Rate your employer's quality assurance capability for agents whose outputs are not identical every run

The mean rating was 3.7 and the median 4.0 (n = 794 respondents). Ratings of 3 or below, the distressed range, made up 47.48% of the answers. The rating was associated with all four study measures. Higher ratings came with less cancellation (r = −0.59, 95% CI −0.68 to −0.47, p < 0.001, n = 150 engagements), higher customer satisfaction among active engagements (r = 0.70, 95% CI 0.58 to 0.79, 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 = −0.61, cluster bootstrap 95% CI −0.68 to −0.53, p < 0.001 with standard errors clustered by engagement, n = 745 respondents). After Holm correction, ratings differed by engagement status and lifecycle stage (Welch F 29.79, Holm p < 0.001, omega squared 0.335) and vendor type (Welch F 3.19, Holm p < 0.050, omega squared 0.012).

The indicator drew 151 written responses from twelve questions. Of these, 126 were negative. Every response to C029 named quality assurance for agent output among the hardest challenges. Responses to the operational questions described output that differed between runs on the same input, staff who would not sign it off as it stood, and checking done by whoever handled the case. A more negative written score came with higher intent to leave.

Ratings of 3 or below, the distressed range, accounted for 377 of the 794 answers (47.48%). The engagement mean fell in the distressed range in 61 of 150 engagements.

The primary responses numbered 35, all written in answer to C135, “Add any comments on your employer that the ratings in this section do not capture.” The remaining 116 responses, 76.82% of the total, were secondary and came from 11 other questions. That secondary share ranked 8th 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 5. Negative responses made up 74.29% of primary responses (26 of 35) and 86.21% of secondary responses (100 of 116).

All 58 responses to C029 named quality assurance or evaluation of agent output, often in terms of output that is not the same on every run. Fifteen responses to C135 said there was no method for checking such output, or that teams invented their own: a quality sheet built for work in which the same input gives the same output, a checklist a lead drew on a whiteboard, an internal audit of a deployment pipeline that used a checklist written for manual releases.

Thirty-one responses to the operational questions described output that differed between runs on the same input and staff who would not put their names to it as it stood. The same sentence recurs in near-identical form: the respondent cannot sign off something that will not reproduce. What respondents did next varied. Nine described choosing one of several runs, the middle result, the best of three or the one that looked right, and others rewrote the output by hand. Nine responses said there was no quality assurance step and the respondent or the team was the check. One respondent screenshotted every case before submitting it.

The six responses to C055 described quality reporting that concealed the gap: a consistency check that compared the output only to the agent's previous run, audit samples drawn from the agent's own high-confidence batch, a fixed weekly sample that shrank as volume grew, and test-pass rates from a suite that never covered the paths the agents took. Of the twenty-one positive responses, several described quality staff who caught errors, a second-read process or a test harness. Three described a fixed routine built to detect drift, such as re-running a known agreement every Monday and stopping the week's work if it missed more than one clause.

Of the 151 responses, 126 (83.44%, 95% CI 77.70% to 88.96%) were negative and 68 (45.03%) carried the strongest negative score of −10. Positive responses numbered 21 (13.91%), and 4 were neutral. The mean written score was −5.7 (95% CI −6.5 to −4.9, n = 151 responses), and the median was −5.0.

Of the 151 responses, 144 came from respondents who also rated the indicator. Among them, the written score rose with the rating (Spearman rho 0.44, 95% CI 0.28 to 0.58, p < 0.001, n = 144 responses). The written responses corroborated the rating. Of the 121 negative responses with a rating, 73 (60.33%) came from respondents whose rating was in the distressed range.

Respondents who wrote about the indicator and gave a rating numbered 133, with a mean rating of 3.4. The 661 respondents who rated the indicator without writing about it had a mean of 3.7. The difference, −0.3 (95% CI −0.7 to 0.1), included zero. Ratings in the distressed range made up 52.63% of the ratings from respondents who wrote (70 of 133) and 46.44% of the ratings from respondents who did not (307 of 661).

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 29.79, Holm p < 0.001, omega squared 0.335). Active engagements in Production rated highest (mean 4.7, n = 98 respondents) and canceled engagements lowest (mean 2.5, n = 302 respondents). Ratings also differed by vendor type (Welch F 3.19, Holm p < 0.050, omega squared 0.012). Pure-play agentic AI firms rated highest (mean 4.3, n = 69 respondents) and BPO specialists lowest (mean 3.4, n = 185 respondents). Outsourcing model, institution type, process family, class of work and respondent role showed no difference.

Written sentiment differed by engagement status and lifecycle stage across the 91 engagements with a written response (Kruskal-Wallis H 9.73, df 3, p = 0.021). Of the groups with 10 or more responses, responses from canceled engagements were the most negative (mean −6.8, 67 responses) and responses from active engagements in Transition/Cut-Over the least negative (mean −3.7, 45 responses).

No test across respondent roles was run, because fewer than two roles had 30 or more written responses. Individual contributors wrote 120, engagement managers wrote 24 and portfolio directors wrote 7 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 = −0.59, 95% CI −0.68 to −0.47, Holm p < 0.001, n = 150 engagements). The association held after Holm correction in eleven 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 = 0.70, 95% CI 0.58 to 0.79, Holm p < 0.001, n = 94 engagements). The association held after Holm correction in fourteen 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 = −0.61, 95% CI −0.65 to −0.56, Holm p < 0.001, n = 745 respondents). The association held after Holm correction in thirty-six of the 37 segment levels tested.

Two comparisons set the engagements with at least one negative response against all other engagements. Of the 69 other engagements, 59 had no written response addressing the indicator. These two comparisons therefore do not isolate written sentiment. Of the 81 engagements with a negative response, 35 (43.21%) were canceled, against 21 of the 69 other engagements (30.43%). The difference did not reach significance (Fisher exact p = 0.128). Customer satisfaction among active engagements did not differ between the two groups (mean 65.07% on 46 engagements against 63.17% on 48, Welch t 0.79, p = 0.433).

Two correlations test written sentiment directly. Across the 91 engagements with a written response, the engagement mean written score was not associated with cancellation (point-biserial r −0.19, p = 0.071). Across 130 respondents, a more negative respondent mean written score came with higher intent to leave (Spearman rho −0.36, p < 0.001). 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 cancellation the weakest. The associations with customer satisfaction and intent to leave did not differ (Holm p = 0.277).

Answers clustered within engagements (ICC 0.63). Clustering reduced the 794 respondent answers to the equivalent of about 214 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, 83.44%, ranked 30th from the highest, against a median of 80.82%. Its mean written score, −5.7, ranked 21st from the most negative, against a median of −4.7.

The indicator most often present in the same responses was C159, “Rate the consistency of the tools' outputs when given the same input and instructions more than once.”, in 53 of the 151 responses (35.10%). C157, “Rate the factual reliability of the AI tools' outputs on this engagement before any human review.” (25 responses), C211, “I have altered AI outputs” (19 responses), C122, “Rate your employer's current supply of qualified agentic AI talent against what its engagements need.” (15 responses) and C123, “Rate the adequacy of your employer's investment in training its workforce.” (15 responses) followed.

The most frequent theme was “AI output cannot be signed off”, in 26 responses (17.22%). The second was “Human rework absorbs the failure”, in 15 responses.

Nearly half of all ratings of the employer's quality assurance capability for agents whose outputs vary fell in the distressed range (377 of 794 answers, 47.48%). 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 2.5, n = 302 respondents).

The written evidence describes quality assurance carried by individuals at the point of sign-off. Respondents could not certify output that changed between runs, and in the absence of a method they reran, chose or rewrote it themselves. Written sentiment was associated with 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 (151 responses on 91 engagements). Intervals for written sentiment therefore used a bootstrap by engagement.