Banks have no shortage of AI experiments. Their harder problem is connecting models to fragmented customer, product, transaction, and compliance information without recreating integration and governance for every use case.
Plumery’s AI Fabric is designed as a standardised layer between generative AI and core banking services. It uses an event-driven, API-first data-mesh architecture intended to expose reusable, domain-oriented streams while preserving security and control.
The product enters a market where fewer than a quarter of banks believe they are ready for AI at scale, according to Boston Consulting Group research cited in the source. Infrastructure, not model availability, is becoming the dividing line between a demonstration and production.
Fragmented Data Makes Every Pilot Start Again
Legacy core systems often sit behind newer mobile and digital channels, creating separate views of customers and products. A new AI initiative must connect those silos, complete security reviews, and establish decision tracing before it can deliver a useful answer.
That duplication raises cost and slows progress. It also makes explanations difficult in regulated areas such as credit and anti-money-laundering, where a bank must reconstruct which information influenced an outcome regardless of who developed the model.
Plumery proposes governed data streams that can serve multiple applications. Separating systems of record from systems of engagement and intelligence allows the core to remain stable while approved services consume consistent events through controlled interfaces.
The domain boundaries are crucial. A “customer” event, for example, must have a defined owner, schema, quality threshold, and permitted uses. Without that discipline, a data mesh can become another layer of inconsistent feeds carrying old ambiguity into faster applications.
Reusable Architecture Supports Incremental Modernisation
Banks already use machine learning for fraud detection, portfolio monitoring, credit risk, and customer service. Citibank has deployed chatbots for routine enquiries, while Santander has described predictive models supporting credit-risk management.
These mature applications depend on timely, high-quality information. Smaller institutions may struggle because each connection requires scarce engineering and compliance capacity. A shared fabric could make another use case less expensive after the first foundations are established.
Plumery’s partnership with open-banking provider Ozone API reflects a broader composable approach. Platforms such as Plumery and Backbase position themselves as orchestration layers around existing cores rather than demanding an immediate replacement.
Incremental change lowers migration risk, but it requires a destination architecture. Otherwise, banks may accumulate a modern engagement layer while the same manual reconciliation and duplicated logic remain underneath. Each new use should retire some bespoke integration or clarify a shared data product.
Standardisation Must Still Prove Governance in Practice
Plumery CEO Ben Goldin says banks want production benefits without sacrificing governance, security, or control. A data fabric can centralise policy, but institutions still need evidence that identity, consent, lineage, retention, and access restrictions follow each event.
Operational resilience matters as well. If many customer journeys depend on one integration layer, its outage becomes a concentration risk. Banks need monitoring, fallback behaviour, version controls, and clear responsibility between platform provider and institution.
Regulatory sandboxes in the UK can help test such architectures under supervision. The valuable result is not simply a successful chatbot, but a repeatable pattern showing that models can access banking data safely and that decisions remain auditable.
Whether AI Fabric becomes a sector standard is uncertain. Banks differ in core systems, risk appetite, and modernisation history. The strategic direction is clearer: bespoke pipes for every experiment will not support enterprise scale.
Operational AI in banking requires reusable data products and shared governance that survive changes in models. Plumery is betting that the most important AI product is not the assistant a customer sees, but the controlled fabric supplying it with trustworthy information.
