What is your role on this engagement?

All 850 respondents answered the question. Individual contributors numbered 667 (78.47%), engagement managers 150 (17.65%), portfolio directors 31 (3.65%) and customer management executives 2. Every engagement had exactly one engagement manager among its respondents and a majority of individual contributors. Individual Contributor was the assigned answer on every engagement, so no test across answer levels was computed at engagement level. The table of principal results therefore carries no engagement-level test.

At respondent level, with standard errors clustered by engagement, role was associated with the respondents' own ratings and with their intent to leave. The respondents' own engagement indicator mean rose from 3.9 among individual contributors to 4.3 among engagement managers and 5.1 among portfolio directors (p < 0.001). Mean intent to leave fell across the same roles, from 6.9 to 6.5 and 6.1 (p < 0.001). Cancellation and customer satisfaction, both properties of the engagement, did not differ by role.

The written responses addressing the indicator numbered 22, from 22 respondents on 21 engagements. Every response was secondary. Of the 22 responses, 7 were negative and 3 positive. Most described where a decision or a task sat in relation to the respondent's role, and the negative responses described decisions that had fallen to individual contributors because no one on either side would make them.

Each engagement was assigned the role given by the largest number of its respondents. Individual Contributor was that answer on all 150 engagements, and no engagement was tied. Engagement managers numbered 150 on 150 engagements, one on each. Portfolio directors answered on 31 engagements (20.67%) and customer management executives on two.

All 22 responses were secondary. Each answered a question outside the indicator's survey section and referred to the respondent's role in its content. They came from 8 questions outside the indicator's own survey section (most often C210 with 5 responses, C156 with 4 responses and C198 with 4 responses). Individual contributors wrote 18 of the responses and engagement managers four. Portfolio directors and customer management executives wrote none.

Six responses described a decision that fell to the respondent, above the respondent's role, because no one on either side would make it: who signs off when an agent records a purchase-order status wrongly, whether an income calculation was safe to pass, what went into the financial institution's file, and how customer-facing flows behaved. One respondent counted most of the sign-off points in a banking flow as not functioning, so that the decisions passed to whoever was closest to the problem. Another had become the person newer colleagues consulted instead of the financial institution, at a cost of about an hour a day. A further respondent, given the tooling and a short handover, was the only escalation path when an agent misfired overnight.

Eight responses said the matter asked about lay outside the respondent's role or view, often because a team lead handled it. Three of these, all to C198, said a rating of the third-party ecosystem recorded the respondent's visibility rather than the suppliers: one respondent who processed invoices was asked to rate suppliers never encountered, and another said half the questions were for someone who could see the contracts. Three responses described catching and correcting the output as the respondent's own job, and two described a move from hands-on correction to signing off on changes or reporting upward. The three positive responses, all to C030, named growth into a stronger role, trust with real response privileges and a manager who removed obstacles.

Of the 22 responses, 7 (31.82%) were negative and 3 (13.64%) positive. Neutral responses numbered 12 (54.55%), and no response carried the strongest negative score of −10. The mean written score was −0.9 (n = 22 responses on 21 engagements), and the median was 0.0.

Six of the seven negative responses described a decision that had fallen to the respondent. The remaining negative response described the respondent as the only escalation path when an agent misfired overnight. Most neutral responses said the matter lay outside the respondent's role or view.

Each share is computed on the respondents in that row. Rows with fewer than 10 respondents report counts only. Among the segments with 30 or more respondents, individual contributors made up between 71.79% and 82.46% of the respondents. The exception was the respondents whose vendor type was not reported and was recovered from the engagement: 56 of 57 (98.25%) were individual contributors.

Every engagement carries the same assigned answer, so the segment families recorded at engagement level could not be tested. Respondent role is the indicator itself. Vendor type, tested on respondents, showed no difference after Holm correction (chi-square 19.38, df = 15, Holm p = 0.187, Cramer's V 0.09, n = 850 respondents).

Responses from canceled engagements numbered 11 (mean −2.3, n = 11 responses), of which 5 were negative. Each active stage drew fewer than 10 responses and is reported as counts only. The test across the stages on engagement means (Kruskal-Wallis H 6.21, df 3, p = 0.102, n = 21 engagements) showed no difference that reached significance at 0.05.

Individual contributors wrote 18 responses (mean −0.8), of which 6 were negative. Engagement managers wrote four, reported as counts only. No test across respondent roles was run, because no role other than individual contributors drew 10 or more responses, and no role drew 30 or more.

Cancellation rate is the share of engagements canceled by the study date, measured at engagement level (150 engagements, 56 canceled).

Individual Contributor was the assigned answer on every engagement, so no test across answer levels was computed at engagement level. At respondent level, with standard errors clustered by engagement, cancellation did not differ by role (F = 0.95, df = 3 and 149, p = 0.419, n = 850 respondents). Individual contributors were on canceled engagements in 256 of 667 (38.38%) cases, engagement managers in 56 of 150 (37.33%) and portfolio directors in 9 of 31 (29.03%).

Customer satisfaction is the percentage recorded against each engagement. The engagement-level estimate uses the 94 active engagements.

Individual Contributor was the assigned answer on every engagement, so no test across answer levels was computed at engagement level. At respondent level, with each respondent carrying the customer satisfaction of the engagement and standard errors clustered by engagement, customer satisfaction did not differ by role (F = 0.65, df = 3 and 149, p = 0.583, n = 850 respondents).

The engagement indicator mean is the mean of an engagement's ratings across the rated indicators. At respondent level the measure is the respondent's own mean rating across the rated indicators. The role is not a rated indicator and does not enter either mean.

Individual Contributor was the assigned answer on every engagement, so no test across answer levels was computed at engagement level. At respondent level, with standard errors clustered by engagement, the respondents' own mean differed by role (F = 82.55, df = 3 and 149, p < 0.001, n = 850 respondents). It was 3.9 among individual contributors, 4.3 among engagement managers and 5.1 among portfolio directors. The mean share of an individual contributor's own ratings in the distressed range was 45.13%, against 36.81% for engagement managers and 20.36% for portfolio directors. Without the two customer management executives, a level of its own in this test, the difference held (F = 24.18, df = 2 and 149, p < 0.001, n = 848 respondents).

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 by up to 798 respondents.

Individual Contributor was the assigned answer on every engagement, so no test across answer levels was computed at engagement level. At respondent level, with standard errors clustered by engagement, intent to leave differed by role (F = 126.25, df = 3 and 149, p < 0.001, n = 798 respondents). Mean intent to leave was 6.9 among individual contributors, 6.5 among engagement managers and 6.1 among portfolio directors. Without the two customer management executives, the difference held (F = 6.49, df = 2 and 149, p = 0.002, n = 797 respondents).

The comparisons set engagements with a negative written response against all other engagements, most of which had no written response addressing the indicator. Cancellation affected 5 of the 7 engagements with a negative response, against 51 of the 143 other engagements (35.66%), a difference that did not reach significance at 0.05 (Fisher exact p = 0.103). The customer satisfaction comparison was not computed: only 2 active engagements had a negative response, fewer than 10. The engagement mean written score against cancellation (r = −0.42, p = 0.060, n = 21 engagements) did not reach significance at 0.05.

Across respondents, the respondent mean written score against intent to leave (rho = −0.52, p = 0.017, n = 21 respondents) reached significance at 0.05: respondents whose written responses were more negative reported higher intent to leave. Of the three comparisons computed, one reached significance at 0.05, the correlation of the respondent mean written score with intent to leave.

No association with a study measure was tested at engagement level, so no association was ranked or compared. At respondent level, the engagement indicator mean and intent to leave differed by role, and cancellation and customer satisfaction did not.

Role did not cluster within engagements (ICC −0.18). Each engagement had one engagement manager and several individual contributors, so respondents on the same engagement gave different answers by the structure of the team. The design effect was 1.00, and the effective n equaled the 850 answers. The respondent-level tests still used standard errors clustered by engagement.

No equivalence test was computed, because no association with a study measure was tested at engagement level. The absence of a test is not evidence of no association.

No primary test was computed, so no check is scored as holding or failing. The checks that compare engagement-level tests were not computed. The respondent-level tests found differences by role for the engagement indicator mean (p < 0.001) and intent to leave (p < 0.001) and none for cancellation (p = 0.419) or customer satisfaction (p = 0.583).

Each of the four measures was computed from data the indicator does not contain. At respondent level the engagement indicator mean is the respondent's own mean rating, and the role does not enter it. Each result is therefore a finding and not an arithmetic consequence.

A categorical indicator has no signed correlation with the study measures and therefore no profile against the study norm. The comparison by role appears in the four association sections.

The indicator, with 22 responses, stands outside the peer set of the 11 indicators of Part 4 with 30 or more written responses. Placed among them, its share of negative responses would rank 7 of 12 from the highest and its mean written score 5 of 12 from the most negative.

The indicator most often present in the same responses was “Rate the factual reliability of the AI tools' outputs on this engagement before any human review” (C157, 3 responses). “Rate how clearly ownership of each workflow step is assigned when agents, vendor staff, and customer staff work in the same process” (C138, 2 responses), “I have altered client-provided data” (C199, 2 responses), “I have altered client processes” (C205, 2 responses) and “I have altered AI outputs” (C211, 2 responses) shared second place.

The most frequent theme was “Occurs every cycle or most days”, in 4 responses, followed by “Insufficient exposure to rate” in three, all responses to C198.

Every respondent answered the question, and the structure of the teams fixed the analysis. Each engagement had one engagement manager and a majority of individual contributors, so every engagement carries the same assigned answer and no engagement-level test could run. At respondent level, the respondents' own ratings rose and intent to leave fell with seniority. Role did not separate respondents on cancellation or customer satisfaction, both properties of the engagement rather than of the respondent.

The written responses place individual contributors at the point where unowned decisions land. Six of the seven negative responses described a decision that the respondent judged to sit above the respondent's role, made because no one on either side would make it. Other responses said the survey asked about matters outside the respondent's view, and three said a rating recorded the respondent's visibility rather than the suppliers. Written sentiment was associated with intent to leave and with none of the other measures computed.

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.

Role varies within every engagement, and the engagement-level design assigns each engagement one answer. The engagement-level tests that carry the other indicators could therefore not be computed, and the findings rest on the respondent-level tests.

The customer management executive level holds two respondents on two engagements, whose own means lay far from those of the other roles (mean intent to leave 3.0). The respondent-level tests include that level, so the size of their F statistics for the engagement indicator mean and intent to leave should not be read as a measure of effect.

The respondents' own mean records how each role rated the engagement. Senior roles may rate it from a different vantage, so the difference by role does not establish which rating is accurate.

The written responses number fewer than 30, so their shares and the written-sentiment tests rest on a small base. Shares of written sentiment are shares of responses and not of the study set of respondents. In the sample read by a second reader, the two readers agreed on the presence of the indicator in 2 of the 3 responses either reader marked, and on the score in 2 of the 2 responses both readers scored.