How many AI agents are running in production on this engagement today?

At least half of respondents reported no AI agents running in production on their engagement (mean 1.2, median 0.0, n = 787 respondents). The association with cancellation was equivalent to zero (TOST p < 0.001). Customer satisfaction on active engagements showed the same tested null (TOST p 0.008). The association with the engagement indicator mean was also equivalent to zero (TOST p < 0.001). Intent to leave returned a tested null as well (TOST p < 0.001). The agent count differed after Holm control by class of work only (Welch F 10.30, Holm p 0.007, omega squared 0.057).

The indicator had 4 written responses coded to it, from 4 of 850 respondents on 4 of 150 engagements. Negative responses numbered 2 of the 4 scored responses. No response carried the strongest negative score, −10. Every response was a secondary response, written in answer to a question outside the indicator’s own survey section.

The study set mean was 1.2 on 787 answering respondents, with a median of 0.0.

All 4 responses were secondary responses. The secondary responses came from 4 respondents on 4 engagements, in answer to 1 question outside the indicator’s own survey section (C135 with 4 responses).

The indicator has no good or bad end. Under the coding rulebook’s rule for such indicators, the written score records the effect each response states on the engagement or on the respondent. The survey question is “How many AI agents are running in production on this engagement today?” The frequency, scope and consequence a response states set the strength of its score. A score of +10 marks a response that describes “this attribute of the engagement as helping the engagement at its best.” A score of −10 marks a response that describes “this attribute as harming the engagement at its worst . . . .”

The distribution was reported as counts only, because fewer than 10 responses carried a score. Of the 4 scored responses, 2 were negative, none carried the strongest negative score and 1 was positive.

Figures are respondent level. Means are on the scale as recorded. A level under 30 prints no interval and a level under 10 prints counts only.

Engagement families were tested on engagement means, and respondent role and vendor type were tested on respondents. A distressed p marked exact is a Fisher exact test for a two by two table, and one marked simulated is a Monte Carlo test run until its p is stable to the third decimal, both used where an expected count is under 5. Class of work differed after Holm control (Welch F 10.30, Holm p 0.007, tested on engagement means). In Descriptive Statistics by Segment, LOB Operational had the highest mean, 1.9 (n = 419 respondents), and Corporate Functional the lowest, 0.5 (n = 368 respondents). No difference held after Holm control for outsourcing model, process family and respondent role.

The share of engagements canceled on the study date, measured at engagement level (base 150 engagements, 56 canceled). The association is a descriptive contrast, inflated because canceled engagements were rated after cancellation, and is not predictive.

The study set r was −0.01 (95% CI −0.17 to 0.15, Holm p 1.000, n = 150 engagements). The association did not hold after Holm control. Equivalence Tests of Null Results found it equivalent to zero inside the bound. Of the 19 tested segment levels, 0 held after Holm control.

The satisfaction score recorded against each engagement on a 100-point scale. The primary estimate uses the 94 active engagements. A descriptive contrast uses all 150 engagements.

The study set r was 0.06 (95% CI −0.15 to 0.26, Holm p 1.000, n = 94 engagements). The association did not hold after Holm control. Equivalence Tests of Null Results found it equivalent to zero inside the bound. Of the 14 tested segment levels, 0 held after Holm control.

The mean of an engagement’s 100 indicator ratings, measured at engagement level (base 150 engagements). Where this indicator is one of the 100, the mean is recomputed without it.

The study set r was 0.03 (95% CI −0.13 to 0.19, Holm p 1.000, n = 150 engagements). The association did not hold after Holm control. Equivalence Tests of Null Results found it equivalent to zero inside the bound. Of the 21 tested segment levels, 0 held after Holm control.

The respondent’s stated likelihood of leaving the current job within 12 months, rated 1 to 10 (C019). Measured at respondent level on up to 798 respondents, with the engagement-level form reported alongside.

The study set r was 0.01 (95% CI −0.06 to 0.08, Holm p 1.000, n = 742 respondents). The association did not hold after Holm control. Equivalence Tests of Null Results found it equivalent to zero inside the bound. Of the 33 tested segment levels, 4 held after Holm control.

Pairs were compared on the engagement level form of every measure, each turned so that a higher value was the better outcome, on the 150 engagements where the indicator and all four measures existed. Of the 6 pairs, 0 differed after Holm control.

Clustering reduced the 787 respondent answers to about 150 independent answers. Respondent level intervals therefore used the cluster bootstrap, and respondent level tests on intent to leave used cluster-robust errors.

Tested null, equivalent to zero: cancellation rate, customer satisfaction, engagement indicator mean and intent to leave. No result was inconclusive. The detection floor column is the study floor at 150 engagements. Intent to leave was tested on its effective n.

Each change was computed before rounding, so it can differ by 0.01 from the difference of the two printed values.

Every check held. A check holds when its test lands on the same side of 0.05 as the primary test and, where the primary test is below 0.05, its r keeps the primary sign. Thin-base rule: engagements in any level under 30 engagements of outsourcing model or process family removed. Institution type is not used, because every one of its levels is under 30.

Each of the four measures was computed from data this indicator does not contain, so each result is a finding rather than an arithmetic consequence.

Each deviation was computed before rounding, so it can differ by 0.01 from the difference of the two printed values.

The comparison ran on 150 engagements. The indicator departed from the study’s typical pattern on cancellation rate, customer satisfaction, engagement indicator mean and intent to leave.

The co-coded indicators tied as most frequent in the responses coded to the indicator were “Rate how well your employer’s training programs build financial industry knowledge” (C125, 3 responses) and “Rate your employer’s ability to operate large numbers of agents in production, including monitoring and exception handling” (C133, 3 responses).

LOB Operational respondents reported more agents in production (mean 1.9, n = 419 respondents) than Corporate Functional respondents (mean 0.5, n = 368 respondents). The tested nulls held under every robustness check run on this item (22 of 22 checks).

The indicator is a survey question that is not a rating, so the written responses were not set against a rating. Written responses were voluntary. They recorded what the respondents who wrote reported, and how common it was is measured by the answers. The analysis of the written responses rested on 4 responses, a base of anecdote strength.

The evidence is reported experience: what the delivery team said about its own engagement, gathered in one cross-section. The figures showed associations, not causes. Indicators about the financial institution were reported at second hand by vendor staff. Cancellation is a descriptive contrast, because canceled engagements were rated after they ended.

Tests across respondent role and vendor type were run at respondent level and treated respondents as independent. The section Clustering and Effective Sample Size states whether, and by how much, clustering within engagements reduced the information in respondent level figures.

Answers matched within engagements, so the item carried about one independent answer per engagement (ICC 1.00, 787 answers worth about 150 independent answers).

Fewer than 10 responses carried a score, so written sentiment was reported as counts, with no statistic. Coding was performed by AI coders under a fixed rulebook. Agreement on whether a response was coded to the indicator rested on 1 response second-coded under rulebook 0.5, with the coders agreeing on 1 of 1. No response coded to the indicator was second-scored under rulebook 0.6, so agreement on the score could not be assessed. This base of second-coded responses is too small for an indicator-level alpha.