Rate the predictability of the tools' token consumption for a given volume of work

The mean rating was 4.6 and the median 4.0 (n = 795 respondents). Ratings of 3 or below, the distressed range, made up 31.32% of the answers. The rating was associated with all four study measures. Higher ratings came with less cancellation (r = −0.65, 95% CI −0.73 to −0.54, p < 0.001, n = 150 engagements), higher customer satisfaction among active engagements (r = 0.62, 95% CI 0.48 to 0.73, p < 0.001, n = 94 engagements), a higher engagement indicator mean computed without the indicator (r = 0.86, 95% CI 0.81 to 0.89, p < 0.001, n = 150 engagements) and lower intent to leave (r = −0.64, cluster bootstrap 95% CI −0.70 to −0.58, p < 0.001 with standard errors clustered by engagement, n = 749 respondents). After Holm correction, ratings differed by engagement status and lifecycle stage (Welch F 43.47, Holm p < 0.001, omega squared 0.407), respondent role (Welch F 7.30, Holm p = 0.007, omega squared 0.014) and vendor type (Welch F 4.68, Holm p = 0.003, omega squared 0.024).

All 30 written responses came from C177, the comment field on the AI tools. Twenty were negative and ten positive, and eleven negative responses carried the strongest score of −10. Respondents described the same volume of work costing different amounts from one run to the next, with no explanation and no estimate before the run. No written outcome test reached significance.

Ratings of 3 or below, the distressed range, accounted for 249 of the 795 answers (31.32%). The engagement mean fell in the distressed range in 32 of 150 engagements.

The primary responses numbered 30, all written in answer to C177, “Add any comments on the agentic AI tools that the ratings in this section do not capture.” No response from any other question addressed the indicator, so every response was primary.

Fifteen responses described the same volume of work costing different amounts from one run, day or week to the next. Four gave a swing of three times: the same query answered twice, the same lead volume on a news-heavy day, a job cheap on Monday and three times heavier on Thursday. Others described spend doubling from one week to the next and consumption differing by a third between two runs of the same deal. Three responses attributed the cost to the same job running or looping several times, in one case a patch check run four times.

Ten responses described cost that could not be known before a run, so that a month could not be planned and one long pitchbook could burn more than a whole day's queue unseen until month end. Four said the respondent could not explain the movement to their own management or director, and one had asked twice without receiving an answer. Respondents asked for a cost estimate before each run, token use visible per job, and a pinned model version with notice of updates.

The ten positive responses described cost as steady or predictable enough, one within about ten percent over a month of provisioning volume, and placed their concerns in consistency of answers, sourcing, logs and confident errors.

Of the 30 responses, 20 (66.67%, 95% CI 48.28% to 82.76%) were negative and 11 (36.67%) carried the strongest negative score of −10. Positive responses numbered 10 (33.33%), and no response was neutral. The mean written score was −3.3 (95% CI −5.6 to −0.9, n = 30 responses), and the median was −5.0.

Of the 30 responses, 30 came from respondents who also rated the indicator. Among them, the written score did not track the rating (Spearman rho 0.32, 95% CI −0.08 to 0.64, p = 0.082, n = 30 responses). Of the 20 negative responses with a rating, 7 (35.00%) came from respondents whose rating was in the distressed range.

Respondents who wrote about the indicator and gave a rating numbered 30, with a mean rating of 4.6. The 765 respondents who rated the indicator without writing about it had a mean of 4.6. The difference, 0.0 (95% CI −0.8 to 0.8), included zero. Ratings in the distressed range made up 26.67% of the ratings from respondents who wrote (8 of 30) and 31.50% of the ratings from respondents who did not (241 of 765).

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 43.47, Holm p < 0.001, omega squared 0.407). Active engagements in Build-Out rated highest (mean 5.5, n = 160 respondents) and canceled engagements lowest (mean 3.2, n = 305 respondents). Ratings also differed by respondent role (Welch F 7.30, Holm p = 0.007, omega squared 0.014). Portfolio directors rated highest (mean 5.6, n = 27 respondents) and individual contributors lowest (mean 4.5, n = 617 respondents). Ratings also differed by vendor type (Welch F 4.68, Holm p = 0.003, omega squared 0.024). Consulting firms rated highest (mean 5.4, n = 112 respondents) and BPO specialists lowest (mean 4.2, n = 185 respondents). Outsourcing model, institution type, process family and class of work showed no difference.

Written sentiment did not differ by engagement status and lifecycle stage across the 27 engagements with a written response (Kruskal-Wallis H 0.68, df 3, p = 0.878). Fewer than two groups held 10 or more responses, so the groups are reported as counts only: active engagements in Build-Out (8), active engagements in Transition/Cut-Over (8), active engagements in Production (6) and canceled engagements (8).

No test across respondent roles was run, because fewer than two roles had 30 or more written responses. Individual contributors wrote 17, engagement managers wrote 11 and portfolio directors wrote 2 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 = −0.65, 95% CI −0.73 to −0.54, 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 = 0.62, 95% CI 0.48 to 0.73, Holm p < 0.001, n = 94 engagements). The association held after Holm correction in ten 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 0.81 to 0.89, 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 = −0.64, 95% CI −0.68 to −0.60, Holm p < 0.001, n = 749 respondents). The association held after Holm correction in thirty-four of the 37 segment levels tested.

Two comparisons set the engagements with at least one negative response against all other engagements. Of the 132 other engagements, 123 had no written response addressing the indicator. These two comparisons therefore do not isolate written sentiment. Of the 18 engagements with a negative response, 5 (27.78%) were canceled, against 51 of the 132 other engagements (38.64%). The difference did not reach significance (Fisher exact p = 0.444). Customer satisfaction among active engagements did not differ between the two groups (mean 66.77% on 13 engagements against 63.67% on 81, Welch t 0.79, p = 0.442).

Two correlations test written sentiment directly. Across the 27 engagements with a written response, the engagement mean written score was not associated with cancellation (point-biserial r 0.07, p = 0.723). Across 29 respondents, the respondent mean written score was not associated with intent to leave (Spearman rho 0.04, p = 0.853). Of the four 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 = 0.951).

Answers clustered within engagements (ICC 0.60). Clustering reduced the 795 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 0.86 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, 66.67%, ranked 63rd from the highest, against a median of 80.82%. Its mean written score, −3.3, ranked 56th from the most negative, against a median of −4.7.

The indicator most often present in the same responses was C157, “Rate the factual reliability of the AI tools' outputs on this engagement before any human review.”, in 6 of the 30 responses (20.00%). C165, “Rate how completely the tools captured material facts located anywhere in long documents, including mid-document content.” (6 responses), C171, “Rate the overall production readiness of the AI tools for bank-grade work as experienced on this engagement.” (5 responses), C175, “Rate the stability of the tools' behavior over the engagement when nothing changed on the engagement side.” (5 responses) and C176, “Rate the quality of the tools' logs, traces, and usage telemetry for detecting and diagnosing failures.” (5 responses) followed.

The most frequent theme was “Comparison to a prior engagement”, in 4 responses (13.33%). The second was “Respondent proposes a remedy”, in 3 responses.

About a third of all ratings of the predictability of token consumption fell in the distressed range (249 of 795 answers, 31.32%). 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 = 305 respondents).

The written evidence separates the cost of the tools from their predictability. Several respondents who wrote negatively about cost described the tools favorably in other respects, and their complaint was that the spend for a given volume could not be stated in advance or explained afterwards. For a financial institution that budgets by volume, this places the failure in planning and governance as much as in expense.

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 1 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 (30 responses on 27 engagements). Intervals for written sentiment therefore used a bootstrap by engagement.