Source: https://akholi.com/outsourcing-2026-agentic-ai-tier-1-banks/study-data/part-5-c211/
Author: Phil Hatch
Publisher: Akholi

I have altered AI outputs

Study Data · Part 5 · Indicator Deep-Dive Analysis

# I have altered AI outputs

Indicator C211 · By Phil Hatch · Akholi · October 9, 2026

← Study Data← Previous indicatorNext indicator →

## On this page

25 sectionsIndicator SpecificationSummary of FindingsDistribution of ResponsesCoverage and Origin of Written ResponsesDistribution of Written SentimentDescriptive Statistics by SegmentTests of Difference Between SegmentsWritten Sentiment by SegmentAssociation with Cancellation RateAssociation with Customer SatisfactionAssociation with the Engagement Indicator MeanAssociation with Intent to LeaveAssociation of Written Sentiment with the Study MeasuresComparison of Associations Across the Four MeasuresClustering and Effective Sample SizeEquivalence Tests of Null ResultsRobustness ChecksMeasure Independence and Part-Whole CorrectionComparison with the Study NormComparison of Written Sentiment with Other Part 5 IndicatorsCo-Occurring Indicators and ThemesInterpretation of the DataLimitationsDefinitionsSources and Provenance

## Indicator Specification

Table 1. Indicator specificationTable

| Field | Value |
| --- | --- |
| Indicator code | C211 |
| Question as asked | I have altered AI outputs |
| Failure category | Not one of the 100 rated indicators |
| Place in the survey | Challenge Deep Dive > 5.6 Alterations > AI Output |
| Indicator class | Respondent level |
| Indicator type | Yes or no |
| Response scale | Levels: Yes, No |
| Answering base | 828 of 850 respondents gave one of the listed answers. 846 answered, counting I do not know and I prefer to not answer, across 150 of 150 engagements |
| Non-answers | I do not know 2, I prefer to not answer 16, blank 4 |
| Written responses addressing the indicator | 407 responses from 276 respondents on 116 engagements (361 primary, 46 secondary) |

## Summary of Findings

Table 2. Key results: association of the indicator with the four study measuresTable

| Measure | Effect size | n | Value | p |
| --- | --- | --- | --- | --- |
| Cancellation rate | Omega squared, 0 or 1 flag | 142 engagements | 0.253 | < 0.001 |
| Customer satisfaction | Omega squared | 88 engagements | 0.072 | 0.005 |
| Engagement indicator mean | Omega squared | 142 engagements | 0.297 | < 0.001 |
| Intent to leave | Omega squared | 142 engagements | 0.259 | < 0.001 |

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.

After Holm correction, the share answering True differed by engagement status and lifecycle stage (chi-square 37.84, Holm p < 0.001, Cramer's V 0.52): the share answering True was highest among respondents on canceled engagements (54.19%, n = 310 respondents) and lowest among respondents on active engagements in Transition/Cut-Over (18.52%, n = 243 respondents). It also differed by respondent role (chi-square 107.09, Holm p < 0.001, Cramer's V 0.36): the share answering True was highest among individual contributors (42.46%, n = 650 respondents) and lowest among portfolio directors (0.00%, n = 31 respondents). Outsourcing model, institution type, process family, class of work and vendor type showed no difference.

The 407 written responses came from 276 respondents on 116 engagements, and 399 were negative. The respondents who answered C214, on why they altered the outputs, described an agent that gave a different answer to the same input from one run to the next, so they chose the version that looked right and corrected it before it moved on. Those who commented under C216 described the alterations as routine, known to the team lead, unrecorded and largely out of the financial institution's sight.

## Distribution of Responses

Table 3. Distribution of responses, respondent levelTable

| Answer | Respondents | Share of answered (base 828) | Cumulative share | Engagements with at least one | Share of 150 engagements | Distressed |
| --- | --- | --- | --- | --- | --- | --- |
| Yes | 278 | 33.57% | 33.57% | 108 | 72.00% | Not applicable |
| No | 550 | 66.43% | 100.00% | 147 | 98.00% | Not applicable |
| I do not know | 2 | Not in base | Not in base | 2 | 1.33% | Not applicable |
| I prefer to not answer | 16 | Not in base | Not in base | 16 | 10.67% | Not applicable |
| Answered, total | 828 | 100.00% | | 150 | 100.00% | |

Of the 828 respondents who answered, 278 (33.57%) answered True and 550 (66.43%) answered False. The non-answers were I do not know (2) and I prefer to not answer (16). Each engagement was assigned the answer given by the largest number of its respondents. True was the assigned answer on 42 engagements and False on 100, and 8 engagements were tied and carry no assigned answer.

## Coverage and Origin of Written Responses

Figure 1. Written responses addressing the indicator, by question answered and written score (n = 407 responses)ExhibitNote: Each bar is one survey question. C214 and C216 and C208 and C210 and C202 and C204 are the written-answer questions of the indicator's survey section, so their responses are primary. The responses to every other question are secondary. Segments wide enough to hold a label carry their count. C214 asked: “Why did you alter AI outputs?” C216 asked: “Comment” C208 asked: “Why did you alter client processes?” C210 asked: “Comment” C202 asked: “Why did you alter client provided data?” C204 asked: “Comment” C177 asked: “Add any comments on the agentic AI tools that the ratings in this section do not capture.” C156 asked: “Add any comments on the joint management of this engagement that the ratings in this section do not capture.” C135 asked: “Add any comments on your employer that the ratings in this section do not capture.” C114 asked: “Add any comments on the customer organization that the ratings in this section do not capture.”Table 4. Written responses by originTable

| Origin | Responses | Respondents | Engagements | Negative | Strongest negative (−10) | Positive | Mean written score |
| --- | --- | --- | --- | --- | --- | --- | --- |
| Primary | 361 | 250 | 109 | 356 of 361 (98.61%) | 98 of 361 (27.15%) | 5 of 361 (1.39%) | −6.2 |
| Secondary | 46 | 44 | 35 | 43 of 46 (93.48%) | 13 of 46 (28.26%) | 3 of 46 (6.52%) | −5.8 |
| All responses | 407 | 276 | 116 | 399 of 407 (98.03%) | 111 of 407 (27.27%) | 8 of 407 (1.97%) | −6.2 |

The primary responses numbered 361, written in answer to C214, C216, C208, C210, C202 and C204, the written-answer questions of the indicator's survey section. The remaining 46 responses, 11.30% of the total, were secondary and came from 4 other questions. That secondary share ranked 1st from the highest among the 13 Part 5 indicators with 30 or more written responses, against a median of 2.00%. Respondents who wrote both a primary and a secondary response numbered 18. Negative responses made up 98.61% of primary responses (356 of 361) and 93.48% of secondary responses (43 of 46).

The 181 responses to C214 gave the reasons for altering the outputs. In 74 responses, respondents said the same input produced a different answer on another run, or that reruns did not match. In 57, they described choosing among the runs: the version that looked right, matched the prior run or matched what the team expected. Respondents said they could not sign off, or put their name to, an output that would not reproduce in 11 responses. Change approval slower than the work, a ticket that took longer than the fix, or a deadline that passed while the team waited appeared in 31 responses across all questions. Eleven responses said leadership had told the team to make the output work and not raise the problem, and 11 responses to C214 placed part of the problem with a vendor, feed, tool version or third-party platform outside the team's control.

The 161 responses to C216 described how the alterations were handled. Respondents used words of frequency, such as every day, most weeks or almost every ticket, in 95 responses, and 8 gave a share out of ten. The team lead or manager appeared in 41 responses, in 38 as someone who knew or was told. Across all questions, respondents said the financial institution was not told, or that they did not know whether it was, in 15 responses, and described alterations that were unlogged, unreported or made quietly in 41. In 9 responses to C216, respondents described the quality assurance score as low, and in 15 they said they caught the errors themselves before the output left the team. Tiredness appeared in 14 responses across all questions, and 5 noted that the engagement had been canceled regardless.

Responses to C177, C156 and C135 that also addressed the indicator described an agent that lost its instructions over a long session or failed on a long document, logs that did not show what the agent had done, and no one with authority to switch it off. Three responses described a side list, private list or own tracker of corrections. Four positive responses said the respondent altered outputs only rarely.

## Distribution of Written Sentiment

Figure 2. Distribution of written sentiment (n = 407 responses)ExhibitNote: Column height is the share of the 407 responses. Labels give the count and the share.Table 5. Summary statistics of written sentimentTable

| Statistic | Value | Basis |
| --- | --- | --- |
| Mean written score | −6.2 | 95% CI −6.4 to −5.9, cluster bootstrap by engagement, 10,000 draws, n = 407 responses |
| Median written score | −5.0 | n = 407 responses |
| Negative responses | 399 of 407 (98.03%) | 95% CI 96.61% to 99.26% of 407 responses, cluster bootstrap by engagement |
| Strongest negative responses (−10) | 111 of 407 (27.27%) | Statements of totality, constancy or serious harm |
| Positive responses | 8 of 407 (1.97%) | +5 or +10 |
| Net balance | −96.07 percentage points | Share positive minus share negative, base 407 responses |
| Respondent-level mean | −6.0 | Mean of each respondent's own mean written score, n = 276 respondents |

Of the 407 responses, 399 (98.03%, 95% CI 96.61% to 99.26%) were negative and 111 (27.27%) carried the strongest negative score of −10. Positive responses numbered 8 (1.97%), and no response was neutral. The mean written score was −6.2 (95% CI −6.4 to −5.9, n = 407 responses), and the median was −5.0.

## Descriptive Statistics by Segment

Table 6. Descriptive statistics by segmentTable

| Segment | Base | Answered | Answer rate | Engagements | Yes, share | No, share |
| --- | --- | --- | --- | --- | --- | --- |
| Study set | 850 | 828 | 97.41% of 850 | 150 | 33.57% | 66.43% |
| Outsourcing model | | | | | | |
| CXM | 232 | 224 | 96.55% of 232 | 42 | 31.70% | 68.30% |
| ITO | 244 | 240 | 98.36% of 244 | 40 | 33.33% | 66.67% |
| BPO | 209 | 203 | 97.13% of 209 | 37 | 35.47% | 64.53% |
| KPO | 165 | 161 | 97.58% of 165 | 31 | 34.16% | 65.84% |
| Institution type | | | | | | |
| Asset management | 132 | 130 | 98.48% of 132 | 24 | 34.62% | 65.38% |
| Wealth management and private banking | 140 | 139 | 99.29% of 140 | 23 | 30.22% | 69.78% |
| Investment banking (advisory and capital-markets origination) | 105 | 100 | 95.24% of 105 | 19 | 44.00% | 56.00% |
| Insurance (life, property and casualty, health) | 78 | 76 | 97.44% of 78 | 14 | 42.11% | 57.89% |
| Commercial and business banking | 82 | 78 | 95.12% of 82 | 13 | 23.08% | 76.92% |
| Payments and merchant acquiring (networks, acquiring, processing) | 71 | 69 | 97.18% of 71 | 13 | 36.23% | 63.77% |
| Credit card and payments issuing | 73 | 70 | 95.89% of 73 | 12 | 35.71% | 64.29% |
| Retail and consumer banking | 73 | 72 | 98.63% of 73 | 12 | 30.56% | 69.44% |
| Markets and trading (sales and trading, global markets) | 57 | 55 | 96.49% of 57 | 11 | 27.27% | 72.73% |
| Private equity and private capital | 39 | 39 | 100.00% of 39 | 9 | 25.64% | 74.36% |
| Process family | | | | | | |
| Corporate and back-office functions | 248 | 240 | 96.77% of 248 | 47 | 35.00% | 65.00% |
| Technology engineering and operations | 204 | 201 | 98.53% of 204 | 35 | 26.37% | 73.63% |
| Client servicing and communications | 94 | 91 | 96.81% of 94 | 16 | 41.76% | 58.24% |
| Trade, settlement and reconciliation | 80 | 77 | 96.25% of 80 | 15 | 41.56% | 58.44% |
| Payments, cards and disputes | 73 | 72 | 98.63% of 73 | 12 | 34.72% | 65.28% |
| Credit, underwriting and suitability | 57 | 55 | 96.49% of 57 | 9 | 20.00% | 80.00% |
| Regulatory, audit, tax and legal | 51 | 50 | 98.04% of 51 | 9 | 36.00% | 64.00% |
| Financial crime and client onboarding | 43 | 42 | 97.67% of 43 | 7 | 40.48% | 59.52% |
| Class of work | | | | | | |
| LOB Operational | 451 | 442 | 98.00% of 451 | 76 | 36.88% | 63.12% |
| Corporate Functional | 399 | 386 | 96.74% of 399 | 74 | 29.79% | 70.21% |
| Engagement status and lifecycle stage | | | | | | |
| Active, Build-Out | 175 | 172 | 98.29% of 175 | 31 | 20.93% | 79.07% |
| Active, Transition/Cut-Over | 250 | 243 | 97.20% of 250 | 43 | 18.52% | 81.48% |
| Active, Production | 103 | 103 | 100.00% of 103 | 20 | 28.16% | 71.84% |
| Canceled | 322 | 310 | 96.27% of 322 | 56 | 54.19% | 45.81% |
| Respondent role | | | | | | |
| Individual Contributor | 667 | 650 | 97.45% of 667 | 150 | 42.46% | 57.54% |
| Engagement Manager | 150 | 147 | 98.00% of 150 | 149 | 1.36% | 98.64% |
| Portfolio Director | 31 | 31 | 100.00% of 31 | 31 | 0.00% | 100.00% |
| Customer Management Executive | 2 | 0 | 0 of 2 | 0 | 0 | 0 |
| Vendor type | | | | | | |
| IT services major | 316 | 309 | 97.78% of 316 | 59 | 33.98% | 66.02% |
| BPO specialist | 198 | 189 | 95.45% of 198 | 37 | 35.98% | 64.02% |
| Consulting firm | 117 | 114 | 97.44% of 117 | 23 | 28.07% | 71.93% |
| Boutique or specialist provider | 87 | 86 | 98.85% of 87 | 17 | 24.42% | 75.58% |
| Pure-play agentic AI firm | 75 | 73 | 97.33% of 75 | 14 | 36.99% | 63.01% |
| Not reported, recovered by engagement | 57 | 57 | 100.00% of 57 | 47 | 43.86% | 56.14% |

Each share is computed on the respondents who answered in that row. Rows with fewer than 10 respondents report counts only.

## Tests of Difference Between Segments

Table 7. Tests of difference between segmentsTable

| Segment family | Unit | Chi-square | df | p | Holm p | Cramer's V | Test for p | Levels excluded |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| Outsourcing model | Engagement | 0.42 | 3 | 0.935 | 1.000 | 0.05 | chi-square | None |
| Institution type | Engagement | 2.44 | 7 | 0.938 | 1.000 | 0.14 | Monte Carlo, simulated | Markets and trading (sales and trading, global markets), n = 9, Private equity and private capital, n = 9 |
| Process family | Engagement | 7.49 | 4 | 0.111 | 0.443 | 0.25 | Monte Carlo, simulated | Credit, underwriting and suitability, n = 9, Regulatory, audit, tax and legal, n = 7, Financial crime and client onboarding, n = 7 |
| Class of work | Engagement | 3.54 | 1 | 0.060 | 0.299 | 0.16 | chi-square | None |
| Engagement status and lifecycle stage | Engagement | 37.84 | 3 | < 0.001 | < 0.001 | 0.52 | chi-square | None |
| Respondent role | Respondent | 107.09 | 2 | < 0.001 | < 0.001 | 0.36 | chi-square | Customer Management Executive, n = 0 |
| Vendor type | Respondent | 8.38 | 5 | 0.137 | 0.443 | 0.10 | chi-square | None |

Segment families recorded at engagement level were tested on engagements, and respondent role and vendor type on respondents. After Holm correction, the share answering True differed by engagement status and lifecycle stage (chi-square 37.84, Holm p < 0.001, Cramer's V 0.52): the share answering True was highest among respondents on canceled engagements (54.19%, n = 310 respondents) and lowest among respondents on active engagements in Transition/Cut-Over (18.52%, n = 243 respondents). It also differed by respondent role (chi-square 107.09, Holm p < 0.001, Cramer's V 0.36): the share answering True was highest among individual contributors (42.46%, n = 650 respondents) and lowest among portfolio directors (0.00%, n = 31 respondents). Outsourcing model, institution type, process family, class of work and vendor type showed no difference.

## Written Sentiment by Segment

Figure 3. Written sentiment by engagement status and lifecycle stage, share of each group's responsesExhibitNote: Each bar is the whole of its group's responses, divided by written score from −10 at the left to +10 at the right. Segments large enough to hold a label carry their share.Table 8. Written sentiment by segmentTable

| Segment | Responses | Respondents | Engagements | Mean written score | Negative | Strongest negative (−10) |
| --- | --- | --- | --- | --- | --- | --- |
| Engagement status and lifecycle stage | | | | | | |
| Active, Build-Out | 49 | 35 | 24 | −5.2 | 47 of 49 (95.92%) | 6 of 49 (12.24%) |
| Active, Transition/Cut-Over | 76 | 51 | 26 | −5.8 | 73 of 76 (96.05%) | 18 of 76 (23.68%) |
| Active, Production | 40 | 27 | 12 | −6.0 | 40 of 40 (100.00%) | 8 of 40 (20.00%) |
| Canceled | 242 | 163 | 54 | −6.5 | 239 of 242 (98.76%) | 79 of 242 (32.64%) |
| Respondent role | | | | | | |
| Individual Contributor | 394 | 266 | 115 | −6.2 | 387 of 394 (98.22%) | 109 of 394 (27.66%) |
| Engagement Manager | 12 | 9 | 9 | −5.0 | 11 of 12 (91.67%) | 2 of 12 (16.67%) |
| Portfolio Director | 1 | 1 | 1 | ‡ | 1 of 1 | 0 of 1 |
| Customer Management Executive | 0 | 0 | 0 | No responses | | |

Note: ‡ Base under 10. No statistic is reported.

Written sentiment differed by engagement status and lifecycle stage across the 116 engagements with a written response (Kruskal-Wallis H 9.64, df 3, p = 0.022). Of the groups with 10 or more responses, responses from canceled engagements were the most negative (mean −6.5, 242 responses) and responses from active engagements in Build-Out the least negative (mean −5.2, 49 responses).

No test across respondent roles was run, because fewer than two roles had 30 or more written responses. Individual contributors wrote 394 responses, engagement managers 12 and portfolio directors 1.

## Association with Cancellation Rate

Cancellation rate is the share of engagements canceled by the study date, measured at engagement level (150 engagements, 56 canceled). Canceled engagements were rated after cancellation. The association is therefore a descriptive contrast and not a forecast of cancellation. Each test compares the engagements by their assigned answer, within the study set and within each segment.

Table 9. Association with cancellation rate by answer levelTable

| Answer level | Engagements | Canceled | Cancellation rate |
| --- | --- | --- | --- |
| Yes | 42 | 32 | 76.19% |
| No | 100 | 22 | 22.00% |
| I do not know | 0 | No engagements | |
| I prefer to not answer | 0 | No engagements | |

Table 10. Association with cancellation rate, tests across answer levels by segmentTable

| Segment | Engagements | Chi-square | df | p | Holm p | Cramer's V | Test for p |
| --- | --- | --- | --- | --- | --- | --- | --- |
| Study set | 142 | 36.86 | 1 | < 0.001 | < 0.001 | 0.51 | chi-square |
| Outsourcing model | | | | | | | |
| CXM | 42 | 7.01 | 1 | 0.013 | 0.013 | 0.41 | Fisher exact |
| ITO | 34 | a | | | | | |
| BPO | 36 | 14.96 | 1 | < 0.001 | < 0.001 | 0.64 | Fisher exact |
| KPO | 30 | a | | | | | |
| Institution type | | | | | | | |
| Asset management | 23 | a | | | | | |
| Wealth management and private banking | 23 | a | | | | | |
| Investment banking (advisory and capital-markets origination) | 18 | a | | | | | |
| Insurance (life, property and casualty, health) | 12 | a | | | | | |
| Commercial and business banking | 11 | a | | | | | |
| Payments and merchant acquiring (networks, acquiring, processing) | 13 | a | | | | | |
| Credit card and payments issuing | 12 | a | | | | | |
| Retail and consumer banking | 12 | a | | | | | |
| Markets and trading (sales and trading, global markets) | 9 | ‡ | | | | | |
| Private equity and private capital | 9 | ‡ | | | | | |
| Process family | | | | | | | |
| Corporate and back-office functions | 47 | 14.33 | 1 | < 0.001 | < 0.001 | 0.55 | chi-square |
| Technology engineering and operations | 30 | a | | | | | |
| Client servicing and communications | 16 | a | | | | | |
| Trade, settlement and reconciliation | 15 | a | | | | | |
| Payments, cards and disputes | 11 | a | | | | | |
| Credit, underwriting and suitability | 9 | ‡ | | | | | |
| Regulatory, audit, tax and legal | 7 | ‡ | | | | | |
| Financial crime and client onboarding | 7 | ‡ | | | | | |
| Class of work | | | | | | | |
| LOB Operational | 74 | 19.83 | 1 | < 0.001 | < 0.001 | 0.52 | chi-square |
| Corporate Functional | 68 | 16.84 | 1 | < 0.001 | < 0.001 | 0.50 | chi-square |
| Engagement status and lifecycle stage | | | | | | | |
| Active, Build-Out | | b | | | | | |
| Active, Transition/Cut-Over | | b | | | | | |
| Active, Production | | b | | | | | |
| Canceled | | b | | | | | |

Note: ‡ Base under 10. No statistic is reported. a Not tested: fewer than two levels of 10 or more engagements. b Not tested: no variation in cancellation.

Of the engagements whose respondents most often answered True, 32 of 42 (76.19%) were canceled, against 22 of 100 (22.00%) of those whose respondents most often answered False (chi-square 36.86, p < 0.001, Cramer's V 0.51, n = 142 engagements). The difference reached significance and held after Holm correction in 5 of the 5 segment levels tested.

## Association with Customer Satisfaction

Customer satisfaction is the percentage recorded against each engagement. The primary estimate uses the 94 active engagements.

Table 11. Association with customer satisfaction by answer levelTable

| Answer level | Engagements | Mean of the measure | SD |
| --- | --- | --- | --- |
| Yes | 10 | 54.7 | 9.0 |
| No | 78 | 65.3 | 11.5 |
| I do not know | 0 | No engagements | |
| I prefer to not answer | 0 | No engagements | |

Table 12. Association with customer satisfaction, tests across answer levels by segmentTable

| Segment | Engagements | Welch F | df | p | Holm p | Omega squared | 90% CI |
| --- | --- | --- | --- | --- | --- | --- | --- |
| Study set | 88 | 11.44 | 1, 13.1 | 0.005 | 0.005 | 0.072 | 0.014 to 0.185 |
| Outsourcing model | | | | | | | |
| CXM | 27 | a | | | | | |
| ITO | 21 | a | | | | | |
| BPO | 22 | a | | | | | |
| KPO | 18 | a | | | | | |
| Institution type | | | | | | | |
| Asset management | 11 | a | | | | | |
| Wealth management and private banking | 15 | a | | | | | |
| Investment banking (advisory and capital-markets origination) | 9 | ‡ | | | | | |
| Insurance (life, property and casualty, health) | 8 | ‡ | | | | | |
| Commercial and business banking | 7 | ‡ | | | | | |
| Payments and merchant acquiring (networks, acquiring, processing) | 10 | a | | | | | |
| Credit card and payments issuing | 7 | ‡ | | | | | |
| Retail and consumer banking | 8 | ‡ | | | | | |
| Markets and trading (sales and trading, global markets) | 6 | ‡ | | | | | |
| Private equity and private capital | 7 | ‡ | | | | | |
| Process family | | | | | | | |
| Corporate and back-office functions | 28 | a | | | | | |
| Technology engineering and operations | 22 | a | | | | | |
| Client servicing and communications | 7 | ‡ | | | | | |
| Trade, settlement and reconciliation | 7 | ‡ | | | | | |
| Payments, cards and disputes | 9 | ‡ | | | | | |
| Credit, underwriting and suitability | 8 | ‡ | | | | | |
| Regulatory, audit, tax and legal | 3 | ‡ | | | | | |
| Financial crime and client onboarding | 4 | ‡ | | | | | |
| Class of work | | | | | | | |
| LOB Operational | 44 | a | | | | | |
| Corporate Functional | 44 | a | | | | | |
| Engagement status and lifecycle stage | | | | | | | |
| Active, Build-Out | 29 | a | | | | | |
| Active, Transition/Cut-Over | 39 | a | | | | | |
| Active, Production | 20 | a | | | | | |
| Canceled | | b | | | | | |

Note: ‡ Base under 10. No statistic is reported. a Not tested: fewer than two levels of 10 or more engagements. b Not tested: canceled engagements fall outside the active population.

Among active engagements, customer satisfaction had a mean of 54.7% (n = 10 engagements) where respondents most often answered True and 65.3% (n = 78 engagements) where they most often answered False (Welch F 11.44, p = 0.005, omega squared 0.072, n = 88 engagements). The difference reached significance and held after Holm correction in 0 of the 0 segment levels tested.

## Association with the Engagement Indicator Mean

The engagement indicator mean is the mean of an engagement's ratings across the rated indicators, measured at engagement level. The indicator is not a rated indicator and does not enter the mean.

Table 13. Association with the engagement indicator mean by answer levelTable

| Answer level | Engagements | Mean of the measure | SD |
| --- | --- | --- | --- |
| Yes | 42 | 2.7 | 0.9 |
| No | 100 | 4.5 | 1.4 |
| I do not know | 0 | No engagements | |
| I prefer to not answer | 0 | No engagements | |

Table 14. Association with the engagement indicator mean, tests across answer levels by segmentTable

| Segment | Engagements | Welch F | df | p | Holm p | Omega squared | 90% CI |
| --- | --- | --- | --- | --- | --- | --- | --- |
| Study set | 142 | 87.54 | 1, 119.0 | < 0.001 | < 0.001 | 0.297 | 0.201 to 0.394 |
| Outsourcing model | | | | | | | |
| CXM | 42 | 15.11 | 1, 27.2 | < 0.001 | 0.001 | 0.204 | 0.061 to 0.389 |
| ITO | 34 | a | | | | | |
| BPO | 36 | 38.28 | 1, 33.4 | < 0.001 | < 0.001 | 0.416 | 0.221 to 0.581 |
| KPO | 30 | a | | | | | |
| Institution type | | | | | | | |
| Asset management | 23 | a | | | | | |
| Wealth management and private banking | 23 | a | | | | | |
| Investment banking (advisory and capital-markets origination) | 18 | a | | | | | |
| Insurance (life, property and casualty, health) | 12 | a | | | | | |
| Commercial and business banking | 11 | a | | | | | |
| Payments and merchant acquiring (networks, acquiring, processing) | 13 | a | | | | | |
| Credit card and payments issuing | 12 | a | | | | | |
| Retail and consumer banking | 12 | a | | | | | |
| Markets and trading (sales and trading, global markets) | 9 | ‡ | | | | | |
| Private equity and private capital | 9 | ‡ | | | | | |
| Process family | | | | | | | |
| Corporate and back-office functions | 47 | 26.46 | 1, 42.2 | < 0.001 | < 0.001 | 0.272 | 0.116 to 0.441 |
| Technology engineering and operations | 30 | a | | | | | |
| Client servicing and communications | 16 | a | | | | | |
| Trade, settlement and reconciliation | 15 | a | | | | | |
| Payments, cards and disputes | 11 | a | | | | | |
| Credit, underwriting and suitability | 9 | ‡ | | | | | |
| Regulatory, audit, tax and legal | 7 | ‡ | | | | | |
| Financial crime and client onboarding | 7 | ‡ | | | | | |
| Class of work | | | | | | | |
| LOB Operational | 74 | 54.98 | 1, 71.8 | < 0.001 | < 0.001 | 0.359 | 0.224 to 0.486 |
| Corporate Functional | 68 | 31.06 | 1, 32.3 | < 0.001 | < 0.001 | 0.226 | 0.102 to 0.370 |
| Engagement status and lifecycle stage | | | | | | | |
| Active, Build-Out | 29 | a | | | | | |
| Active, Transition/Cut-Over | 39 | a | | | | | |
| Active, Production | 20 | a | | | | | |
| Canceled | 54 | 4.30 | 1, 49.1 | 0.043 | 0.043 | 0.054 | 0.000 to 0.201 |

Note: ‡ Base under 10. No statistic is reported. a Not tested: fewer than two levels of 10 or more engagements.

The engagement indicator mean was 2.7 (n = 42 engagements) on engagements whose respondents most often answered True and 4.5 (n = 100 engagements) on those whose respondents most often answered False (Welch F 87.54, p < 0.001, omega squared 0.297, n = 142 engagements). The difference reached significance and held after Holm correction in 6 of the 6 segment levels tested.

## Association with Intent to Leave

Intent to leave is the answer to C019, “How likely are you to leave your current job in the next 12 months?”, rated 1 to 10 by up to 798 respondents. For the indicator, the comparison uses the mean intent to leave of each engagement.

Table 15. Association with intent to leave by answer levelTable

| Answer level | Engagements | Mean of the measure | SD |
| --- | --- | --- | --- |
| Yes | 42 | 8.7 | 1.3 |
| No | 100 | 6.0 | 2.4 |
| I do not know | 0 | No engagements | |
| I prefer to not answer | 0 | No engagements | |

Table 16. Association with intent to leave, tests across answer levels by segmentTable

| Segment | Engagements | Welch F | df | p | Holm p | Omega squared | 90% CI |
| --- | --- | --- | --- | --- | --- | --- | --- |
| Study set | 142 | 79.99 | 1, 131.4 | < 0.001 | < 0.001 | 0.259 | 0.166 to 0.357 |
| Outsourcing model | | | | | | | |
| CXM | 42 | 13.55 | 1, 26.0 | 0.001 | 0.002 | 0.191 | 0.053 to 0.377 |
| ITO | 34 | a | | | | | |
| BPO | 36 | 24.68 | 1, 34.0 | < 0.001 | < 0.001 | 0.298 | 0.117 to 0.486 |
| KPO | 30 | a | | | | | |
| Institution type | | | | | | | |
| Asset management | 23 | a | | | | | |
| Wealth management and private banking | 23 | a | | | | | |
| Investment banking (advisory and capital-markets origination) | 18 | a | | | | | |
| Insurance (life, property and casualty, health) | 12 | a | | | | | |
| Commercial and business banking | 11 | a | | | | | |
| Payments and merchant acquiring (networks, acquiring, processing) | 13 | a | | | | | |
| Credit card and payments issuing | 12 | a | | | | | |
| Retail and consumer banking | 12 | a | | | | | |
| Markets and trading (sales and trading, global markets) | 9 | ‡ | | | | | |
| Private equity and private capital | 9 | ‡ | | | | | |
| Process family | | | | | | | |
| Corporate and back-office functions | 47 | 24.77 | 1, 44.6 | < 0.001 | < 0.001 | 0.241 | 0.092 to 0.413 |
| Technology engineering and operations | 30 | a | | | | | |
| Client servicing and communications | 16 | a | | | | | |
| Trade, settlement and reconciliation | 15 | a | | | | | |
| Payments, cards and disputes | 11 | a | | | | | |
| Credit, underwriting and suitability | 9 | ‡ | | | | | |
| Regulatory, audit, tax and legal | 7 | ‡ | | | | | |
| Financial crime and client onboarding | 7 | ‡ | | | | | |
| Class of work | | | | | | | |
| LOB Operational | 74 | 47.88 | 1, 71.4 | < 0.001 | < 0.001 | 0.314 | 0.181 to 0.446 |
| Corporate Functional | 68 | 28.90 | 1, 39.0 | < 0.001 | < 0.001 | 0.186 | 0.072 to 0.330 |
| Engagement status and lifecycle stage | | | | | | | |
| Active, Build-Out | 29 | a | | | | | |
| Active, Transition/Cut-Over | 39 | a | | | | | |
| Active, Production | 20 | a | | | | | |
| Canceled | 54 | 0.34 | 1, 51.9 | 0.564 | 0.564 | −0.013 | 0.000 to 0.079 |

Note: ‡ Base under 10. No statistic is reported. a Not tested: fewer than two levels of 10 or more engagements.

The engagement mean of intent to leave was 8.7 (n = 42 engagements) on engagements whose respondents most often answered True and 6.0 (n = 100 engagements) on those whose respondents most often answered False (Welch F 79.99, p < 0.001, omega squared 0.259, n = 142 engagements). The difference reached significance and held after Holm correction in 5 of the 6 segment levels tested.

## Association of Written Sentiment with the Study Measures

Table 17. Written sentiment and the study measuresTable

| Comparison | With a negative response | Without a negative response | Test | Statistic | n | p |
| --- | --- | --- | --- | --- | --- | --- |
| Cancellation rate, engagements | 54 of 115 (46.96%) | 2 of 35 (5.71%) | Fisher exact | | 150 | < 0.001 |
| Customer satisfaction, active engagements | 60.84%, n = 61 | 70.12%, n = 33 | Welch t | −3.95 | 94 | < 0.001 |
| Engagement mean written score and cancellation, engagements | | | Point-biserial r | −0.26 | 116 | 0.005 |
| Respondent mean written score and intent to leave, respondents | | | Spearman rho | −0.27 | 254 | < 0.001 |

Two comparisons set the engagements with at least one negative response against all other engagements. Of the 35 other engagements, 34 had no written response addressing the indicator. These two comparisons therefore do not isolate written sentiment. Of the 115 engagements with a negative response, 54 (46.96%) were canceled, against 2 of the 35 other engagements (5.71%). The difference reached significance (Fisher exact p < 0.001). Customer satisfaction among active engagements differed between the two groups (mean 60.84% on 61 engagements against 70.12% on 33, Welch t −3.95, p < 0.001).

Two correlations test written sentiment directly. Across the 116 engagements with a written response, a more negative engagement mean written score came with cancellation (point-biserial r −0.26, p = 0.005). Across 254 respondents, a more negative respondent mean written score came with higher intent to leave (Spearman rho −0.27, p < 0.001). Of the four comparisons, all reached significance at 0.05.

## Comparison of Associations Across the Four Measures

Table 18. Associations across the four measuresTable

| Measure | Effect size | n | Value | p | Equivalent to zero | Rank among the four |
| --- | --- | --- | --- | --- | --- | --- |
| Cancellation rate | Omega squared, 0 or 1 flag | 142 | 0.253 | < 0.001 | No | 3 |
| Customer satisfaction | Omega squared | 88 | 0.072 | 0.005 | No | 4 |
| Engagement indicator mean | Omega squared | 142 | 0.297 | < 0.001 | No | 1 |
| Intent to leave | Omega squared | 142 | 0.259 | < 0.001 | No | 2 |

Steiger's z compares two correlations. A categorical indicator yields an effect size across its levels and not a correlation, so the pairs of measures were not compared by test. On omega squared, the association was strongest with the engagement indicator mean (0.297) and weakest with customer satisfaction (0.072).

## Clustering and Effective Sample Size

Table 19. Clustering and effective sample sizeTable

| Statistic | Value | Reading |
| --- | --- | --- |
| Clusters (engagements) | 150 | Engagements with one or more answers. |
| Mean respondents per engagement | 5.6 | Answers per engagement, among engagements with one or more answers. |
| ICC(1) | 0.23 | Share of the variance that lies between engagements. |
| Design effect | 2.05 | One plus the product of the ICC and the mean cluster size less one. |
| Nominal respondent n | 846 | Every answer, with “I do not know” and “I prefer to not answer” counted as levels, as in the plurality rule. |
| Effective n | 412 | Nominal n divided by the design effect. |
| Cluster bootstrap applied | Yes | Applied to every respondent-level interval where the ICC is 0.05 or above. |

Answers clustered within engagements (ICC 0.23). Clustering reduced the 846 respondent answers to the equivalent of about 412 independent answers. Respondent-level intervals therefore used the cluster bootstrap. Respondent-level tests used standard errors that allow for clustering by engagement.

## Equivalence Tests of Null Results

Table 20. Equivalence tests of null resultsTable

| Measure | n | Omega squared | Equivalence bound | Equivalence p | Equivalent to zero | Achieved power | MDE at power 0.80 |
| --- | --- | --- | --- | --- | --- | --- | --- |
| Cancellation rate | 142 | 0.253 | Omega squared < 0.090 | 0.999 | No | 0.96 | 0.053 |
| Customer satisfaction | 88 | 0.072 | Omega squared < 0.090 | 0.437 | No | 0.83 | 0.084 |
| Engagement indicator mean | 142 | 0.297 | Omega squared < 0.090 | 1.000 | No | 0.96 | 0.053 |
| Intent to leave | 142 | 0.259 | Omega squared < 0.090 | 0.999 | No | 0.96 | 0.053 |

No association was equivalent to zero. Each of the four differed from zero. The equivalence bound is an omega squared of 0.090, the study's threshold for a negligible association expressed as variance explained.

## Robustness Checks

Table 21. Robustness checksTable

| Check | Measure | Primary p | Check p | Check statistic | Conclusion holds |
| --- | --- | --- | --- | --- | --- |
| Kruskal-Wallis in place of Welch F | Customer satisfaction | 0.005 | 0.007 | 7.38 | Yes |
| Kruskal-Wallis in place of Welch F | Engagement indicator mean | < 0.001 | < 0.001 | 45.98 | Yes |
| Kruskal-Wallis in place of Welch F | Intent to leave | < 0.001 | < 0.001 | 36.25 | Yes |
| Active engagements only | Engagement indicator mean | < 0.001 | 0.001 | 17.53 | Yes |
| Active engagements only | Intent to leave | < 0.001 | 0.003 | 14.30 | Yes |
| Respondent level, cluster-robust Wald F by engagement | Cancellation rate | < 0.001 | < 0.001 | 55.00 | Yes |
| Respondent level, cluster-robust Wald F by engagement | Customer satisfaction | 0.005 | < 0.001 | 68.20 | Yes |
| Respondent level, cluster-robust Wald F by engagement | Engagement indicator mean | < 0.001 | < 0.001 | 225.98 | Yes |
| Respondent level, cluster-robust Wald F by engagement | Intent to leave | < 0.001 | < 0.001 | 110.71 | Yes |

All nine checks held. A check holds when its test falls on the same side of 0.05 as the primary test. The checks replace Welch's F with a Kruskal-Wallis test, repeat the tests on active engagements alone, and repeat them at respondent level with a Wald F clustered by engagement.

## Measure Independence and Part-Whole Correction

Each of the four measures was computed from data the indicator does not contain. Each result is therefore a finding and not an arithmetic consequence.

## Comparison with the Study Norm

A categorical indicator has no signed correlation with the study measures and therefore no profile against the study norm. The comparison by answer appears in the four association sections.

## Comparison of Written Sentiment with Other Part 5 Indicators

Figure 4. Share of negative written responses, Part 5 indicators with 30 or more written responses (n = 13 indicators)ExhibitNote: Each column is one indicator, ordered from the highest share of negative responses. The indicator is the amber column. The dashed line marks the median share, 98.03%.

Among the 13 Part 5 indicators with 30 or more written responses, the indicator's share of negative responses, 98.03%, ranked 7th from the highest, against a median of 98.03%. Its mean written score, −6.2, ranked 10th from the most negative, against a median of −6.3.

## Co-Occurring Indicators and Themes

Figure 5. Other indicators and themes in the same written responses (n = 407 responses)ExhibitNote: Themes carry no score. Labels give the count and the share of the 407 responses. The figure shows each indicator and each theme among the five most frequent of its kind, with every tie at the cut. C157: “Rate the factual reliability of the AI tools' outputs on this engagement before any human review.” C159: “Rate the consistency of the tools' outputs when given the same input and instructions more than once.” C212: “I discussed the AI output alteration with internal management” C213: “I or my management discussed AI output alteration with the client” C175: “Rate the stability of the tools' behavior over the engagement when nothing changed on the engagement side.” Emotional states in the same responses, as counts: frustrated 21, resigned 17, tired 16, calm or fine 14, stressed 8, proud 1.

The indicator most often present in the same responses was C157, “Rate the factual reliability of the AI tools' outputs on this engagement before any human review.”, in 313 of the 407 responses (76.90%). C159, “Rate the consistency of the tools' outputs when given the same input and instructions more than once.” (175 responses), C212, “I discussed the AI output alteration with internal management” (54 responses), C213, “I or my management discussed AI output alteration with the client” (28 responses) and C175, “Rate the stability of the tools' behavior over the engagement when nothing changed on the engagement side.” (26 responses) followed.

The most frequent theme was “Human rework absorbs the failure”, in 323 responses (79.36%). The second was “Occurs every cycle or most days”, in 93 responses.

## Interpretation of the Data

Of the respondents who answered, 33.57% said they had altered AI outputs. Engagements whose respondents most often said so were canceled more often, recorded lower customer satisfaction and a lower engagement indicator mean, and their respondents reported higher intent to leave. Every association held under each of the robustness checks.

The written evidence describes the alteration of outputs as the delivery team's answer to an agent that would not give the same answer twice, approval cycles slower than the work, and an instruction to keep the output moving. The alterations were made routinely, known within the team and rarely reported to the financial institution. The associations do not establish that the alterations produced the outcomes. The written responses describe them as one response to the same instability and the same gaps in change control that the Part 5 ratings record.

## Limitations

The evidence is the reported experience of the vendor's delivery teams, collected in a single cross-section. The associations reported here are not causal. The answer is the respondent's own report of an alteration and not a record of it.

Each engagement carries the answer given by the largest number of its respondents, and tied engagements are excluded from the engagement-level tests.

Shares of written sentiment are shares of responses and not of the study set of respondents. The written score takes five fixed values, so its mean and median are coarse. Responses clustered within engagements (407 responses on 116 engagements). Intervals for written sentiment therefore used a bootstrap by engagement.

The written responses are counted by sets of words stated in each rule. A response that described a reason in other words is not counted, so each count is a floor.

## Definitions

Table 22. DefinitionsTable

| Term | Definition |
| --- | --- |
| Response scale | The answer is True or False to the statement shown in the indicator specification. The indicator is not a rating. No distressed range applies. |
| Answering base | Respondents who answered True or False. |
| Non-answers | “I do not know” and “I prefer to not answer” are counted separately. They are excluded from the answering base and from the shares. |
| Indicator class | Respondent level: a report of the respondent's own action. Each engagement also carries the answer given by the largest number of its respondents, for the engagement-level tests. |
| Failure category | The five failure categories of the Study Taxonomy apply only to the rated indicators. The indicator carries none. |
| Engagement indicator mean | The mean of an engagement's ratings across the rated indicators. The indicator does not enter it. |
| Written response | A written answer about the respondent's own engagement that addresses the indicator. Non-answers and answers about another engagement are excluded. |
| Primary and secondary | A primary response answered the written-answer question of the indicator's survey section. A secondary response answered a question in another section and addressed the indicator in its content. |
| Written sentiment score | Each written response carries one of five scores: −10, −5, neutral, +5 or +10. The sign records whether the response described the state of the indicator as harmful or helpful to the engagement. The magnitude records its strength. A score of −10 marks a statement of totality, constancy or serious harm. |
| Cluster bootstrap | An interval computed by resampling engagements rather than responses, so that the responses from one engagement stay together. |

## Sources and Provenance

Table 23. Sources and provenanceTable

| Source | Value |
| --- | --- |
| Study data | Study Data.xlsx, Data sheet, column C211, sha256 219eba0457274d04 |
| Statistics engine | compute_statistics.py, sha256 738319eef764602b |
| Statistics record | Computed Statistics.json, record C211, sha256 1ae50a3637fc9370 |
| Computed | October 13, 2026, 04:57 UTC |
| Template | Indicator Template v2, version 2.0 of 10 October 2026 |
| Figures placed by script | 1,130 |
| Written responses | Quotes Master.jsonl, sha256 27da38085e2b13fd |
| Written sentiment scores | Quote Coding Rulebook, version 0.6, with section 21 |
| Written-response computation | qa_stats.py, sha256 74a61f3a96169da9. wfacts.py, sha256 b85fae20adf812a1. build_indicator.py, sha256 a97ad6e74627e11e |
| Written-response records | C211.qa.json, sha256 19e2817d652fd91a. C211.wfacts.json, sha256 36a002de8fad95d0. Hand-written text handcraft/C211.json, sha256 6ff0aaa2f1a72601 |


## Links to related study evidence

- [Part 5](https://akholi.com/outsourcing-2026-agentic-ai-tier-1-banks/part-5-manual-alteration/)
- [Customer satisfaction](https://akholi.com/outsourcing-2026-agentic-ai-tier-1-banks/study-data/part-1-measure-customer-satisfaction/)
- [Engagement indicator mean](https://akholi.com/outsourcing-2026-agentic-ai-tier-1-banks/study-data/part-1-measure-engagement-indicator-mean/)
- [Intent to leave](https://akholi.com/outsourcing-2026-agentic-ai-tier-1-banks/study-data/part-4-c019/)
- [C019](https://akholi.com/outsourcing-2026-agentic-ai-tier-1-banks/study-data/part-4-c019/)
- [C157](https://akholi.com/outsourcing-2026-agentic-ai-tier-1-banks/study-data/part-2-c157/)
- [C159](https://akholi.com/outsourcing-2026-agentic-ai-tier-1-banks/study-data/part-2-c159/)
- [C212](https://akholi.com/outsourcing-2026-agentic-ai-tier-1-banks/study-data/part-5-c212/)
- [C213](https://akholi.com/outsourcing-2026-agentic-ai-tier-1-banks/study-data/part-5-c213/)
- [C175](https://akholi.com/outsourcing-2026-agentic-ai-tier-1-banks/study-data/part-2-c175/)