6 minute read
The next wave of supervisory pressure isn't about what firms report. It's about whether they can prove how their systems produced it. This is what emerged from the RegTech Conference 2026: Mastering AI-driven solutions.
At the RegTech FS conference in London (26 March 2026), regulators, financial institutions, and technology providers gathered to examine how regulatory obligations are being translated into operational, auditable control frameworks. The discussions were case-led and regulator-informed, revealing a consistent pattern across the industry.
Most firms can produce the number. Very few can prove how their system produced it.
That gap between output and proof is not a reporting problem. It is a regulatory data architecture problem. And it is now the defining challenge for CTOs, CDOs, and product leaders at regulated financial institutions.
In this article, we explore the 3 dimensions of this challenge, drawing on insights from RegTech FS 2026 by RegRisk.
The core failure: data without proof
Ask a regulated financial institution a simple question: how was this number produced?
In most cases, the answer does not come from the system. It is rebuilt. Data is pulled from multiple sources, logic is retraced, assumptions are checked, and subject matter experts are brought in to explain what the architecture cannot.
That is not an edge case. It is the operating model across much of the industry.
Supervisors have noticed. The regulatory test has shifted from validating reported outputs to testing system behavior. Firms are now expected to demonstrate, on demand, how data flows, how decisions are made, how controls are executed, and how all of it can be evidenced.
The absence of this capability follows a predictable chain: fragmented data leads to inconsistent outputs, which drives duplicated controls and repeated remediation cycles. At scale, this translates into audit challenges, rising operational costs, regulatory findings, and ultimately capital consequences.
The industry has invested heavily in data platforms, pipelines, and reporting tools. But these improve access and speed. They do not create a governed, structured, explainable representation of data that can be used consistently across reporting, risk, and AI.
The missing layer is not more data. It is usable, defensible knowledge: data that is structured, connected, and capable of explaining itself under supervisory scrutiny.
This is why it keeps happening:
Why AI programs are stalling, and it's not the models
AI investment across financial services is accelerating. Confidence is not.
The pattern is now well-documented: AI programs move from pilot to proof of concept, then stall before reaching production. The most common diagnoses are model performance, integration complexity, or organizational readiness. But RegTech FS 2026 surfaced a more fundamental issue.
Most AI programs are failing because the data beneath them cannot support explainable decisions.
When supervisors test AI-driven systems, they are not asking whether the model works. They are asking whether it can be trusted. They want to know whether the firm can demonstrate what data the model used, how it was structured and interpreted, how decisions are derived, and whether outputs can be reproduced.
Most firms cannot answer these questions. Training data is unclear. Transformations are not traceable. Business context is missing from model outputs. And when scrutiny arrives, the response is the same: reconstruction.
AI is now inside the supervisory perimeter. If a firm cannot evidence how its AI behaves, that system will be constrained, challenged, or removed. This is not a model risk issue. It is a data architecture issue.
The firms seeing AI succeed are investing not in better LLMs but in better data foundations: semantic enrichment, consistent definitions, transparent lineage, and the ability to explain outputs in business and regulatory terms. Organizations with successful AI initiatives invest up to four times more in data quality, governance, and AI-ready foundations than those that fail.
The implication for CTOs and CDOs is direct: the constraint on AI deployment in regulated financial services is not technical capability. It is the absence of a regulatory data architecture that makes AI explainable, traceable, and defensible by design.
3 challenges firms are running into right now:
From data management to digital assurance
The challenges above, data without proof and AI without explainability, point to the same structural gap. The industry is scaling infrastructure without scaling control.
Firms are delivering faster pipelines, larger platforms, and broader AI capability. But without embedding control into the architecture, this creates fragmentation at scale: more data, more outputs, more complexity, but not more trust.
RegTech FS 2026 framed this as a decision point. Firms can continue managing data as an operational asset, or engineer it into the control fabric. The latter is what the conference called Digital Assurance: an operating model where data, control, and evidence are engineered into the architecture rather than bolted on afterward.
Digital Assurance requires 3 conditions to hold together.
-
Data pipelines must be provable, consistent, and reliable
-
Every decision must be explainable and traceable.
-
And processing must meet regulatory and jurisdictional constraints.
When these conditions are met, control becomes observable, repeatable, explainable, and testable under supervision.
Without this, firms remain dependent on reconstruction. And reconstruction is evidence that the system cannot explain itself.
For product managers building in regulated environments, this reframes the compliance conversation entirely.
Regulatory compliance is not a feature to be bolted onto a product roadmap. It is an architectural property of the data layer beneath the product.
When 50+ regulatory change drivers land on your roadmap in a single year, the question is not "how do we comply with each one?" It is "Does our data architecture absorb regulatory change, or does every change require a rebuild?"
What regulatory data architecture actually requires
Across all 3 dimensions (proof, AI governance, and digital assurance), the missing capability is the same. Firms need an architectural layer that transforms fragmented enterprise data into structured, governed knowledge that can be queried, explained, and defended.
This layer must:
- Unify structured and unstructured data into a single, consistent view across silos, including trades, transactions, disclosures, customer records, and operational data
- Apply semantic structure and shared definitions across domains, so data means the same thing regardless of where it sits
- Make lineage and transformations transparent, with every data point traceable from origin to output and every rule and transformation visible
- Connect data to regulatory obligations, not as a post-hoc mapping exercise, but as a persistent, queryable relationship built into the architecture
- Enable data to be searched, explained, and reused in business and regulatory terms, not just technical ones
This is the layer that most financial institutions do not have. And without it, neither proof, nor AI governance, nor digital assurance can be achieved at scale.
How Datavid builds this layer
Datavid operates at exactly this point in the architecture.
Using knowledge graphs, semantic structuring, and metadata-first governance, Datavid transforms disconnected enterprise data into the governed, searchable, AI-ready knowledge layer that regulatory data architecture demands.
For financial institutions, this means:
- Provable lineage on demand. Knowledge graphs connect fragmented data sources into a unified view with full traceability. When a supervisor asks "how was this produced?", the system answers without reconstruction.
- AI that can explain itself. Semantic enrichment applies domain-specific ontologies and controlled vocabularies to raw data, creating the contextual layer that makes AI outputs explainable in regulatory terms. Clean, metadata-rich pipelines feed GenAI, RAG systems, and predictive models with full auditability.
- Control embedded in the architecture. FAIR-aligned metadata governance enables regulatory mapping, multi-jurisdictional reporting, and continuous compliance evidence from a single platform. Supervisory exams become a query against existing architecture, not a crisis mobilization.
Datavid's accelerator platform, Rover, makes this transition practical, delivering production pilots in 6 to 8 weeks rather than requiring multi-year transformation programs.
"We waited more than 2 years for internal teams to create our compliance system.
Datavid, with Rover, did it in a few weeks."
- Data Officer, Roche
The supervisory environment described at RegTech FS 2026 is not a future scenario. It is the current reality. Firms that cannot prove how their systems behave, cannot explain their AI, and cannot evidence control continuously will face capital consequences, operational constraints, and sustained regulatory scrutiny.
The question is not whether to invest in regulatory data architecture. It is whether your current architecture can pass the test that supervisors are already applying.
Can your data withstand regulatory scrutiny?
.jpeg?width=40&height=40&name=Avinash%20Dixit%20datavid%20(Custom).jpeg){kind=small}
