84% of Banks Claim Enterprise AI. Most Can't Show the Returns.

Eighty-four percent of banking executives say their institution has achieved enterprise-level AI deployment. A new report from banking software company nCino finds that most of them are measuring the wrong thing — and that legacy infrastructure is silently consuming the budget that should be funding real AI transformation.

Adoption Metrics Are Not the Same as ROI

The nCino report, published June 2 and covered by Fintech Global, surveyed banking leaders across retail, commercial, and investment banking segments. The headline — 84% claiming enterprise AI — looks impressive until you examine what these institutions are actually measuring.

The majority of respondents described AI progress in terms of adoption metrics: how many employees are using AI tools, how many workflows have been touched by automation, how many pilots have reached production status. What most are not tracking: measurable return on investment — revenue attributable to AI, costs reduced by a specific dollar amount, loan decisions improved by a verifiable margin.

This distinction matters. Adoption means the tools are running. ROI means the tools are working.

A bank can deploy AI-assisted loan underwriting across 500 branches and still have no clean answer to "did this reduce credit losses?" if it never established the baseline metrics needed to measure the before-and-after. Most banks have not established those baselines.

The Legacy Infrastructure Problem

The structural finding in the nCino report is where the money goes. Roughly 70% of bank IT spending is consumed by maintaining legacy systems — core banking platforms, data warehouses, and integration layers built decades ago that cannot natively support modern AI use cases.

That leaves 30% of the IT budget to fund everything new: cloud migration, fresh applications, security, and the AI initiatives executives are claiming as enterprise-scale. For most midsize institutions, that 30% is not sufficient to both run the infrastructure AI requires and build the measurement systems needed to prove it's working.

The consequence is a feedback loop. Banks deploy AI on top of fragmented data environments. The AI produces inconsistent outputs because the data feeding it is inconsistent. The institution cannot cleanly attribute business outcomes to the AI deployment. Without attribution, ROI measurement becomes guesswork. Without ROI data, the institution can't justify the infrastructure investment needed to fix the data environment. The loop closes on itself.

Three Components of the Execution Gap

The nCino report defines the "execution gap" as the distance between where bank leaders say they are and where actual infrastructure can support them being. The gap has three distinct components:

Data fragmentation. AI models require clean, unified data. Most banks have customer data distributed across dozens of systems built at different times by different vendors — core banking, CRM, loan origination, fraud, compliance. A loan officer's AI assistant pulling from five different databases with different schemas will produce inconsistent recommendations. The AI isn't broken; the data environment is.

Integration debt. Connecting modern AI tools to legacy core systems requires custom API layers that take months to build and create new failure points. Each integration is a cost center that doesn't appear in an AI deployment announcement but consumes budget and engineering time that could go toward measurement infrastructure.

Missing baselines. To calculate ROI, you need to know what performance looked like before the AI was deployed. Many banks lack clean baselines for the processes they are now automating — manual underwriting cycle times, false positive rates in fraud screening, loan officer throughput — making before-and-after comparison methodologically impossible.

The Competitive Risk

The danger is not abstract. Fintechs were built natively on cloud infrastructure with unified data models. They do not have this execution gap.

A fintech lending platform running AI underwriting on a clean data stack can process a loan application in minutes with consistent, auditable decision logic. A traditional bank running nominally the same use case — AI underwriting on a legacy core with a middleware layer on top — may produce the same headline capability claim but slower, less consistent outcomes that it cannot cleanly measure.

The 84% of banks claiming enterprise AI deployment are competing against fintechs that didn't have to claim it because they were built that way.

What to Watch

The shift from adoption metrics to ROI metrics will accelerate when regulators demand it. Watch for OCC and Federal Reserve examiners to begin asking specifically about AI return on investment — not just deployment rates — in examination cycles over the next 12 to 18 months. That regulatory pressure, when it arrives, will force banks to build the measurement infrastructure they currently lack.

The institutions building it proactively now will have a head start. The ones waiting for the examiner's question will be explaining a gap rather than showing results.

By Hector Herrera