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.

Of the 253 respondents who answered, 150 (59.29%) answered True and 103 (40.71%) answered False. The non-answers were I prefer to not answer (17). Each engagement was assigned the answer given by the largest number of its respondents. True was the assigned answer on 57 engagements and False on 36, “I prefer to not answer” on 4, and 53 engagements were tied or unanswered and carry no assigned answer.

The primary responses numbered 93, written in answer to C216, C214 and C204, the written-answer questions of the indicator's survey section. The remaining 2 responses, 2.11% of the total, were secondary and came from 2 other questions. That secondary share ranked 6th 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 0. Negative responses made up 26.88% of primary responses (25 of 93) and 1 of the 2 secondary responses.

The 86 responses to C216 described how the alteration was handled within the vendor. In 62, the respondent said they told the team lead, manager or team, or that the lead or manager already knew. Respondents placed the alteration at a frequency, such as every cycle, most weeks or seven out of ten, in 50 responses, and said that telling the lead changed nothing at the financial institution in 3.

Across all questions, respondents said the financial institution was not told in 15 responses. No response said the financial institution had been told. In 21 of the 26 negative responses, the respondent said they had told no one, or no one above them, of the alteration. One respondent who had told the lead once said it felt held against them, and no longer mentioned it.

Of the 95 responses, 26 (27.37%, 95% CI 18.39% to 37.37%) were negative and 23 (24.21%) carried the strongest negative score of −10. Positive responses numbered 67 (70.53%), and 2 were neutral. The mean written score was 1.4 (95% CI −0.1 to 2.7, n = 95 responses), and the median was 5.0.

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

Segment families recorded at engagement level were tested on engagements, and respondent role and vendor type on respondents. The share answering True did not differ by any segment family after Holm correction.

Written sentiment did not differ by engagement status and lifecycle stage across the 61 engagements with a written response (Kruskal-Wallis H 5.74, df 3, p = 0.125). Of the groups with 10 or more responses, responses from canceled engagements were the most negative (mean 0.2, 50 responses) and responses from active engagements in Production the least negative (mean 4.5, 10 responses).

No test across respondent roles was run, because fewer than two roles had 30 or more written responses. Individual contributors wrote 93 responses, engagement managers 2.

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.

Of the engagements whose respondents most often answered True, 28 of 57 (49.12%) were canceled, against 23 of 36 (63.89%) of those whose respondents most often answered False (chi-square 1.94, p = 0.163, Cramer's V 0.14, n = 93 engagements). The difference did not reach significance.

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

Among active engagements, customer satisfaction had a mean of 63.3% (n = 29 engagements) where respondents most often answered True and 57.5% (n = 13 engagements) where they most often answered False (Welch F 3.67, p = 0.063, omega squared 0.033, n = 42 engagements). The difference did not reach significance.

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.

The engagement indicator mean was 3.6 (n = 57 engagements) on engagements whose respondents most often answered True and 3.2 (n = 36 engagements) on those whose respondents most often answered False (Welch F 3.13, p = 0.080, omega squared 0.018, n = 93 engagements). The difference did not reach significance.

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.

The engagement mean of intent to leave was 7.3 (n = 57 engagements) on engagements whose respondents most often answered True and 8.0 (n = 36 engagements) on those whose respondents most often answered False (Welch F 2.32, p = 0.131, omega squared 0.012, n = 93 engagements). The difference did not reach significance.

Two comparisons set the engagements with at least one negative response against all other engagements. Of the 127 other engagements, 89 had no written response addressing the indicator. These two comparisons therefore do not isolate written sentiment. Of the 23 engagements with a negative response, 15 (65.22%) were canceled, against 41 of the 127 other engagements (32.28%). The difference reached significance (Fisher exact p = 0.004). The comparison of customer satisfaction was not computed, because one of the two groups held too few active engagements.

Two correlations test written sentiment directly. Across the 61 engagements with a written response, the engagement mean written score was not associated with cancellation (point-biserial r −0.25, p = 0.050). Across 90 respondents, the respondent mean written score was not associated with intent to leave (Spearman rho −0.16, p = 0.121). Of the three comparisons, one reached significance at 0.05.

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 customer satisfaction (0.033) and weakest with cancellation (0.010).

Answers clustered within engagements (ICC 0.39). Clustering reduced the 270 respondent answers to the equivalent of about 170 independent answers. Respondent-level intervals therefore used the cluster bootstrap. Respondent-level tests used standard errors that allow for clustering by engagement.

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

Seven of the nine checks computed 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.

Two checks failed. The respondent-level clustered check for the engagement indicator mean gave p < 0.001, against p = 0.080 for the primary test. The respondent-level clustered check for intent to leave gave p = 0.013, against p = 0.131 for the primary test.

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.

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.

Among the 13 Part 5 indicators with 30 or more written responses, the indicator's share of negative responses, 27.37%, ranked 11th from the highest, against a median of 98.03%. Its mean written score, 1.4, ranked 11th from the most negative, against a median of −6.3.

The indicator most often present in the same responses was C211, “I have altered AI outputs”, in 54 of the 95 responses (56.84%). C157, “Rate the factual reliability of the AI tools' outputs on this engagement before any human review.” (51 responses), C213, “I or my management discussed AI output alteration with the client” (45 responses), C159, “Rate the consistency of the tools' outputs when given the same input and instructions more than once.” (16 responses) and C215, “My greater team regularly alters AI outputs.” (8 responses) followed.

The most frequent theme was “Told vendor-side management”, in 72 responses (75.79%). The second was “Human rework absorbs the failure”, in 45 responses.

Of the respondents who answered, 59.29% said they had discussed the AI output alteration with internal management. At engagement level, the answer was not associated with any of the four study measures. Two of the respondent-level clustered checks, for the engagement indicator mean and for intent to leave, reached significance where the engagement-level tests did not, so the absence of an association did not hold at respondent level.

The written evidence describes disclosure that stopped inside the vendor. Respondents who discussed the alteration told the team lead or manager, and few responses said the financial institution had been told. Respondents who told no one described the same corrections made in silence, some because raising it had changed nothing or would mean a hard conversation with their own manager.

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 (95 responses on 61 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 practice in other words is not counted, so each count is a floor.