How financial services firms can win by moving at two AI speeds

Financial institutions need to balance productivity gains today with AI-led innovation that drives enterprise-scale value tomorrow.

In brief

• Incremental AI deployments have delivered efficiency gains, but most financial institutions have yet to achieve enterprise-scale impact.
• A two-speed approach pairs near-term productivity improvements with lighthouse AI deployments that unlock growth and risk reduction.
• An AI-led operating model integrates data, workflows, governance and security to enable faster, safer and better informed decisions.

As AI continues to reshape the financial services sector, senior executives and boards are pushing for enterprise-scale deployments that create exponential value, rather than focusing on narrow improvements that have been the norm to date. Many firms have automated manual work, increased productivity and reduced costs. They’ve launched training and upskilling programs focused on the use of new tools. Those are necessary steps — and it’s worth continuing along this path to harvest the gains — but most firms have yet to achieve large-scale adoption from these initial efforts.

IT and business leaders recognize that the value of incremental approaches ultimately plateaus, especially if localized improvements add to technical debt, reinforce existing silos and inhibit new ways of working. They also understand that stronger technical foundations, more extensive interconnectivity and deeper strategic coordination are necessary to fulfill more ambitious, enterprise-scale objectives.

For chief information officers (CIOs) at financial services firms, the imperative is to translate AI ambition into measurable outcomes across three value pools:

• Efficiency and human productivity — primarily through straight-through-processing and cycle-time compression
• Risk reduction and increased resilience via decreased fraud and credit losses and stronger protections against operational risk events
• Growth through faster innovation cycles, increased share-of-wallet, pricing precision, improved client experience

Envisioning the future: the AI-led operating model in action

An AI operating model defines how organizations structure and link cross-functional processes, deploy talent, engage partners, and use data and technology to embed AI directly into day-to-day decision-making and execution. Operating models can also specify responsibility for outcomes, establish accountability for investment decisions, determine risk and governance requirements, and promote cross-functional collaboration.

Refined operating models are necessary to harness the power of AI’s rapid advancements, enabling firms to reimagine how business gets done based on intelligent, integrated operations, rather than siloed functions and fragmented data environments. And they address many of the challenges firms face in seeking to drive higher AI ROI.

AI-native workflows

Rather than a side tool, AI links daily operations across the front, middle and back offices, enabling continuous human-machine collaboration. AI agents make decisions inside workflows (e.g., know your customer (KYC), anti-money laundering (AML), underwriting, treasury), eliminating manual decision bottlenecks and reducing cycle time from days to seconds. They also flag process exceptions, document interactions and handle routine communications. AI agents prepare relationship managers for conversations with commercial and corporate clients, delivering tailored insights and prompts for next-best actions. Underwriters and credit teams have ready access to more precise information about and clearer visibility into overall exposures and individual risks. Product and software teams use AI-native approaches and agents at every phase of the development cycle to create mature software products (including code, tests, documentation and infrastructure) in days, rather than months.

Data-driven decision engines

A unified fabric of live, dynamic and well-governed data, featuring transactional, market, behavioral and operational signals, feeds decision systems across treasury, risk, pricing, fraud and market-facing operations. This data mesh is layered with semantic models and vectorized knowledge that promote accuracy and trust. Emerging risks and opportunities can be identified sooner. More sophisticated scenario modeling can explore a broader range of outcomes and implications. Those capabilities result in faster, better informed and more confident decision-making about pricing, capital allocation and liquidity.

Automated operations

When standard, straight-through execution is the rule across functions, humans focus on those issues that require strategic judgment and big-picture analysis. AI agents will orchestrate and oversee the workflows for new accounts and onboarding, routine payments, claims investigation, reconciliations, regulatory reporting, documentation and other core processes. The benefits for customers include shorter cycle times, more accurate processing, fewer repeat requests for the same information and consistent outcomes across channels. Within IT operations, incident, problem and change management processes could all be further automated and streamlined.

Zero-trust, cyber-resilient platforms

AI gives bad actors new tools and expands attack surfaces, but it also supports new capabilities for security and resilience, including AI-driven red teaming, automated patching and threat containment, and secure-by-design model endpoints. Every data object and model action is continuously verified and logged. Continuous authentication, more rigorous access controls and automatic responses to anomalous behaviors and suspicious actions will become standard practices. Platforms and infrastructure will be designed for resilience, based on smarter network monitoring and response capabilities, with almost-instantaneous identification and isolation of attacks. Security becomes event-driven and dynamic, rather than static and reactive.

Always-compliant operations

Compliance is designed into the flow of everyday processes, continuously monitored and verifiable in real time, rather than managed as a separate oversight activity. Evidence is generated and documented by agents at the point of transaction, communication and decision-making, freeing risk and compliance pros to focus on adjusting controls and advising business leaders. Institutions benefit from reduced regulatory risk, while clients and customers enjoy faster processing, fewer delays and greater convenience.

Bringing the future forward: accelerating AI innovation for enterprise impact

Given that today’s incremental approaches to AI adoption will only take firms so far in realizing the vision, the most pressing question for IT leaders is how to shorten the path to industrialized, AI-led solutions at scale. At most firms, it’s not just a matter of allocating more resources.

Rather, the goal should be to establish a new operating model that drives greater collaboration among formerly independent IT roles and functions to create the intelligence networks that will enable the business to execute at “two speeds.” CIOs will orchestrate closer integration among all IT domains (technology, data, risk, cyber, operations) to create a unified execution fabric that connects data flows, model behavior, workflow automation and security controls within a single, unified system that supports every part of the business.

Institutions should pair a focused foundation with a small set of lighthouse deployments designed for breakthrough performance gains. Beyond proving value, such deployments typically harden controls and create reusable components, changing not only how work is performed, but how risk is created, detected and governed. An AI operating model should explicitly align to model risk management and technology risk expectations, with clear accountability across the first, second and third lines of defense.

So how can firms move forward? A practical path often involves the following steps:

These actions demonstrate that speed and safety are not trade-offs: the platform team can standardize guardrails (e.g., model gateway, policy enforcement, logging, evaluation and monitoring), while domain and product teams can deliver differentiated solutions within boundaries.

In modernizing the IT operating model, CIOs and other leaders should seek to replicate proven practices that early adopters have helped define:

• Assessing and tiering use cases based on materiality, customer impact and regulatory sensitivity
• Embracing policy-by-design so that rules for data classification, retention, privacy and permissible use are embedded in the development pipeline
• Setting specific human-in-the-loop thresholds for high-impact decisions (e.g., credit, AML/KYC disposition, pricing overrides)
• Segregating duties and approvals for agent actions that move money or change records
• Establishing end-to-end auditability (e.g., prompt and context logging, model and version traceability, decision rationale capture)
• Instilling third-party and IP risk controls for external models and tools
• Continuous monitoring for drift, bias, performance, and control effectiveness, with clear escalation paths
• Preparing the culture and adopting new ways of working to capitalize on exponentially faster timelines of AI-native product development lifecycles

Conclusion

Collectively, these actions can facilitate the development of a new operating model for the future of intelligent, autonomous financial businesses. They set the stage for new value creation at a large scale and accelerated pace, even as incremental gains proceed along the same lines. The ability to move at pace will only become more important as large language model providers and tech firms deliver ever more powerful capabilities that accelerate coding and development to unprecedented speed.

This isn’t simply about adding new tools and equipping the workforce to use them. Rather, it’s about using AI to fundamentally change how decisions are made, how work is executed and how risk is governed. As such, we believe it’s the way for firms to exponentially increase their returns of huge AI investments.

Matthew Mahaffy, Senior Manager, Technology Consulting, Ernst & Young LLP contributed to this article.