Rate how well human review points are placed across the two firms at the moments where agent errors would matter most

The mean rating was 3.6 and the median 4.0 (n = 797 respondents). Ratings of 3 or below, the distressed range, made up 49.31% of the answers. The rating was associated with all four study measures. Higher ratings came with less cancellation (r = −0.62, 95% CI −0.71 to −0.51, p < 0.001, n = 150 engagements), higher customer satisfaction among active engagements (r = 0.60, 95% CI 0.46 to 0.72, p < 0.001, n = 94 engagements), a higher engagement indicator mean computed without the indicator (r = 0.84, 95% CI 0.79 to 0.88, p < 0.001, n = 150 engagements) and lower intent to leave (r = −0.66, cluster bootstrap 95% CI −0.72 to −0.60, p < 0.001 with standard errors clustered by engagement, n = 747 respondents). After Holm correction, ratings differed by engagement status and lifecycle stage (Welch F 41.35, Holm p < 0.001, omega squared 0.382) and respondent role (Welch F 10.14, Holm p < 0.001, omega squared 0.014).

Written responses addressing the indicator numbered 105, from 103 respondents on 75 engagements. Of these, 75 were negative and 25 positive. Respondents described three faults in human review. Some review points sat at the end of the process, after an error had already gone out. Some approvals were ticked through without anyone reading what the agent did. Some output went unchecked by anyone except the respondent. Others said the review points were in the right place and the gap lay elsewhere. Responses to the question on manipulated reports described dashboards that hid what the agent had closed without a human check. Respondents whose written responses were more negative reported higher intent to leave.

Ratings of 3 or below, the distressed range, accounted for 393 of the 797 answers (49.31%). The engagement mean fell in the distressed range in 65 of 150 engagements.

The primary responses numbered 70, all written in answer to C156, “Add any comments on the joint management of this engagement that the ratings in this section do not capture.” The remaining 35 responses, 33.33% of the total, were secondary and came from 7 other questions. That secondary share ranked 28th from the highest among the 71 Part 2 indicators with 30 or more written responses, against a median of 18.92%. Respondents who wrote both a primary and a secondary response numbered 2. Negative responses made up 68.57% of primary responses (48 of 70) and 77.14% of secondary responses (27 of 35).

Nine responses stated that the review point sat at the end of the process: a wrong dispute reason found after the letter was sent, a file reviewed after it had been packaged for the investor, checks that caught typing errors and missed rating rules. Seven described review that was ticked through: forty findings cleared in an hour, a same-day approval that let a wrong tier rate pass both firms, approvers who did not read the message before it went out. Seventeen responses, from the joint management field and the alterations and employer fields, stated that no one checked the agent's output before it went out, or that the respondent was the only check: a cash match on the wrong value date sent out with the pack, a denial letter sent to a claimant with no check, a forecast pack whose only reviewer was the analyst who built it. Seven described the load this placed on a few people, including engineers reading generated code until nine at night and colleagues who resigned or asked to move.

Against these, thirteen responses said the review points were placed correctly and located the gap elsewhere: in sign-off on changes, in visibility, or in speed. Five described checks the team had built for itself, such as a two-person reading rota each morning, sample checks against the source file, small batches, or an agent that could not commit anything without a person approving it. The six responses to C055, the question on how the vendor manipulates performance reports and dashboards, described reports that omitted breaks the agent resolved without a human check, alerts auto-closed without sign-off, and tickets the agent closed logged as resolved.

Of the 105 responses, 75 (71.43%, 95% CI 62.00% to 80.53%) were negative and 27 (25.71%) carried the strongest negative score of −10. Positive responses numbered 25 (23.81%), and 5 were neutral. The mean written score was −3.4 (95% CI −4.6 to −2.1, n = 105 responses), and the median was −5.0.

Of the 105 responses, 100 came from respondents who also rated the indicator. Among them, the written score rose with the rating (Spearman rho 0.45, 95% CI 0.24 to 0.62, p < 0.001, n = 100 responses). The written responses corroborated the rating. Of the 71 negative responses with a rating, 40 (56.34%) came from respondents whose rating was in the distressed range.

Respondents who wrote about the indicator and gave a rating numbered 98, with a mean rating of 3.8. The 699 respondents who rated the indicator without writing about it had a mean of 3.6. The difference, 0.1 (95% CI −0.4 to 0.7), included zero. Ratings in the distressed range made up 48.98% of the ratings from respondents who wrote (48 of 98) and 49.36% of the ratings from respondents who did not (345 of 699).

Every figure in the segment tables is at respondent level. Segments with fewer than 30 respondents carry no interval. Segments with fewer than 10 report counts only.

Segment families recorded at engagement level were tested on engagement means. Respondent role and vendor type were tested on respondents. Where an expected count fell below 5, the test of the distressed share used Fisher's exact test for a two-by-two table, marked exact, or a Monte Carlo test, marked simulated. After Holm correction, ratings differed by engagement status and lifecycle stage (Welch F 41.35, Holm p < 0.001, omega squared 0.382). Active engagements in Transition/Cut-Over rated highest (mean 4.7, n = 229 respondents) and canceled engagements lowest (mean 2.2, n = 303 respondents). Ratings also differed by respondent role (Welch F 10.14, Holm p < 0.001, omega squared 0.014). Portfolio directors rated highest (mean 4.7, n = 31 respondents) and individual contributors lowest (mean 3.5, n = 614 respondents). Outsourcing model, institution type, process family, class of work and vendor type showed no difference.

Written sentiment did not differ by engagement status and lifecycle stage across the 75 engagements with a written response (Kruskal-Wallis H 2.87, df 3, p = 0.412). Of the groups with 10 or more responses, responses from canceled engagements were the most negative (mean −4.6, 41 responses) and responses from active engagements in Production the least negative (mean −1.3, 12 responses).

No test across respondent roles was run, because fewer than two roles had 30 or more written responses. Individual contributors wrote 75, engagement managers wrote 25 and portfolio directors wrote 5 responses.

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.

Across the study set, higher ratings came with less cancellation (r = −0.62, 95% CI −0.71 to −0.51, Holm p < 0.001, n = 150 engagements). The association held after Holm correction in thirteen of the 20 segment levels tested.

Customer satisfaction is the percentage recorded against each engagement. The primary estimate uses the 94 active engagements. A descriptive contrast uses all 150 engagements.

Across the active engagements, higher ratings came with higher customer satisfaction (r = 0.60, 95% CI 0.46 to 0.72, Holm p < 0.001, n = 94 engagements). The association held after Holm correction in twelve of the 16 segment levels tested.

The engagement indicator mean is the mean of an engagement's ratings across the rated indicators, measured at engagement level (150 engagements). The indicator is one of the rated indicators. Each comparison therefore uses the mean recomputed without it.

Higher ratings came with a higher engagement indicator mean (r = 0.84, 95% CI 0.79 to 0.88, Holm p < 0.001, n = 150 engagements). The association held after Holm correction in twenty-four of the 24 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. It is measured at respondent level on up to 798 respondents, with the engagement-level form reported alongside.

Respondents who gave higher ratings reported lower intent to leave (r = −0.66, 95% CI −0.70 to −0.62, Holm p < 0.001, n = 747 respondents). The association held after Holm correction in thirty-seven of the 37 segment levels tested.

Two comparisons set the engagements with at least one negative response against all other engagements. Of the 92 other engagements, 75 had no written response addressing the indicator. These two comparisons therefore do not isolate written sentiment. Of the 58 engagements with a negative response, 21 (36.21%) were canceled, against 35 of the 92 other engagements (38.04%). The difference did not reach significance (Fisher exact p = 0.864). Customer satisfaction among active engagements did not differ between the two groups (mean 65.27% on 37 engagements against 63.33% on 57, Welch t 0.80, p = 0.423).

Two correlations test written sentiment directly. Across the 75 engagements with a written response, the engagement mean written score was not associated with cancellation (point-biserial r −0.13, p = 0.270). Across 96 respondents, a more negative respondent mean written score came with higher intent to leave (Spearman rho −0.27, p = 0.007). Of the four comparisons, one reached significance at 0.05.

The four associations were compared in pairs on their engagement-level forms, each oriented so that a higher value is the better outcome, across the 150 engagements with all four measures. After Holm correction, five of the six pairs differed. The association with the engagement indicator mean was the strongest and the association with customer satisfaction the weakest. The associations with customer satisfaction and intent to leave did not differ (Holm p = 0.723).

Answers clustered within engagements (ICC 0.65). Clustering reduced the 797 respondent answers to the equivalent of about 209 independent answers. Respondent-level intervals therefore used the cluster bootstrap. Respondent-level tests of intent to leave used standard errors that allow for clustering by engagement.

No association was equivalent to zero. Each of the four differed from zero. The detection floor shown is the study floor at 150 engagements. Intent to leave was tested on its effective sample size.

Each change was computed before rounding and may differ by 0.01 from the difference between the two printed values.

All 22 checks held. A check holds when its test falls on the same side of 0.05 as the primary test. Where the primary test falls below 0.05, the check must also keep the sign of the primary correlation. The thin-base check removes the engagements in any level of outsourcing model or process family with fewer than 30 engagements. Institution type is not used for that check, because every one of its levels has fewer than 30 engagements.

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. The indicator is one of the rated indicators, and the engagement indicator mean was recomputed without it. The correction changed r by 0.00, from 0.85 to 0.84.

Each deviation was computed before rounding and may differ by 0.01 from the difference between the two printed values.

The comparison covered 150 engagements. The indicator followed the study's typical pattern on all four measures, and no deviation held after Holm correction.

Among the 71 Part 2 indicators with 30 or more written responses, the indicator's share of negative responses, 71.43%, ranked 56th from the highest, against a median of 80.82%. Its mean written score, −3.4, ranked 55th from the most negative, against a median of −4.7.

The indicator most often present in the same responses was C136, “Rate how effectively the customer and vendor jointly manage this engagement now that agents perform part of the work.”, in 15 of the 105 responses (14.29%). C148, “Rate how well joint human review capacity keeps up with the volume of output the agents produce.” (15 responses), C157, “Rate the factual reliability of the AI tools' outputs on this engagement before any human review.” (15 responses), C138, “Rate how clearly ownership of each workflow step is assigned when agents, vendor staff, and customer staff work in the same process.” (13 responses) and C149, “Rate how much genuine scrutiny agent outputs receive at human review points rather than routine approval.” (11 responses) followed.

The most frequent theme was “Respondent proposes a remedy”, in 19 responses (18.10%). The second was “Raised it, nothing came back”, in 9 responses.

Nearly half of all ratings of how well human review points are placed across the two firms fell in the distressed range (393 of 797 answers, 49.31%). The rating was associated with all four study measures, and every association held under each of the 22 robustness checks. Canceled engagements rated far below the three active stages (mean 2.2, n = 303 respondents).

The written responses separate where review sits from whether it happens. In some engagements the review point was correctly placed and still reduced to a signature. In others it sat too late to stop an error, or did not exist, and the individual analyst became the control. The reports described in answer to C055 indicate that institutions could not always see how much work the agent completed without any human check.

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.

Tests by respondent role and by vendor type treated respondents as independent. The section on clustering reports how far clustering within engagements reduced the information in respondent-level figures.

Customer management executives gave 2 answers, a base of anecdote strength.

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 (105 responses on 75 engagements). Intervals for written sentiment therefore used a bootstrap by engagement.