Rate the consistency of the tools' outputs when given the same input and instructions more than once

The mean rating was 4.6 and the median 5.0 (n = 800 respondents). Ratings of 3 or below, the distressed range, made up 32.38% of the answers. The rating was associated with all four study measures. Higher ratings came with less cancellation (r = −0.65, 95% CI −0.74 to −0.55, p < 0.001, n = 150 engagements), higher customer satisfaction among active engagements (r = 0.66, 95% CI 0.52 to 0.76, p < 0.001, n = 94 engagements), a higher engagement indicator mean computed without the indicator (r = 0.88, 95% CI 0.83 to 0.91, p < 0.001, n = 150 engagements) and lower intent to leave (r = −0.63, cluster bootstrap 95% CI −0.69 to −0.57, p < 0.001 with standard errors clustered by engagement, n = 750 respondents). After Holm correction, ratings differed by engagement status and lifecycle stage (Welch F 44.38, Holm p < 0.001, omega squared 0.414), respondent role (Welch F 5.60, Holm p = 0.027, omega squared 0.011) and vendor type (Welch F 4.17, Holm p = 0.007, omega squared 0.021).

The indicator drew 383 written responses, and 363 were negative. Respondents described the same input producing a different result on another run, and staff who would not sign off such output, chose among runs, or corrected it by hand before it went on. Engagements with a negative response were canceled far more often than engagements without one, and written sentiment was associated with all four study comparisons.

Ratings of 3 or below, the distressed range, accounted for 259 of the 800 answers (32.38%). The engagement mean fell in the distressed range in 33 of 150 engagements.

The primary responses numbered 53, all written in answer to C177, “Add any comments on the agentic AI tools that the ratings in this section do not capture.” The remaining 330 responses, 86.16% of the total, were secondary and came from 9 other questions. That secondary share ranked 2nd 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 4. Negative responses made up 62.26% of primary responses (33 of 53) and 100.00% of secondary responses (330 of 330).

All 217 responses to C214 were negative, and nearly all described the same input giving a different result on another run, followed by a correction before the output went on. Eighty of these responses described choosing among runs: the one that looked right, the closest of several, the best of three, a blend of a few. Eighteen chose the run that matched history, a prior case or what had already been sent or agreed, so that the earlier answer settled the later one. Twenty-one responses across the operational questions said the respondent could not sign off, or put a name to, output that would not reproduce.

Responses to C202, C204, C208 and C210 added data delivered wrong before the agent touched it, decisions that arrived after their deadlines, and changes made first and documented afterward, with the variation between runs on top. Responses to C216 described corrections made quietly, not logged and not raised. The responses to C135 and C029 described the absence of any method for checking output that varies between runs.

Responses to C177 were divided. Negative responses gave specific cases: two different card-network reason codes on two runs of the same dispute file, the same alert scored low on one day and high three days later, a full quarter rerun as a control with residuals that did not match. One respondent said a regulatory filings pilot never produced the same report twice from the same documents and the financial institution canceled it. The nineteen positive responses said the same input now gave the same output, often as an improvement since the early months, and placed the weakness in long jobs, completion reports or logging.

Of the 383 responses, 363 (94.78%, 95% CI 92.41% to 96.95%) were negative and 131 (34.20%) carried the strongest negative score of −10. Positive responses numbered 19 (4.96%), and 1 was neutral. The mean written score was −6.2 (95% CI −6.5 to −5.8, n = 383 responses), and the median was −5.0.

Of the 383 responses, 358 came from respondents who also rated the indicator. Among them, the written score rose with the rating (Spearman rho 0.27, 95% CI 0.17 to 0.37, p < 0.001, n = 358 responses). The written responses corroborated the rating. Of the 338 negative responses with a rating, 203 (60.06%) came from respondents whose rating was in the distressed range.

Respondents who wrote about the indicator and gave a rating numbered 307, with a mean rating of 3.4. The 493 respondents who rated the indicator without writing about it had a mean of 5.3. The difference, −1.9 (95% CI −2.2 to −1.6), excluded zero. Ratings in the distressed range made up 56.35% of the ratings from respondents who wrote (173 of 307) and 17.44% of the ratings from respondents who did not (86 of 493).

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 indicator worst.

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 44.38, Holm p < 0.001, omega squared 0.414). Active engagements in Transition/Cut-Over rated highest (mean 5.5, n = 230 respondents) and canceled engagements lowest (mean 3.2, n = 309 respondents). Ratings also differed by respondent role (Welch F 5.60, Holm p = 0.027, omega squared 0.011). Portfolio directors rated highest (mean 5.4, n = 31 respondents) and individual contributors lowest (mean 4.5, n = 617 respondents). Ratings also differed by vendor type (Welch F 4.17, Holm p = 0.007, omega squared 0.021). Consulting firms rated highest (mean 5.0, n = 113 respondents) and BPO specialists lowest (mean 4.1, n = 188 respondents). Outsourcing model, institution type, process family and class of work showed no difference.

Written sentiment differed by engagement status and lifecycle stage across the 121 engagements with a written response (Kruskal-Wallis H 11.68, df 3, p = 0.009). Of the groups with 10 or more responses, responses from canceled engagements were the most negative (mean −6.8, 208 responses) and responses from active engagements in Transition/Cut-Over the least negative (mean −5.0, 76 responses).

No test across respondent roles was run, because fewer than two roles had 30 or more written responses. Individual contributors wrote 352, engagement managers wrote 27 and portfolio directors wrote 4 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 = −0.65, 95% CI −0.74 to −0.55, Holm p < 0.001, n = 150 engagements). The association held after Holm correction in fourteen 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 = 0.66, 95% CI 0.52 to 0.76, 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 = 0.88, 95% CI 0.83 to 0.91, 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 = −0.63, 95% CI −0.67 to −0.59, Holm p < 0.001, n = 750 respondents). The association held after Holm correction in thirty-five of the 37 segment levels tested.

Two comparisons set the engagements with at least one negative response against all other engagements. Of the 33 other engagements, 29 had no written response addressing the indicator. These two comparisons therefore do not isolate written sentiment. Of the 117 engagements with a negative response, 55 (47.01%) were canceled, against 1 of the 33 other engagements (3.03%). The difference reached significance (Fisher exact p < 0.001). Customer satisfaction among active engagements differed between the two groups (mean 62.03% on 62 engagements against 68.09% on 32, Welch t −2.54, p = 0.013).

Two correlations test written sentiment directly. Across the 121 engagements with a written response, a more negative engagement mean written score came with cancellation (point-biserial r −0.28, p = 0.002). Across 306 respondents, a more negative respondent mean written score came with higher intent to leave (Spearman rho −0.21, p < 0.001). Of the four comparisons, all 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, four of the six pairs differed. The association with the engagement indicator mean was the strongest and the association with intent to leave the weakest. The associations with cancellation and intent to leave did not differ (Holm p = 0.051). The associations with customer satisfaction and intent to leave did not differ (Holm p = 0.252).

Answers clustered within engagements (ICC 0.60). Clustering reduced the 800 respondent answers to the equivalent of about 223 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 0.88 to 0.88.

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, 94.78%, ranked 2nd from the highest, against a median of 80.82%. Its mean written score, −6.2, ranked 13th from the most negative, against a median of −4.7.

The indicator most often present in the same responses was C211, “I have altered AI outputs”, in 175 of the 383 responses (45.69%). C157, “Rate the factual reliability of the AI tools' outputs on this engagement before any human review.” (152 responses), C127, “Rate your employer's quality assurance capability for agents whose outputs are not identical every run.” (53 responses), C199, “I have altered client-provided data” (22 responses) and C109, “Rate how quickly the customer responds when this engagement raises issues that need customer action.” (20 responses) followed.

The most frequent theme was “Human rework absorbs the failure”, in 124 responses (32.38%). The second was “Occurs every cycle or most days”, in 67 responses.

About a third of all ratings of the consistency of the tools' outputs on repeated input fell in the distressed range (259 of 800 answers, 32.38%). 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 3.2, n = 309 respondents).

The written evidence describes a choice made by the person handling the case whenever runs disagreed, guided by what looked right or what had been sent before. Written sentiment was associated with all four study comparisons: engagements with a negative response were canceled more often, active engagements with one had lower customer satisfaction, and more negative written scores came with cancellation and with higher intent to leave.

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 (383 responses on 121 engagements). Intervals for written sentiment therefore used a bootstrap by engagement.