A Practical Path to Govern Data, Stream in Real Time, and Modernize Beside the Core
Markets run on two clocks. One measures reliability: the systems that clear, reconcile, and settle every day. The other measures change: the pressure to ship new technologies such as AI that can spot risk sooner and automate the dull work. Those clocks don’t tick at the same speed. The banks that are moving fast with AI aren’t tearing out the past; they’re layering governed, real-time data on top of it, so models have something trustworthy to stand on.
History backs that up. Progress came from abstraction that made old systems useful in new ways: Client–server in the 1990s scaled front ends; Java’s virtual machine simplified deployment across mixed hardware; FIX standardized real-time trading flows starting in 1992; and web GUIs moved controlled functionality closer to users. Each wave solved a real problem by layering on top of what came before, a useful reminder now that AI is moving from pilot to production.
The Real AI Prerequisite: Governed, Reliable, Current Data
If AI is going to help with risk detection, control-room visibility, and operational triage, it has to draw from data that is auditable, governed, and fresh—not stitched together after the fact. Analyst estimates cited by IBM indicate that 70% of global transaction value still runs on IBM Z, which is why bringing governed AI to the mainframe matters. IBM also reports z16 can score up to 300B AI inferences per day, making real-time risk scoring and AML checks feasible where the data already lives.
Regulators are converging on the same point. The Basel Committee on Banking Supervision has highlighted the need for strong governance and supervisory cooperation around AI in banking. Meanwhile, the EU’s AI Act phases in obligations for general-purpose and high-risk AI starting in 2025 and tightening over the following years, timelines that make “governed first, AI second” more than a slogan.
Progress While Everything Stays Online
Banks don’t get maintenance windows for modernization. The workable path isn’t a wholesale rebuild; it’s evolving next to the core. Insulate new work from brittle interfaces, prove value where users feel it, and unwind the old in contained pieces once the new route is carrying real traffic.
Start at the edges where latency and clarity matter. Put a governed, real-time layer in front of the systems you have so trading, risk, and operations are looking at the same live publications with history attached. When teams can spot exceptions as they emerge, and resolve them without swivel-chairing across stacks, you earn the right to tackle the heavier restructuring behind the scenes.
Treat each success as fuel. Retire a bounded workflow, a desk, or a region; redirect the released budget and specialist time into the next slice. The cadence becomes repeatable: shield, show, then shed. The U.S. move to a T+1 settlement cycle in May 2024, and other markets exploring same-day options, only tightens tolerance for data gaps and delays.
That discipline comes down to three moves:
- Decouple new work from old interfaces.
Insert a high-performance access layer so teams publish consistent, entitlement-aware endpoints, and get a real-time source of truth, without touching mainframe contracts or bespoke middleware every time they ship a feature. - Replace in bounded slices.
Once users are on the modern path, decommission legacy by workflow, desk, product, or region. Each slice you retire trims “run the bank” spend and frees specialist cycles for the next wave. - Instrument by design.
Treat observability, lineage, and access control as part of the build. Track where numbers came from and how quickly they propagate so today’s changes don’t create tomorrow’s fragility.
Using Streaming and Virtualization to Modernize in Motion
Streaming and data virtualization reset the math by letting firms show value at the edges of the stack while the core is upgraded step by step. A governed, real-time layer means front office, risk, and operations all see the same live feeds and history; exceptions are flagged as they occur; derived measures update alongside the stream; and audit trails stay clean. That lets teams deliver “last-mile” improvements while the deeper foundation is refitted safely underneath.
Virtualization and a governed abstraction layer avoid hardwired point-to-point integrations that turn into expensive knots later. They shorten build time, keep interfaces consistent, and reduce the number of systems every change has to touch.
Build and Run Together
Teams move fastest when data access, processing, storage, and visualization live in a single, integrated environment so iteration doesn’t hardwire business logic into brittle connectors.
Think of it as an application engine for real-time finance: a place to connect sources, compose logic, design role-aware views, and publish safely with entitlements and lineage enforced throughout. The payoff shows up as lower latency, fewer manual reconciliations, quicker time-to-change, fewer frameworks to stitch together, and less custom glue code to maintain.
When front office, risk, and compliance share one governed source of truth, decisions and audit trails line up. Interfaces stay responsive under stress, with views updating in step with the stream so teams see risk in time instead of reconstructing it later. Poor data quality can cost firms millions annually in remediation, overtime, and mispriced risk; governance built into the workflow is cheaper than governance applied after the fact.
AI on Top of Governance, Not Instead of It
Once the foundation is in place (real-time, auditable data; resilient systems; clear entitlements) AI becomes the next layer. Models add value because they draw from governed, current information, whether for alert triage, operational summarization, or anomaly detection. This allows AI to become practical, safe, and explainable. Skip that step and you risk automating fragility.
Firms that can show clear lineage, entitlement checks, and timely data flows will also have an easier time explaining model outputs when they’re challenged.
How to Start Now
Pick a desk-level workflow where real-time visibility will pay immediate dividends, such as intraday exposure, reconciliations, or post-trade exceptions. Stand up the shield layer, define publications and entitlements, stream the data, and present a shared live view. Measure what improves: time to detect, reconcile, and remediate.
When the new path is stable, retire a bounded slice of the old and reinvest the freed budget in the next priority. Repeat that cycle. The aim isn’t to erase the past; it’s to build on it, layer by layer, so AI has something trustworthy to stand on and teams stop paying a premium to keep fragile systems wired together.
