by Ryan Donom
Nearly every financial institution already has decided that artificial intelligence matters. In KPMG’s research, 81% of banking and insurance CEOs name generative AI a top investment priority. The harder truth is that conviction has not translated into results. Accenture finds that only about 8% of companies globally qualify as “front-runners” — organizations that have scaled AI effectively and embedded it into core strategy. The gap between ambition and impact is the real story.
That gap is why the question we hear most often — “Should we become an AI-native company or simply an AI-enabled one?” — is the wrong place to start. For an established institution with legacy cores, regulators, and a book of business, a ground-up native rebuild isn’t realistically on the table. The useful question is narrower and more actionable: for a given use case, how far down the AI-native path does the return actually justify going? At evolv, we see the firms that win let ROI — not ideology — set that depth, one use case at a time.
Defining the Choice: Enabled vs. Native
“AI-enabled” means layering AI onto processes and systems that already exist: a copilot for advisors, fraud scoring bolted onto the legacy core, automated document review. AI is a feature that improves what you already do. “AI-native” means designing the workflow, data architecture, and sometimes the product itself on the assumption that AI is the default actor — autonomous servicing, AI-first underwriting, model-driven product design. AI becomes the operating model rather than an add-on.
For incumbents, this is a spectrum, not a switch. Most firms cannot — and should not — rebuild their system of record, but they can push specific, high-value workflows much further toward native. The starting infrastructure is sobering: only 25% of banks have enterprise-wide cloud or hybrid platforms supporting data-driven services, and just 18% use AI as a core driver of product and service development. You cannot go native on a foundation that isn’t there yet.
Where AI-enabled is the Right Call
Augmentation is the right answer for regulated, human-in-the-loop work: lending decisions that require sign-off, compliance review, fraud and AML detection, document processing, and advisor productivity. The trajectory is clear in wealth management, where Cerulli reports 42% of bank advisors already use AI in their practice, a figure projected to reach 77% within two years — overwhelmingly as augmentation, not autonomy.
The advantages are speed and safety: faster time-to-value, minimal disruption, auditability, compatibility with the existing core, and lower regulatory risk. The trade-off is that gains stay incremental. AI bolted onto a broken process inherits the broken process, technical debt accumulates, and differentiation is limited. The warning sign is everywhere: Forrester finds that while more than 70% of firms have generative or predictive AI in production, few measure its financial impact. “Enabled” can quietly become activity without return.
Where AI-native Earns its Cost
Going native makes sense in greenfield digital units, high-volume and low-touch segments exposed to fintech competition, and data-rich domains such as personalization and risk modeling. Accenture’s analysis shows AI value concentrates in five functions — customer servicing, lead origination, IT engineering, product development, and risk management — which together account for 59% of the total impact. Those are the candidates worth taking native first.
The upside is structural: a durable cost advantage, a defensible moat, and a magnet for technical talent. Accenture’s front-runners expect roughly 13% productivity gains and 12% revenue growth, and McKinsey sizes the banking-wide generative AI opportunity at $200–$340 billion annually, equal to 9–15% of operating profits. But the costs are real: heavy capital and time, intense model-risk and explainability scrutiny, governance immaturity, and cultural resistance. Cost is the single biggest barrier to deeper technology integration in wealth units, cited by 55% of executives. Client trust also caps how far autonomy can go — only about 30% of affluent investors currently use AI to inform decisions, and just 39% of investors say they are comfortable with AI in their provider relationship.
The Hybrid Reality — and Where Most Firms Get Stuck
In practice, the answer for nearly every institution is hybrid: native at the edges, enabled at the core. Rebuild a handful of high-ROI workflows toward native while keeping the system of record stable, and let the return on each use case set the depth of the rebuild. The data argues for exactly this discipline — only 34% of organizations have scaled AI for even a single core process. Spreading native ambition across everything is how firms end up scaling nothing.
The reason most firms stall, in our experience, is organizational rather than technical. Business and IT run separate backlogs — one framed in business outcomes, the other in technical enablement — and the two are rarely reconciled. The business asks for an outcome; IT prioritizes a different set of foundations; the use case stalls between them. evolv’s work is to draw a straight line from the business outcome, to the specific technology foundation required to support it, to a realistic ROI expectation benchmarked against what we see across the industry.
That line almost always runs through the same place: data governance and semantic modeling. The gaps there are the recurring root cause behind stalled pilots, and closing them is typically the unglamorous prerequisite to any use case — enabled or native — actually working. Forrester’s own roadmap points the same direction, predicting that AI will automate more than a third of manual processes such as data handling, reporting, and reconciliation, with agentic and explainable AI arriving over the medium term. None of that lands without a trustworthy data and semantic layer underneath it.
Conclusion: Sequence Beats Slogans
The wrong question is “are we native or enabled?” The right one is “which use cases justify which depth, in what sequence — and is our data foundation ready to support them?” The pragmatic path is to enable first, building data maturity and governance muscle, then go native precisely where the ROI justifies the cost and the trust exists to support it.
The institutions that pull ahead won’t be the ones with the boldest AI slogan. They’ll be the ones that reconciled the business and IT backlogs, set honest ROI expectations from cross-industry evidence, and shored up the governance and semantic foundation that determines their ceiling. That is the work evolv exists to do by leveraging Snowflake Cortex/Cowork or Claude Cowork — and it’s where the conversation with your team should start.
Further Reading
McKinsey — Capturing the full value of generative AI in banking
McKinsey — The State of AI 2025: rewiring to capture value
Forrester — The State of AI, 2025
Forrester — Predictions 2026: How Financial Services Can Thrive Amid AI Disruption
KPMG — AI Quarterly Pulse Survey / Financial Services Insights
KPMG — Intelligent Banking Report
Accenture — The Front-Runners’ Guide to Scaling AI
Accenture — Banking in the Age of Generative AI
Ryan Donom is a client partner and director at evolv Consulting.
