AI in financial services 2026: where banks and insurers are seeing returns

Executive summary

AI is now standard equipment in banks, insurers and payment firms, but evidence of financial return is still scarce. Most institutions can point to tools in use, far fewer can show the effect in their accounts, and the gap between the leaders and everyone else is widening. This report sets out what named institutions have disclosed, where AI is in production, what supervisors now expect, and what to do before the main regulatory deadlines in late 2027.

Adoption is close to universal; measured returns are rare. 75% of UK financial firms already used AI in 2024 [1], and 65% of financial institutions in NVIDIA’s 2026 survey are actively using it [2]. Yet Evident’s 2026 review of 50 large banks found only 12% of reported use cases show impact against operational KPIs, and barely 1% disclose concrete financial returns [3].
A small group is publishing real numbers. JPMorgan Chase said AI-driven benefits are heading towards $2 billion [4], DBS reported about SGD 1 billion in economic value from data and AI in 2025 [5], and Lloyds put generative AI value at around £50 million for 2025 [6].
Insurers and payment firms disclose even less. Only three of the 30 insurers in Evident’s 2026 insurance index quantify group-level AI returns [7]. No payments company in Evident’s 2026 payments index has disclosed realised or projected group ROI [8].
The main compliance deadlines now fall in late 2027. EU high-risk rules for credit scoring now apply from 2 December 2027 [9]. Singapore’s AI risk guidelines take effect on 7 October 2027 [10]. US banking agencies left generative and agentic AI outside their revised model risk guidance in April 2026 [11].
AI-enabled fraud is rising faster than defences. UK fraud cases reached 4.1 million in 2025, up 11% [12]. Deloitte projects US fraud losses enabled by generative AI could reach $40 billion by 2027 [13].
Agents are moving from pilots to early production. 42% of financial institutions are using or assessing agentic AI [2], and six of the 50 Evident-tracked banks announced their first agentic use case in the second quarter of 2026 alone [16].

Key findings

01Only 12% of reported bank AI use cases show impact against operational KPIs, and barely 1% disclose financial returns

02JPMorgan Chase raised its projection of AI-driven benefits from $1 billion to heading towards $2 billion

03Only three of 30 large insurers disclose group-level AI returns, led by Manulife at CA$300 million in fiscal 2025

Evidence of AI returns in 50 large banks

Show impact on operational KPIs

12%

Disclose financial returns

Source: Evident AI Index, October 2026

What leaders should do

Tie each AI use case to a line in the P&L or a regulatory outcome before funding it.
Bring model risk teams in at the design stage as well as for validation.
Prepare for agent-initiated payments and the authentication changes they require.

In the full report

Adoption is broad; evidence of value is thin
Who is reporting returns, and how
Use cases in production
Model risk and governance
Regulation and supervision
Deepfakes and AI-enabled fraud
What to do in 2027
Sources

Methodology

We reviewed public sources published between 2023 and October 2026, favouring material from 2025 and 2026. These include supervisory surveys and publications (Bank of England, FCA, PRA, MAS, FSB, BIS, IMF, EBA, FinCEN), benchmarks (Evident AI Indices), consultancy and vendor surveys (McKinsey, BCG, Deloitte, NVIDIA, IDC, Statista) and company disclosures. Each figure is cited to a dated source we opened. Where we relied on a secondary report of a primary document, the source list says so. Vendor and consultancy surveys are self-reported and should be read as indicators of sentiment, not audited results. We did not survey institutions directly.