Financial Services Industry Forum

Firms seeking a deeper grounding in how these principles apply to their sector can explore Databricks for financial services, where the platform's approach to regulated-industry AI is laid out in full.

The core message was blunt. The model is not the moat. Durable advantage comes from the data layer beneath it, plus architectures and controls that let institutions ship fast where ready, and iterate deliberately where risk dictates a slower lane.

Build beyond the model: choice, context, and control

Leaders are avoiding lock‑in and keeping optionality across models and clouds, because the top model changes quickly and different tasks want different capabilities. Strong evaluation lets teams start with a best model to prove value, then optimize down to smaller, faster, cheaper models while maintaining quality. Capabilities like Mosaic AI help enforce that discipline across providers.

Real business value hinges on safe access to proprietary context. That means bringing institutional knowledge and semantics directly to AI, and doing so under tight governance to balance risk, compliance, and cost. Many firms are now operating two lanes in parallel: a slow lane for intentional testing, and a fast lane for high‑confidence projects engineered to scale globally.

Data modernization is the non‑negotiable prerequisite

Morgan Stanley and JPMorgan Chase framed clean, curated, well‑governed data as the gating factor for AI impact. Unifying data into high performance stores and building semantic layers are unlocking tangible wins, from cutting deposit processing times by 95 percent to shrinking counterparty risk calculations from days to hours. Two thirds of JPMorgan's 300,000 employees now use large language models daily, with governance and data products embedded directly into workflows via Unity Catalog.

Mastercard's lakehouse foundation unifies 175 billion annual transactions into an AI‑ready base. That base is enabling large tabular models for fraud that better separate genuine anomalies, such as once‑in‑a‑lifetime purchases, from fraud, improving approval rates and customer trust.

Governing agentic AI in a regulated reality

Agentic AI changes the governance paradigm. Agents require persistent identities, scoped entitlements, and real‑time controls, because after‑the‑fact checks are insufficient for non‑deterministic systems. With formal AI rulebooks still emerging, firms are building to regulator‑grade standards now: rigorous evaluations, extended parallel runs, comprehensive documentation, and full audit trails.

Cost discipline is integral. Long‑running agents can generate significant inference and token spend, so teams are matching the right model to each job and optimizing once value is proven, rather than defaulting to the frontier.

From productivity helpers to an AI‑native operating model

RBC Capital Markets illustrated what mature production looks like with Aiden, its platform serving 8,000 professionals. Analysts have asked 4.7 million questions and saved 1.7 million hours. Aiden Research compresses an hour of earnings analysis per company into minutes using agents attuned to each analyst's voice, live transcript catch‑up during calls, a virtual notetaker, and Excel‑native agents that update models with source citations. The outcome is measurable uplift, including industry recognition for research quality tied to this transformation.

Across the firms, the vision converges on an AI‑native operating model where humans focus on relationships, strategy, and judgment, while agents handle the structured, time‑consuming work.

What will separate tomorrow's leaders

Technology alone will not deliver returns. The institutions pulling ahead treat data as a strategic asset, keep architectures open and flexible, embed governance into real‑time operations, and invest in culture, education communities, and leadership support. Bringing AI into the tools where people already work is what turns pilots into compounding business impact.