Rate how much genuine scrutiny agent outputs receive at human review points rather than routine approval

The mean rating was 3.6 and the median 4.0 (n = 794 respondents). The rating was associated with all four study measures. Engagements with higher ratings were canceled less often (r = −0.64, 95% CI −0.73 to −0.53, p < 0.001, n = 150 engagements). Higher ratings accompanied higher customer satisfaction among active engagements (r = 0.62, 95% CI 0.48 to 0.73, p < 0.001, n = 94 engagements) and a higher engagement indicator mean, computed without the indicator (r = 0.85, 95% CI 0.80 to 0.89, p < 0.001, n = 150 engagements). Respondents who gave higher ratings reported lower intent to leave (r = −0.69, cluster bootstrap 95% CI −0.74 to −0.63, p < 0.001, n = 747 respondents). After Holm correction, ratings differed by engagement status and lifecycle stage (Welch F 43.93, Holm p < 0.001, omega squared 0.404) and by respondent role (Welch F 11.11, Holm p < 0.001, omega squared 0.020).

The written responses addressing the indicator numbered 66, from 63 respondents on 50 engagements. Of these, 59 (89.39%) were negative and 23 (34.85%) carried the strongest negative score of −10. The most common account, in 36 of the 50 primary responses, was a review point that approved agent output without reading it. Respondents who wrote rated the indicator lower than respondents who did not (mean 2.6, n = 62 respondents, against mean 3.7, n = 732 respondents).

Ratings of 3 or below, the distressed range, accounted for 393 of the 794 answers (49.50%). The engagement mean fell in the distressed range in 64 of 150 engagements.

The primary responses numbered 50, all written in answer to C156, the request for comment on the joint management of the engagement. The remaining 16 responses, 24.24% of the total, were secondary and came from 4 other questions. That secondary share ranked 32nd from the highest among the 71 Part 2 indicators with 30 or more written responses, against a median of 18.92%. One respondent wrote both a primary and a secondary response.

Of the 50 primary responses, 36 described the review point as an approval given without reading the output: a click, a tick, a signature or a formality. Respondents reported approvals returned in minutes for batches of hundreds of lines. Nine primary responses attributed the shallow review to too few reviewers for the volume of agent output and not to the reviewers themselves.

The largest secondary sources were C055, on how the vendor manipulates performance reports and dashboards (6 responses), and C216, the comment field on the regular alteration of AI outputs (6 responses). Every response to C055 was negative and described reporting that counted an approval as a completed review. The responses to C216 described output that passed without an independent check, so that respondents corrected it themselves. Negative responses made up 94.00% of primary responses (47 of 50) and 75.00% of secondary responses (12 of 16). The difference did not reach significance (Fisher exact p = 0.053, n = 66 responses).

Of the 66 responses, 59 (89.39%, 95% CI 81.25% to 95.95%) were negative and 23 (34.85%) carried the strongest negative score of −10. Positive responses numbered 6 (9.09%), and one response was neutral. The mean written score was −5.5 (95% CI −6.7 to −4.3, n = 66 responses), and the median was −5.0.

Each of the six positive responses described a check that was actually performed, by a reviewer, a lead, a supervisor or the respondent. Three of these came from C156 and still described other failures of joint management, such as the absence of a named owner for each step.

Of the 66 responses, 65 came from respondents who also rated the indicator. Among them, the written score showed a weak positive correlation with the rating that did not reach significance (Spearman rho 0.21, 95% CI −0.04 to 0.43, p = 0.091, n = 65 responses). Of the 58 negative responses with a rating, 45 (77.59%) came from respondents whose rating was in the distressed range of 3 or below.

Respondents who wrote about the indicator and gave a rating numbered 62, with a mean rating of 2.6. The 732 respondents who rated the indicator without writing about it had a mean of 3.7. The difference, −1.1 (95% CI −1.6 to −0.6), excluded zero. Ratings in the distressed range made up 75.81% of the ratings from respondents who wrote (47 of 62) and 47.27% of the ratings from respondents who did not (346 of 732).

The respondents who wrote came disproportionately from the distressed end of the scale. The written evidence therefore gives more weight to the respondents who rated the scrutiny of agent output lowest.

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 43.93, Holm p < 0.001). Active engagements in Build-Out rated highest (mean 4.7, n = 160 respondents) and canceled engagements lowest (mean 2.2, n = 300 respondents). Ratings also differed by respondent role (Welch F 11.11, Holm p < 0.001). Portfolio directors rated highest (mean 5.0, n = 30 respondents) and individual contributors lowest (mean 3.5, n = 612 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 50 engagements with a written response (Kruskal-Wallis H 0.81, df 3, p = 0.847). Negative responses were the majority in every stage group. Active engagements in Production produced eight responses, reported as counts only.

No test across respondent roles was run, because only individual contributors wrote 30 or more responses (55). All nine responses from engagement managers were negative, reported as counts only.

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 = −0.64, 95% CI −0.73 to −0.53, Holm p < 0.001, n = 150 engagements). The association held after Holm correction in 16 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 = 0.62, 95% CI 0.48 to 0.73, Holm p < 0.001, n = 94 engagements). The association held after Holm correction in 12 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 = 0.85, 95% CI 0.80 to 0.89, Holm p < 0.001, n = 150 engagements). The association held after Holm correction in all 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 = −0.69, Holm p < 0.001, n = 747 respondents). The association held after Holm correction in all 37 segment levels tested.

Two comparisons set the engagements with at least one negative response against all other engagements. Of the 105 other engagements, 100 had no written response addressing the indicator. These two comparisons therefore do not isolate written sentiment. Of the 45 engagements with a negative response, 20 (44.44%) were canceled. Of the 105 other engagements, 36 (34.29%) were canceled. The difference did not reach significance (Fisher exact p = 0.271). Customer satisfaction among active engagements did not differ between the two groups (mean 62.12% on 25 engagements against 64.81% on 69, Welch t −1.01, p = 0.320).

Two correlations test written sentiment directly. Across the 50 engagements with a written response, the engagement mean written score was not associated with cancellation (point-biserial r −0.02, p = 0.870). Across 57 respondents, the respondent mean written score was not associated with intent to leave (Spearman rho −0.09, p = 0.487). None of the four comparisons 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, 5 of the 6 pairs differed.

Clustering within engagements reduced the 794 respondent answers to the equivalent of about 203 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. No result was inconclusive. 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 0.86 to 0.85.

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 every measure.

Among the 71 Part 2 indicators with 30 or more written responses, the indicator's share of negative responses, 89.39%, ranked 13th from the highest, against a median of 80.82%. Its mean written score, −5.5, ranked 24th from the most negative, against a median of −4.7.

Two indicators were most often present in the same responses, each in 16 of the 66 responses (24.24%): C136, on how effectively the financial institution and the vendor jointly manage the engagement, and C148, on whether joint human review capacity keeps up with the volume of agent output. C147, on the placement of human review points across the two firms, followed with 11 responses. C052 and C053, on whether the performance reports given to the financial institution are accurate and whether the vendor manipulates them, appeared only in secondary responses.

The most frequent theme was “The scale or question does not fit”, in 8 responses (12.12%). “Respondent proposes a remedy” and “Positive with an embedded negative” followed with 6 responses each.

Ratings in the distressed range made up 49.50% of the 794 answers. The rating was associated with all four study measures, and most strongly with the engagement indicator mean (r = 0.85, n = 150 engagements). Every association held under each of the 22 robustness checks. Ratings were highest among active engagements in Build-Out (mean 4.7, n = 160 respondents) and lowest among canceled engagements (mean 2.2, n = 300 respondents). Individual contributors, the role that performs the engagement work, gave the lowest ratings (mean 3.5, n = 612 respondents).

The written responses show what lay behind the low rating: approval at the human review point without a reading of the agent's output. The reason respondents gave most often was review capacity that did not match the volume of output (9 primary responses). Negative written sentiment was not associated with any study measure. The rating was. Written responses were voluntary. They recorded what the respondents who wrote reported, and how common it was is measured by the ratings. The written evidence rests on 66 responses from 50 engagements.

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. Respondents reported on the financial institution at second hand.

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. Answers clustered within engagements (ICC 0.68). The 794 answers carried the information of about 203 independent answers.

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 (66 responses on 50 engagements). Intervals for written sentiment therefore used a bootstrap by engagement.