Source: https://akholi.com/outsourcing-2026-agentic-ai-tier-1-banks/introduction/executive-summary/
Author: Phil Hatch
Publisher: Akholi

Executive Summary

The 2026 Report · Introduction

# Executive Summary

Principal outcomes, manual alteration, governance, experience, and process redesign

By Phil Hatch · Akholi · Version 1.8.2

Complete study · PDFIntroduction · PDFStudy overview / Introduction / Executive Summary

The surveyed engagements had a high cancellation rate and low reported customer satisfaction. By the study date, 56 of 150 engagements had been canceled (37.33%). Delivery teams rated cancellation as at least moderately likely for another 58 of the 94 active engagements. Together, 114 of 150 engagements (76.00%) had either been canceled or met the study's threshold for perceived cancellation risk. Vendor-reported customer satisfaction averaged 43.85 points on a 100-point scale across the 150 engagements.

Positive measured returns were reported for 17 of the 30 engagements that had reached Production, with a mean recorded return of 24.71% among those 17. The mean is the arithmetic average of the recorded values. These are reported return figures, not independently verified financial results. Only 84 of 782 respondents (10.74%) rated their engagement as achieving its stated key performance indicators, with a rating of 3 or below on the statement that it was not doing so.

Four findings warrant particular attention:

- Manual alteration by the delivery team was reported on all 150 engagements when the five manual alteration methods were combined: manual alteration of financial-institution-provided data, manual alteration of financial institution processes, manual alteration of AI outputs, reporting alteration, and third-party work done by hand. Respondents described these interventions as necessary to sustain delivery. In the structured disclosure questions, discussion of data, process, or AI-output manual alteration with the financial institution was reported in 13 of 1,004 valid answers (1.29%).

- Respondents reported that financial institution governance, oversight, and control arrangements were not being used as designed. The 19 selected governance indicators from financial institution readiness and co-management had a mean rating of 3.63 on the 10-point scale (n = 15,201 ratings). Here n denotes the number of ratings. Every engagement had at least one respondent reporting that financial institution governance, oversight, and control arrangements were not implemented as designed. This selected subset is distinct from the 47-indicator governance, oversight, and control score examined in Part 3.

- Experience with agentic AI was limited. No respondent reported more than one year of personal agentic AI experience, and no respondent reported that the vendor had more than two years of agentic AI delivery experience. Vendor delivery-playbook maturity was distressed in 392 of 807 ratings (48.57%). Respondents also reported limited agentic AI knowledge among financial institution staff, with 418 of 793 ratings classified as distressed (52.71%).

- Respondents reported limits in the financial institution's willingness to redesign processes for agentic AI. Distressed ratings were given in 429 of 804 answers (53.36%).

The study organized 100 engagement indicators into five study-defined failure categories. Financial institution readiness had the lowest category mean (3.55), followed by co-management (3.65) and vendor capability (3.67). AI technology and the third-party ecosystem were rated more favorably, at 4.60 and 4.64, respectively, and both category means were below 5.0.

Figure 1. Mean indicator rating of the five failure categories on the 10-point scale, 150 engagements

Scroll horizontally to view the complete figure.

Financial institution readiness, co-management, and vendor capability received lower mean ratings

Exact chart data

| Series | Point index | Category | Value as stored |
| --- | --- | --- | --- |
| Mean indicator rating of the five failure categories | 0 | Financial institution readiness | 3.55 |
| Mean indicator rating of the five failure categories | 1 | Co-management | 3.65 |
| Mean indicator rating of the five failure categories | 2 | Vendor capability | 3.67 |
| Mean indicator rating of the five failure categories | 3 | AI technology | 4.5999999999999996 |
| Mean indicator rating of the five failure categories | 4 | Third-party ecosystem | 4.6399999999999997 |

Note: Each category mean averages the engagement-level means of its 20 indicators across 150 engagements. The amber bars identify organizational categories proposed as priorities for institutional review. The colors do not represent a measured degree of institutional control.

The study does not establish a regulatory or legal breach, and it contains no account from a financial institution. Its findings support the recommendation that financial institutions audit their agentic AI outsourcing engagements. Part 6 sets out the proposed response.

> The vendor's change log recorded release counts and cycle time but not which releases were later rolled back for validation failures, so the reported release velocity looked strong right up until the engagement was terminated.” INDIVIDUAL CONTRIBUTOR · ITO, Trade, settlement and reconciliation

← Introduction overviewStudy Design, Evidentiary Basis, and Definitions →