Agentic AI outsourcing is already impacting T1 banks financially.

Ten primary financial-institution types appear, from asset management with 24 engagements to private equity with 9. All four primary outsourcing models are represented, with customer experience management (CXM) the largest at 28.00% of the study set. No institution type or outsourcing model dominates the study set, by design.

The mean engagement age is 5.69 months and the median is 5. Some 68.67% of the study set is six months old or newer, only 10 engagements have passed the one-year mark, and the longest runs to 22 months. Every figure in this report is an early reading of a young market.

Conversion to live operation remains limited. Only 30 engagements have reached production, and those 30 run a combined 181 agents; the remaining 120 divide evenly between planning and proof-of-concept. Production work consistently concentrates on high-volume processes whose outputs can be checked against a defined standard.

Cancellation has claimed 56 of the 150 engagements, a rate of 37.33% — and attrition falls unevenly across the lifecycle. Planning-stage engagements cancel at 48.33%, proof-of-concept at 28.33%, production at 33.33%. The heaviest losses come before the first milestone, where the concept first meets the client’s real data and systems.

Among the 30 engagements that reached production, 17 measured returns. Those returns average 24.71%, ranging from 6.00% to 47.00%, and all 17 remained active at the time of the study. Customer satisfaction splits to the extremes rather than clustering around the 43.85% mean: 56 engagements score below 40 on the 100-point scale, 59 score at 60 or above, and only 35 fall in the middle band.

Cancellation rates vary widely across the ten institution types, from 22.22% at private-equity firms to 54.17% at asset managers. None of that variation survives testing at these sample sizes: with 9 to 24 engagements per type, differences of this size arise by chance, and satisfaction behaves the same way. Problem composition, by contrast, holds nearly constant — client-preparation issues account for 38.00% to 45.00% of the prioritized problems across every institution type. The failure pattern is associated with the work itself, not with the type of institution buying it.

Operational scope shows a timing difference with no material performance difference. Institutions placed line-of-business operations into agentic outsourcing first; a corporate-functional wave followed roughly two months behind and has had less time to mature. The two waves are nearly equal in size — 74 corporate-functional engagements and 76 line-of-business — and line-of-business work averages 6.63 months of age against 4.73 for corporate-functional. Maturity alone separates them.

Cancellation rates are 35.14% for corporate-functional and 39.47% for line-of-business, a difference within sampling noise, and active-engagement satisfaction is flat at 64.35% and 63.83%. The ROI gap is mechanical: line-of-business work holds 21 of the 30 production engagements and 11 of the 17 reporting returns, a queue position rather than a quality signal. Mean scores across all five problem areas differ by just 0.34 points between the two.

Outsourcing-model selection is not a risk lever. The four models differ by 3.00 points on cancellation and 5.73 on active-engagement satisfaction. Pricing model and deal size behave the same way — neither the five pricing forms nor the annual-contract-value bands separate outcomes.

Engagements on 24-month terms cancel at 56.25%, against 28.00% for all other terms pooled. The difference survives correction for multiple comparisons at a corrected p of 0.010, and the available checks do not explain it away: 24-month engagements are younger than average, their composition resembles the rest of the study set, and survivor satisfaction is flat across every term. One reading is that a two-year term is long enough to judge an engagement and short enough to abandon it — one-year deals are treated as disposable pilots, while four- and five-year deals carry switching costs that keep them in place.

Vendor headcount across the study set peaked at 1,502 and now stands at 1,253, with planned headcount a year out of 856 — a forecast 31.68% decline for the full study set, almost all of it driven by canceled engagements winding down. The 94 active engagements plan growth of 11.90%, from 765 today to 856. Within the study set, a shrinking team signals an engagement approaching cancellation. Two boundaries apply: respondents staff the agentic delivery itself, so the record says nothing about employment inside the client institutions or the vendor, and canceled engagements still reported teams largely in place or in early transition.

The study identified material compliance, oversight, regulatory, and legal (CORAL) concerns. Individual contributors frequently lacked any awareness of CORAL topics, and respondents regularly noted the significant human effort needed to address preparation and agentic-AI challenges — decisions made and acted on by a junior team without a basic understanding of CORAL, most of it carried out without the bank’s involvement in the decision-making or basic awareness.

While the risk is material, the data indicated that CORAL concerns would not have materially altered the performance or final disposition of any engagement in the study. Given respondents’ lack of awareness and the young mean age of agentic AI outsourcing, however, it is highly likely that this topic is a greater concern than the data indicate. This report does not further develop the CORAL findings.