The 2026 Report

Agentic AI outsourcing is failing on the buyer’s side of the boundary.

The delivery-side record from 150 engagements at Tier 1 financial institutions — named, dated, and method-transparent. 850 delivery staff; 37.33% of engagements already canceled.

By Phil Hatch · Akholi · First Edition, July 2026 · DOI 10.6084/m9.figshare.33005630

Where failure concentrates

Part 2 →
The five problem areas, ranked by share of distressed ratingsExhibit
20%40%60% Bank-side (client) preparationBank–vendor co-managementVendor preparation & mgmtAI tools (technology)3rd-party ecosystem 3.523.653.664.584.62 51.48%49.26%48.22%32.14%31.74%
Source: Akholi Outsourcing 2026, Table 6 (n = 850). Bars show the share of ratings in the distressed band (rated 3 or below). The figure inside each bar is the mean condition score on a 1–10 scale, where 1 is worst and 10 is best; the institution- and relationship-side areas (amber) score materially worse than technology and the third-party ecosystem (grey).
Key findings

What the record settles.

The industry is new. Every participant is inventing the practice in live engagements. No respondent reported more than one year of experience with agentic AI, and no vendor had delivered the work for more than two years.

Failure is material now. Cancellation is 37.33% for a study set with a median engagement length of five months. Client preparation rated the weakest of all measured factors; vendor capability second weakest; co-management becomes the primary success factor once in production.

Reporting is more likely manipulated than accurate. The people who assemble client-facing reporting put the likelihood of manipulation at 6.26 of 10.

Oversight rests on a workforce that is leaving. The agents are overseen by a junior and dissatisfied workforce — job satisfaction 4.19 of 10 — most of whom intend to leave. What holds these engagements together is undocumented knowledge, and departures remove it.

In summary

The delivery-side record.

Between April and June 2026, 850 delivery staff at major outsourcing firms gave structured ratings and written testimony on 150 agentic AI outsourcing engagements at Tier 1 financial institutions, spanning ten institution types and four outsourcing models. 37.33% of engagements had already been canceled; realized cancellations and the high-risk band together account for half the study set. Customer satisfaction averages 43.85%. Outcomes track the conditions institutions build rather than the deals they sign — and the account is vendor-side testimony, uncorroborated by client records.

37.33%Of engagements already canceled (56 of 150), median age five months
850Delivery staff surveyed, April–June 2026
150Engagements at Tier 1 financial institutions
43.85%Mean customer satisfaction across the study set

The executive agenda

Part 5 →

The record supports six immediate executive actions — each a governance choice already within the institution’s authority, none requiring a technology decision or a regulator’s permission.

Action 1

Gate every agentic contract behind an institutional readiness standard

Readiness
Action 2

Build co-management discipline before signature, and improve it throughout

Co-management
Action 3

Audit the design and execution of vendor reporting, not its arithmetic

Reporting
Action 4

Restructure commercial terms around outcomes and exit

Terms
Action 5

Treat the human oversight layer as a control function

Oversight
Action 6

Listen directly to the vendor’s delivery team

Signal
Read the full report

The report, in eight parts.

Summary

Executive Summary

The sample, the four findings, and the six executive actions.

Read →
Part 1

The Market

Adoption is broad and young, and the losses are already realized.

Read →
Part 2

Why Engagements Fail

Five problem areas; the institutions on both sides come before the technology.

Read →
Part 3

The Integrity of Performance Information

Reporting the producers rate more likely manipulated than accurate.

Read →
Part 4

The Delivery Workforce

The oversight layer is junior, dissatisfied, and leaving.

Read →
Part 5

Conclusions & the Executive Agenda

Six actions, all within the institution’s own authority.

Read →
Method

Study Information & AI Disclosure

Design, independence, statistics, limits, and the AI-assistance disclosure in full.

Read →
Definitions

A fixed vocabulary.

The study set. The 150 agentic AI outsourcing engagements at Tier 1 financial institutions that this report draws on.

Active and canceled engagements. 94 engagements were still running on the study date; 56 had been canceled.

Lifecycle stage. Every engagement sits in planning, proof of concept, or production. 30 had reached production; the remaining 120 were divided between planning and proof-of-concept.

The 10-point scale. Respondents rated 100 engagement-condition questions from 1 to 10, where 1 is the worst condition a respondent can report and 10 is the best.

Distressed. Any condition rated 3 or below.

Overall condition. An engagement’s mean rating across all 100 conditions; ranked by it, the study set divides into a weakest, middle, and strongest third of 50 engagements each.

High risk. An active engagement with a team-assessed cancellation likelihood of 70% or higher. 19 engagements qualify; with the 56 canceled, they account for half the study set.

Customer satisfaction. The instrument’s satisfaction metric, with a study mean of 43.85%.

Return. A measured return on engagement, reported for 17 of the 30 production engagements, averaging 24.71%.

Report information

Publication. Outsourcing 2026: Agentic AI and Tier 1 Financial Institutions — The Delivery-Side Record from 150 Engagements. Akholi, First Edition, July 2026. Study period Q1–Q2 2026. DOI 10.6084/m9.figshare.33005630 · ORCID 0000-0003-1709-8860.

How to cite. Hatch, Phil. 2026. Outsourcing 2026: Agentic AI and Tier 1 Financial Institutions: The Delivery-Side Record from 150 Engagements. Akholi.

Rights. © 2026 Akholi. This report may be quoted and cited with attribution to the author and Akholi; reproduction or adaptation of substantial portions requires written permission (p.hatch@akholi.com).

Contact. Media and general inquiries: info@akholi.com. Permissions, citation, and author correspondence: p.hatch@akholi.com.