Ai In Finance

AI in Finance: Adoption, Barriers & Path to ROI

Skip to Main Menu

Skip to Main Content

Skip to Footer

Back to Blog

-->

Back to Blog

It’s nearly 5:00 pm on Friday. Your compliance team is staring at more than a thousand AML alerts, each needing to be cleared. Everyone knows most will prove false, yet every item must be reviewed. Down the hall, a client-onboarding manager is explaining to a potential depositor why a straightforward KYC check has stretched into a third week.

These are not theoretical frustrations. They are daily realities in financial services, where off-the-shelf AI tools often crumble under regulatory scrutiny, fragmented data, and the need for precision.

This piece takes a clear look at where AI adoption in finance really stands, why so many projects stall, and what it takes to move from pilots to production.

From pilots to production

Adoption is widespread. McKinsey reports that 71% of financial services firms are using generative AI in at least one function. But maturity is a different story. BCG found that 75% of firms remain stuck in pilots or proofs of concept, with most investments flowing into “basic” AI activities rather than transformative ones (BCG 2025).

That tension — enthusiasm vs. execution — is where many banks, insurers, and capital markets firms sit today. Models can impress in demos, but when exposed to production workloads and regulatory oversight, they often falter: fragmented data, opaque logic, weak governance.

Yet momentum is undeniable. Evident Insights reports that banks announced 117 new AI use cases between January and June 2025 — more than double the six months prior. Most are internal — fraud detection, compliance monitoring, onboarding efficiency — where risk can be contained while value is proven.

Leaders are showing what scale looks like. Bank of America invested for years before seeing returns. Today, its virtual assistant Erica handles more than 58 million interactions a month, with more than 90% of employees relying on it for routine tasks. The bank also deployed AI coding tools to 17,000 developers, cutting costs and accelerating delivery. These results show that domain-tuned, embedded AI can deliver measurable impact across both customer and employee experience.

The pattern is clear: adoption is growing fast, but real ROI only comes when AI is designed for transparency, control, and productivity.

Trust: the decisive barrier

Lack of trust in AI was identified by Deloitte as a key barrier to adoption in finance workflows. We see the same pattern with our prospects: concerns around data quality, validation, and oversight stall promising initiatives.

For firms operating under Basel III, SR 11-7, the EU AI Act, or SEC oversight, the ability to prove data lineage, enable human-in-the-loop controls, and generate auditable outputs is non-negotiable. Trust is not a soft issue. Without it, projects rarely progress past pilots.

Trust gaps exist because expectations of AI are often mismatched with its actual capabilities. Financial leaders hear promises of automation and efficiency, but in practice discover that models are powerful in narrow, well-defined contexts and brittle everywhere else. Making progress means being clear-eyed about where AI delivers value today and where it still needs human oversight, governance, and supporting systems.

What AI can and cannot do

Where AI adds value today:

Automates repetitive knowledge work and surfaces insights that speed decision-making

Supports compliance reviews, claims drafting, and customer service through copilots

Extends enterprise data value with retrieval-augmented generation (RAG), context engineering, and reasoning linked to trusted sources

Where AI falls short on its own:

Resolving fragmented, siloed, or sensitive data

Guaranteeing accuracy without validation—critical when errors create compliance or reputational risk

Embedding seamlessly into production workflows that must withstand regulatory workloads

Replacing governance, security, or auditability requirements

The lesson is clear. AI creates impact only when tied to trusted data, embedded in workflows, and governed by strong controls.

Crossing the chasm

Moving from pilots to production takes more than models. It requires orchestration, domain expertise, and governance.

That’s why building with a partner often makes the difference. A joint approach accelerates delivery with proven components and integration patterns, ensures governance and validation are built in from the start, and provides options for private deployment with structured knowledge transfer to your team. For institutions aiming for production-grade automation across multiple workflows, it’s the fastest and most reliable path forward.

This is where our approach fits. With Maestro , we help financial institutions design secure AI systems that plan, validate, and execute reliably in regulated workflows. With Jamba , we bring long-context models that handle complex retrieval and reasoning without breaking under data fragmentation. And because we’ve spent years implementing AI in regulated industries, we understand the operational realities — how to adapt systems to complex workflows, safeguard sensitive data, and make governance practical rather than burdensome.

Together, Maestro and Jamba form a foundation that is tailored to your domain, validated at every step, and deployable in your environment to keep sensitive data under your control.

For financial institutions, the next wave of AI is about earning trust while delivering measurable value. Those who get this right won’t just clear today’s backlogs, but be the first to realize the true ROI of AI.

Discover more

Jun 25, 2026

Token spend isn’t going down. You need more than naive routing to manage it

Labs in Front

-->

Jun 24, 2026

Tipping the scales: Merging weak agents into a state-of-the-art deep researcher

Labs in Front

-->

Jun 4, 2026

First scale, then enrich: How the right execution strategy helped us reach state-of-the-art on SWE-rebench