Reported delivery difficulties, recurring manual alteration, and limited visibility into delivery work connect the engagement outcomes, governance concerns, and workforce experience examined in this study. Delivery teams described work sustained through human intervention alongside weaknesses in preparation and control. Their accounts require the reported performance of agentic AI outsourcing to be considered together with how that performance was achieved and what the two firms could observe. The report presents the structured results and written accounts from engagements at Tier 1 financial institutions, including null and conflicting findings.
The evidence is the reported experience of 850 members of vendor delivery teams on 150 agentic AI outsourcing engagements serving Tier 1 financial institutions, collected once, between January and June 2026.
The report is organized into six Parts.
01 Engagement Outcomes and Their Correlates
Engagement outcomes on the four core study measures, with mixed evidence on their relationships to structural attributes and careful distinction among supported, equivalent, and inconclusive results.
02 The Five Failure Categories
Financial institution readiness, co-management, vendor capability, AI technology, and the third-party ecosystem.
03 Governance, Oversight, and Control
Reported weaknesses in governance, oversight, and control, their relationship to engagement performance, and the operational significance of authority, review, and visibility.
The delivery workforce's experience, working conditions, intent to leave, and the continuity of knowledge used during delivery.
05 Manual Alteration by the Delivery Team
Manual alteration by the delivery team, the reasons given for it and its disclosure to the financial institution.
Seven proposed controls and a phased implementation schedule for assessing engagements, correcting identified gaps, and verifying that controls operate effectively.
The definitions, the indicator inventory, the method, the list of companion analyses, and an index.
I feel we are being judged on something nobody in this market has finished learning yet. Our people can read a medical file and see what matters in it, that part is solid, and the staff they put on this desk inside three weeks is more than I expect. But the agent side we are building while we fly. There is nobody in the firm I can ring at ten at night when the summariser produces a different risk note on the same chart, and the QA method we use is still the one written for human summarisers. So I score those lower. It is not that leadership is hiding from it. But wanting the capability and having it are two different things, and right now we are somewhere in between.”
ENGAGEMENT MANAGER · KPO, Credit, underwriting and suitability