Rate how reliably the tools' built-in guardrails and safety controls held during production use

The mean rating was 4.7 and the median 5.0 (n = 787 respondents). Ratings of 3 or below, the distressed range, made up 32.02% of the answers. The rating was associated with all four study measures. Higher ratings came with less cancellation (r = −0.66, 95% CI −0.74 to −0.56, p < 0.001, n = 150 engagements), higher customer satisfaction among active engagements (r = 0.65, 95% CI 0.52 to 0.76, p < 0.001, n = 94 engagements), a higher engagement indicator mean computed without the indicator (r = 0.86, 95% CI 0.82 to 0.90, p < 0.001, n = 150 engagements) and lower intent to leave (r = −0.66, cluster bootstrap 95% CI −0.71 to −0.60, p < 0.001 with standard errors clustered by engagement, n = 742 respondents). After Holm correction, ratings differed by engagement status and lifecycle stage (Welch F 51.61, Holm p < 0.001, omega squared 0.424) and vendor type (Welch F 4.64, Holm p = 0.003, omega squared 0.024).

Of the 17 written responses, 10 were positive and 7 negative, the reverse of the balance in most Part 2 indicators. Most positive responses said the guardrails held and placed the respondent's concern elsewhere. The negative responses described guardrails that let through data exposure, prohibited wording or actions on accounts that should have had a second review. No written outcome test reached significance.

Ratings of 3 or below, the distressed range, accounted for 252 of the 787 answers (32.02%). The engagement mean fell in the distressed range in 32 of 150 engagements.

The primary responses numbered 15, 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 2 responses, 11.76% of the total, were secondary and came from 1 other questions. Respondents who wrote both a primary and a secondary response numbered 0. Negative responses made up 46.67% of primary responses (7 of 15) and 0 of the 2 secondary responses.

Eight responses to C177 said the guardrails or safety controls held, one over a whole quarter. One respondent described a sandbox that caught an agent reaching for a settlement system, and another said the controls held except twice. These respondents placed their concerns in memory, consistency between runs, long multi-step jobs and the absence of an audit trail. C030 asked respondents what they loved most about their job, and its two responses listed guardrails that hold among the conditions they valued.

Six responses described a guardrail that let something through. An employee's home address went into a public ticket note, and the respondent said the guardrails caught nothing and the team caught it. Real investor account numbers went into a test file although the guardrail was meant to stop it. On hardship cases the guardrails let through wording the team was not allowed to use with a debtor. One agent pulled statements from a second custodian account because an instruction was loose, and another would action a customer cash account without a second review. One respondent said the safety controls had never been built for the team's kind of tickets, and another said they sometimes blocked clean work.

Of the 17 responses, 7 (41.18%, 95% CI 16.67% to 68.75%) were negative and 2 (11.76%) carried the strongest negative score of −10. Positive responses numbered 10 (58.82%), and no response was neutral. The mean written score was 1.2 (95% CI −2.5 to 4.7, n = 17 responses), and the median was 5.0.

Of the 17 responses, 17 came from respondents who also rated the indicator. Among them, the written score rose with the rating (Spearman rho 0.57, 95% CI 0.14 to 0.84, p = 0.018, n = 17 responses). The written responses corroborated the rating. Of the 7 negative responses with a rating, 4 came from respondents whose rating was in the distressed range.

Respondents who wrote about the indicator and gave a rating numbered 17, with a mean rating of 5.4. The 770 respondents who rated the indicator without writing about it had a mean of 4.6. The difference, 0.8 (95% CI −0.3 to 1.9), included zero. Ratings in the distressed range made up 23.53% of the ratings from respondents who wrote (4 of 17) and 32.21% of the ratings from respondents who did not (248 of 770).

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 51.61, Holm p < 0.001, omega squared 0.424). Active engagements in Transition/Cut-Over rated highest (mean 5.6, n = 238 respondents) and canceled engagements lowest (mean 3.2, n = 292 respondents). Ratings also differed by vendor type (Welch F 4.64, Holm p = 0.003, omega squared 0.024). Consulting firms rated highest (mean 5.3, n = 109 respondents) and BPO specialists lowest (mean 4.2, n = 187 respondents). Outsourcing model, institution type, process family, class of work and respondent role showed no difference.

Written sentiment did not differ by engagement status and lifecycle stage across the 14 engagements with a written response (Kruskal-Wallis H 2.85, df 3, p = 0.415). Fewer than two groups held 10 or more responses, so the groups are reported as counts only: active engagements in Build-Out (4), active engagements in Transition/Cut-Over (9), active engagements in Production (2) and canceled engagements (2).

No test across respondent roles was run, because fewer than two roles had 30 or more written responses. Individual contributors wrote 11, engagement managers wrote 5 and portfolio directors wrote 1 responses, 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 = −0.66, 95% CI −0.74 to −0.56, Holm p < 0.001, n = 150 engagements). The association held after Holm correction in sixteen 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 = 0.65, 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 = 0.86, 95% CI 0.82 to 0.90, 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 −0.69 to −0.61, Holm p < 0.001, n = 742 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 144 other engagements, 136 had no written response addressing the indicator. These two comparisons therefore do not isolate written sentiment. Of the 6 engagements with a negative response, 0 were canceled, against 56 of the 144 other engagements (38.89%). The difference did not reach significance (Fisher exact p = 0.084). The comparison of customer satisfaction was not computed, because one of the two groups held too few active engagements.

Two correlations test written sentiment directly. Across the 14 engagements with a written response, the engagement mean written score was not associated with cancellation (point-biserial r 0.39, p = 0.168). Across 16 respondents, the respondent mean written score was not associated with intent to leave (Spearman rho 0.04, p = 0.890). Of the three comparisons, none 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 = 0.651).

Answers clustered within engagements (ICC 0.61). Clustering reduced the 787 respondent answers to the equivalent of about 220 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 0.87 to 0.86.

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, 41.18%, ranked 71st from the highest, against a median of 80.82%. Its mean written score, 1.2, ranked 71st from the most negative, against a median of −4.7.

The indicator most often present in the same responses was C168, “Rate how consistently the agents took only intended and authorized actions when operating connected tools and systems.”, in 5 of the 17 responses (29.41%). C157, “Rate the factual reliability of the AI tools' outputs on this engagement before any human review.” (4 responses), C160, “Rate the end-to-end reliability of the tools on multi-step tasks, from first step to finished output.” (4 responses), C175, “Rate the stability of the tools' behavior over the engagement when nothing changed on the engagement side.” (4 responses) and C164, “Rate how reliably the tools retained instructions and constraints given earlier in long tasks and sessions.” (3 responses) followed.

The most frequent theme was “Occurs every cycle or most days”, in 1 responses (5.88%). The second was “Effort is invisible in the numbers”, in 1 responses.

About a third of all ratings of how reliably the built-in guardrails held fell in the distressed range (252 of 787 answers, 32.02%). 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 = 292 respondents).

The written responses were few and leaned positive, with a negative share of 41.18% against a Part 2 median of 80.82% among indicators with 30 or more written responses. The ratings and the written evidence therefore point in different directions, and the ratings, on the full base, carry the weight. Where respondents did describe a failure, the control had not anticipated the situation, such as a hardship case or a loose instruction, and a person found the problem.

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 (17 responses on 14 engagements). Intervals for written sentiment therefore used a bootstrap by engagement.

The written responses number fewer than 30, so their shares and the written-sentiment tests rest on a small base.