The role of a shared knowledge layer across AI systems

In my last post, I described the agent sprawl pattern: five lines of business, five AI vendors and five sets of agents, each operating without a shared understanding of the institution’s data.

A common next step is to introduce orchestration, often in the form of a central agent that routes requests across systems. This can help coordinate activity across agents, but it does not address how those agents interpret data or how their decisions align.

A shared knowledge layer aligns data across systems and agents

What’s missing is a shared knowledge layer: a structured, governed semantic model of the institution’s data that sits beneath every agent, from every vendor, across every line of business.

This layer provides a consistent way to understand how data relates across systems. It defines how entities connect, how relationships are represented and how business context is applied, so that each system is working from the same underlying interpretation.

Without that shared structure, each agent relies on its own view of the data, shaped by how and when that data is accessed, transformed and interpreted within its environment. Over time, those differences lead to multiple versions of the same customer, each informing decisions in slightly different ways.

Ontology defines how data connects and what it means in context

An ontology enables this shared layer by defining how data connects without moving it. Loan data remains in the origination system, transaction data remains in the core and customer records remain in the CRM. The ontology maps the relationships between those systems and defines what each piece of data represents in context.

That context shapes how customer activity is interpreted across the bank. A customer is not just a record in a system, but a set of relationships that span products, accounts, transactions and prior activity across the institution. Without that broader view, each system reflects only a portion of the customer’s profile.

For example, the same individual may be a commercial borrower with a $4.2M CRE loan, a retail checking customer, a wealth management client with a trust account and the subject of resolved AML alerts from Q3 2024 that were closed as false positives.

This is one customer, one identity and one governed view. Each agent — whether for spreading, fraud detection, AML triage or mortgage processing — needs to reason from that same view.

Without shared context, the same activity is interpreted differently

When agents operate from different contexts, the same customer activity can lead to different interpretations. A $50,000 wire may appear anomalous when viewed only through retail transaction history, while in a commercial context, it reflects routine inventory purchases. When agents operate on different slices of the same customer’s activity, their decisions can diverge.

A shared knowledge layer brings those perspectives together

A shared knowledge layer brings those perspectives together. It aligns how entities, relationships and business rules are defined across the institution, so that agents operate with a consistent context while continuing to use data where it resides. The priority is ensuring that all systems operate from a shared understanding of the data they rely on.

This is a data architecture decision that shapes every AI system

This is fundamentally a data architecture decision. It shapes how your systems connect, how your data is interpreted and how decisions are produced across your organization. It also influences whether those decisions remain consistent as they move across lines of business, where each system contributes a different perspective on the same underlying data.

In many organizations, this layer is still undefined, in part because it sits between traditional areas of ownership. It spans data, applications and AI systems, which can make it difficult to assign to a single team or function.

Explore how Rackspace Technology helps define and implement the shared data and AI layers that connect systems, align decisions and support governance across your organization. Get started →