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.

After Holm correction, the share answering True differed by engagement status and lifecycle stage (chi-square 37.31, Holm p < 0.001, Cramer's V 0.53): the share answering True was highest among respondents on canceled engagements (69.62%, n = 316 respondents) and lowest among respondents on active engagements in Transition/Cut-Over (36.59%, n = 246 respondents). It also differed by respondent role (chi-square 215.77, Holm p < 0.001, Cramer's V 0.51): the share answering True was highest among individual contributors (63.43%, n = 659 respondents) and lowest among portfolio directors (0.00%, n = 30 respondents). It also differed by vendor type (chi-square 19.90, Holm p = 0.007, Cramer's V 0.15): the share answering True was highest among respondents at vendors of the type “BPO specialist” (55.33%, n = 197 respondents) and lowest among respondents at vendors of the type “Pure-play agentic AI firm” (42.67%, n = 75 respondents). Outsourcing model, institution type, process family and class of work showed no difference.

The 578 written responses came from 378 respondents on 128 engagements, and 571 were negative. The respondents who answered C202, on why they altered the data, described data from the financial institution that arrived unusable, every cycle and after earlier cleaning, and access too limited to correct it at source. The respondents who commented under C204 described the alteration as routine and largely out of the financial institution's sight.

Of the 839 respondents who answered, 421 (50.18%) answered True and 418 (49.82%) answered False. The non-answers were I do not know (1) and I prefer to not answer (5). Each engagement was assigned the answer given by the largest number of its respondents. True was the assigned answer on 76 engagements and False on 59, and 15 engagements were tied and carry no assigned answer.

The primary responses numbered 548, written in answer to C202, C204, C208, C210 and C214, the written-answer questions of the indicator's survey section. The remaining 30 responses, 5.19% of the total, were secondary and came from 2 other questions. That secondary share ranked 2nd 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 14. Negative responses made up 98.91% of primary responses (542 of 548) and 96.67% of secondary responses (29 of 30).

The 281 responses to C202 gave the reasons for altering the data. Across 94 responses, respondents named access: a legacy platform with no way in for an agent, a security team that would not approve a connection, read-only or expired credentials, or data copied off the display because no export existed. Respondents named missing, blank or dropped rows and values in 48 responses, a format or layout that changed between deliveries in 42, data that did not match between systems or a field that meant different things in different feeds in 44, and data that arrived dirty again after the team had cleaned it in 41. Reference data years out of date appeared in 33 responses and duplicate records in 28. Several added that decisions came after the deadline they were needed for, or that the AI tool gave a different result on each run.

The 247 responses to C204 described how the alteration was handled. Respondents placed it every day, cycle, week or file, or on most cases, in 161 responses. Respondents said the alteration was not reported, logged or raised, or was done quietly, in 78 responses, and that the financial institution was not told, or that they did not know whether it was, in 38. The team lead or manager appeared in 88, most often as someone who knew and could not change the source. Five responses across the indicator said the financial institution had been told or had asked for the corrections. Several respondents said they had become quick at spotting a bad file, and several said the work had worn them down.

Twelve responses described a private sheet, list, mapping file or tracker kept by the respondent to correct the data before each run. The responses to C208, C210 and C214 described altering the financial institution's processes or AI outputs as well as data, often because no one would make a decision in time.

Of the 578 responses, 571 (98.79%, 95% CI 97.86% to 99.62%) were negative and 246 (42.56%) carried the strongest negative score of −10. Positive responses numbered 6 (1.04%), and 1 was neutral. The mean written score was −7.0 (95% CI −7.2 to −6.8, n = 578 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. After Holm correction, the share answering True differed by engagement status and lifecycle stage (chi-square 37.31, Holm p < 0.001, Cramer's V 0.53): the share answering True was highest among respondents on canceled engagements (69.62%, n = 316 respondents) and lowest among respondents on active engagements in Transition/Cut-Over (36.59%, n = 246 respondents). It also differed by respondent role (chi-square 215.77, Holm p < 0.001, Cramer's V 0.51): the share answering True was highest among individual contributors (63.43%, n = 659 respondents) and lowest among portfolio directors (0.00%, n = 30 respondents). It also differed by vendor type (chi-square 19.90, Holm p = 0.007, Cramer's V 0.15): the share answering True was highest among respondents at vendors of the type “BPO specialist” (55.33%, n = 197 respondents) and lowest among respondents at vendors of the type “Pure-play agentic AI firm” (42.67%, n = 75 respondents). Outsourcing model, institution type, process family and class of work showed no difference.

Written sentiment did not differ by engagement status and lifecycle stage across the 128 engagements with a written response (Kruskal-Wallis H 6.50, df 3, p = 0.090). Of the groups with 10 or more responses, responses from canceled engagements were the most negative (mean −7.3, 305 responses) and responses from active engagements in Transition/Cut-Over the least negative (mean −6.7, 125 responses).

No test across respondent roles was run, because fewer than two roles had 30 or more written responses. Individual contributors wrote 561 responses, engagement managers 16 and portfolio directors 1.

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, 47 of 76 (61.84%) were canceled, against 6 of 59 (10.17%) of those whose respondents most often answered False (chi-square 37.19, p < 0.001, Cramer's V 0.52, n = 135 engagements). The difference reached significance and held after Holm correction in 8 of the 9 segment levels tested.

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 57.4% (n = 29 engagements) where respondents most often answered True and 67.8% (n = 53 engagements) where they most often answered False (Welch F 17.74, p < 0.001, omega squared 0.176, n = 82 engagements). The difference reached significance and held after Holm correction in 4 of the 4 segment levels tested.

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 3.1 (n = 76 engagements) on engagements whose respondents most often answered True and 5.1 (n = 59 engagements) on those whose respondents most often answered False (Welch F 93.60, p < 0.001, omega squared 0.415, n = 135 engagements). The difference reached significance and held after Holm correction in 10 of the 10 segment levels tested.

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 8.0 (n = 76 engagements) on engagements whose respondents most often answered True and 5.4 (n = 59 engagements) on those whose respondents most often answered False (Welch F 55.34, p < 0.001, omega squared 0.293, n = 135 engagements). The difference reached significance and held after Holm correction in 9 of the 10 segment levels tested.

Two comparisons set the engagements with at least one negative response against all other engagements. Of the 22 other engagements, 22 had no written response addressing the indicator. These two comparisons therefore do not isolate written sentiment. Of the 128 engagements with a negative response, 55 (42.97%) were canceled, against 1 of the 22 other engagements (4.55%). The difference reached significance (Fisher exact p < 0.001). Customer satisfaction among active engagements differed between the two groups (mean 61.89% on 73 engagements against 71.76% on 21, Welch t −3.83, p < 0.001).

Two correlations test written sentiment directly. Across the 128 engagements with a written response, a more negative engagement mean written score came with cancellation (point-biserial r −0.20, p = 0.024). Across 346 respondents, a more negative respondent mean written score came with higher intent to leave (Spearman rho −0.18, p < 0.001). Of the four comparisons, all 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 the engagement indicator mean (0.415) and weakest with customer satisfaction (0.176).

Answers clustered within engagements (ICC 0.20). Clustering reduced the 845 respondent answers to the equivalent of about 434 independent answers. Respondent-level intervals therefore used the cluster bootstrap. Respondent-level tests used standard errors that allow for clustering by engagement.

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.

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.

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, 98.79%, ranked 6th from the highest, against a median of 98.03%. Its mean written score, −7.0, ranked 5th from the most negative, against a median of −6.3.

The indicator most often present in the same responses was C094, “Rate the quality of the customer's data across the systems this engagement depends on.”, in 341 of the 578 responses (59.00%). C200, “I discussed the data alteration with internal management” (128 responses), C108, “Rate the customer's ability to provision the system access and identities that AI agents require.” (92 responses), C096, “Rate how well the customer's data is normalized, with consistent formats and definitions across source systems.” (88 responses) and C201, “I or my management discussed data alteration with the client” (74 responses) followed.

The most frequent theme was “Occurs every cycle or most days”, in 215 responses (37.20%). The second was “Told vendor-side management”, in 102 responses.

Half of the respondents who answered said they had altered data provided by the financial institution. 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 answer was given more often by individual contributors and on canceled engagements.

The written evidence describes alteration as routine work forced by the state of the data and the lack of access to its source, done every cycle and mostly kept within the team. Respondents rarely said the financial institution had been told. The associations do not establish that alteration produced the outcomes. The written responses describe alteration as one response to the same data and access problems that the other Part 5 indicators record.

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 (578 responses on 128 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.