondents on canceled engagements reported a mean of 3.5 (n = 308).

Higher self-rated expertise was associated with a lower cancellation rate (r = −0.61, n = 150 engagements), higher customer satisfaction among active engagements (r = 0.74, n = 94 engagements), and a higher engagement indicator mean (r = 0.78, n = 150 engagements). At respondent level, higher ratings were associated with lower intent to leave (r = −0.78, n = 753 respondents). The conclusions were retained under the applicable robustness checks. The mean rating was distressed on 26 of 150 engagements.

Thirty-seven written answers addressed agentic AI expertise. Thirty came from the question about valued aspects of the job and seven from comments on vendor capability. Twenty-four were positive and four negative. All 30 answers about valued aspects of the job identified agentic AI skills acquired during the engagement, most often quality assurance of agent output. Comments on vendor capability described brief training, instruction through videos, and respondents who considered their own AI knowledge greater than that of people in more senior roles.

“The technical stretch of learning how to evaluate an agent's own output” -Engagement Manager, ITO, Technology engineering and operations

“… was my first job out of university. They sent me to training. I knew more about AI than the trainer.” -Individual Contributor, BPO, Corporate and back-office functions

the training was two days of slides on the ticket tool and nothing about how the ai decides a routing, so i am learning from seniors on the floor. it slows me down and i feel behind. rest of it i have no complaint about”

INDIVIDUAL CONTRIBUTOR · CXM, Corporate and back-office functions

5 Certifications and Sources of AI Knowledge

Professional certifications closely followed role, more than half of those answering the AI-certification question reported none, and source-of-knowledge comparisons did not identify statistically significant outcome differences. These findings do not isolate preparation from role or establish the contribution of certification to engagement outcomes.

“what they did is put five of us on agent test automation with nobody who has built it before, my lead learn it from the vendor documents same as me. we had one trainer for whole floor and his diary is full for weeks.” -Individual Contributor, ITO, Technology engineering and operations

5.1 Professional Certifications and Respondent Role

Indicator C015 · What professional certifications do you have? · Full statistical analysis ›

Professional-certification answer categories each occurred within a single role, so their associations could not be separated from role in these comparisons. Professional certifications were reported by 793 of 850 respondents. Of those who answered, 236 (29.76%) reported none. The most frequent named certifications were Six Sigma Green Belt (127 respondents), Information Technology Infrastructure Library (ITIL) Foundation (126), and Amazon Web Services (AWS) Certified Cloud Practitioner (124).

Individual contributors reported no certification or one of the three certifications identified above. Engagement managers reported Project Management Professional (PMP), alone or with ITIL or Scrum Master, or Six Sigma Black Belt. Portfolio directors reported PMP with a Master of Business Administration (MBA) or ITIL Expert, or Program Management Professional (PgMP). Engagement-level comparisons assigned the unique most frequent answer to each engagement. Tied assignments and answer categories below the required base were excluded.

The chi-square test assesses whether the distribution across categories differs from that expected under independence. Welch’s F test compares group means while allowing the spread of values within each group to differ. Engagement-level comparisons of professional certifications did not identify statistically significant differences in cancellation (chi-square 4.29, p = 0.235, n = 99 engagements), customer satisfaction among active engagements (Welch F 0.00, p = 0.992, n = 50), engagement indicator mean (Welch F 1.60, p = 0.211, n = 99), or intent to leave (Welch F 1.15, p = 0.346, n = 99). Respondent-level differences closely followed role and do not isolate the contribution of certification.

5.2 Artificial Intelligence Certifications

Indicator C016 · What AI certifications do you have? · Full statistical analysis ›

More than half of respondents answering the AI-certification question, 437 of 788 (55.46%), reported no AI certification. AI certifications were reported by 788 of 850 respondents. The remaining 351 respondents named one of five AI certifications, each reported by between 66 and 78 respondents.

AI-certification answers did not differ statistically significantly by role, vendor type, or the engagement-level segment families examined. Engagement-level comparisons with the four core study measures could not be estimated under the study’s eligibility rule. Fewer than two answer categories contained at least 10 engagements. The certification distribution remains descriptive and does not establish the contribution of certification to capability or engagement outcomes.

5.3 Sources of Agentic AI Knowledge

Indicator C017 · Where did you gain your AI knowledge? · Full statistical analysis ›

Selected sources of AI knowledge did not differ statistically significantly by role, vendor type, or the engagement-level segment families examined. A source of AI knowledge was reported by 800 of 850 respondents. Each of the six sources was selected by between 125 and 140 respondents.

The source of AI knowledge assigned to each engagement was not statistically significantly associated with cancellation (chi-square 2.39, p = 0.792, n = 94 engagements), customer satisfaction among active engagements (Welch F 0.43, p = 0.521, n = 22 engagements), engagement indicator mean (Welch F 0.24, p = 0.940, n = 94 engagements), or intent to leave (Welch F 0.36, p = 0.871, n = 94 engagements). The cancellation, engagement indicator mean, and intent-to-leave comparisons met the equivalence criterion for categorical answers. That criterion uses omega squared, an estimate of the variation associated with differences between groups, with a study-defined bound of 0.09. The customer-satisfaction comparison did not meet the equivalence criterion and remains inconclusive. The conclusions were retained under the reported robustness checks.

Thirty-seven written answers addressed sources of AI knowledge, all from individual contributors on 33 engagements. Nineteen were negative and 14 positive. They described brief formal instruction supplemented by colleagues, external material, and experience during delivery. The structured question recorded one selected source, whereas the written answers described learning that could involve several sources.

“i did the agent course after i was already three months on the desk, i learned the hard way first, then the course told me what i should have done. Order is wrong, content is fine” -Individual Contributor, BPO, Corporate and back-office functions