How AI Agent Sprawl Impacts Cost and Governance Across the Bank

Across the banking industry, AI adoption is expanding within individual lines of business, often through vendors selected to solve specific problems. Each vendor brings its own pricing model and reporting structure. Usage may be measured in tokens, API calls, processed documents, active seats or proprietary units that don’t align with standard financial reporting.

Taken together, these differences make it harder to form a clear view of overall spend. As a result, it becomes difficult to answer a basic question: What is our total AI spend, and what is it producing?

The information is available, but it is captured in separate systems, each using its own usage model, timing and reporting format. A commercial lending platform may charge per document processed, while an AML solution charges per alert triaged and a fraud platform charges per transaction scored. These models reflect how each system operates, but they do not translate easily into a unified view of cost or business outcome.

As a result, understanding how AI investments perform across the institution requires pulling these inputs together across systems. Each vendor is optimizing its own model, while the broader picture of cost, value and efficiency remains distributed across the bank.

Cost visibility is difficult to establish

This fragmentation makes evaluating AI spend at the institutional level more complex. Each vendor uses a pricing model aligned to how its system operates, which means forming a unified view of cost, value and efficiency across the bank requires a consistent way to compare those inputs.

Cost, however, is only part of the picture. Governance introduces a second layer of complexity that becomes more apparent as these systems scale.

Governance is hard to apply across independent AI systems

As these systems scale, governance becomes more complex to apply consistently across the institution. When an examiner asks how AI systems are governed, the response often points to existing model risk management frameworks.

Those frameworks are effective for internally developed models, but they are not always designed for a distributed set of third-party agents operating across different lines of business, each with its own data access patterns, update cycles and explainability standards.

In practice, this creates gaps in the application of oversight. Its vendor may update a fraud model while an AML model continues to run on an earlier version. A mortgage processing system may be fine-tuned on institution-specific data and then remain unchanged for months.

This variation makes it more difficult to track model behavior across systems, understand how decisions evolve and maintain a consistent audit trail across the agent fleet.

Gaps in visibility can lead to inconsistent outcomes

As systems operate independently, differences in data timing, model behavior and system context can lead to inconsistent outcomes across the institution. In some cases, similar transactions may be evaluated differently by separate systems because they are operating on different data snapshots or interpreting customer activity in different ways. These variations are not always visible at the point of decision, which makes them harder to identify and reconcile across systems.

Over time, this lack of alignment can affect reporting, compliance and control processes. When decisions are not consistently connected across lines of business, it becomes more difficult to demonstrate how outcomes were produced and how they align with institutional policies.

Regulatory expectations continue to evolve in this area. Financial institutions are expected to maintain clear oversight of how models are used, how decisions are made and how controls are applied across systems. As AI adoption grows, these expectations extend across a broader set of technologies and decision points.

A connected foundation brings cost, governance and decisions into alignment

As AI systems continue to expand across the bank, the way they connect and operate together becomes increasingly important. Cost visibility, governance and decision consistency are all shaped by whether these systems share a common foundation for how data is interpreted and how outcomes are produced.

Without that alignment, each additional system introduces more variation in how data is used, how decisions are made and how results are measured across the institution. With the right foundation in place, these systems can operate with shared context, enabling a clearer view of cost, more consistent governance and greater confidence in how decisions are produced across the bank.

As you evaluate AI across your organization, consider how cost visibility, governance and decision-making come together across systems. We can help you build the foundation to support that alignment. Get started →