Published in: Case Study

Case Study: How OakNorth Matches Its AI-Powered Credit Intelligence With AI-Powered UBO Resolution

Author Brown Olivia

Published on: February 16, 2026

Last updated: October 22, 2026

A bank famous for zero credit losses on £12.5 billion in lending needed UBO verification that could keep pace with its AI-driven underwriting.

The Challenge

OakNorth is not a typical bank. It lends to mid-market businesses — the “missing middle” with revenues between £1 million and £100 million — using a proprietary AI-powered Credit Intelligence platform that analyses borrower risk across 270 industry sub-sectors. The result: over £12.5 billion in cumulative lending with zero credit losses. Loans that traditional banks take months to approve, OakNorth closes in days.

But there is a tension. OakNorth’s credit engine moves fast. Its AML and KYB compliance process did not. Every loan still required the compliance team to identify Ultimate Beneficial Owners, trace ownership chains, screen against sanctions and PEP lists, and document the entire process for the FCA. For straightforward borrowers, this was manageable. For the complex ones — property developers with multi-layered holding structures, hospitality groups with cross-border ownership, family businesses with trust arrangements — manual UBO resolution could take longer than the credit analysis itself.

OakNorth’s competitive advantage is speed. A borrower who waits three days for a credit decision but then waits two weeks for compliance clearance is a borrower who considers other lenders. The bank needed UBO resolution that was as fast and intelligent as its credit engine.

This isn’t a problem the FCA is indifferent to. In April 2026 the regulator published findings from a multi-firm review of customer due diligence, enhanced due diligence and ongoing monitoring controls. Firms that scored well had documented, risk-based criteria for when EDD applied and kept an evidence trail of what was actually checked; firms that scored poorly had gaps between their stated policy and what the file could prove happened. A lender closing commercial facilities in days, for borrowers whose ownership sits inside SPVs, trusts and offshore holding companies, is exactly the kind of firm that review was aimed at — the speed that wins the deal is the same speed that can outrun the paper trail a supervisor will later ask to see.

Why Credit Risk Models Need Accurate Ownership Data

Most explanations of UBO verification treat it as a compliance checkbox bolted onto lending: find the real owner, screen them, file the evidence, move on. That undersells what ownership data actually does inside a commercial credit process. For a lender like OakNorth, who owns and controls a borrower isn’t a side question to the credit decision — it’s an input to it, in at least three distinct ways.

It’s an entity resolution problem before it’s a risk problem

A credit model scores the borrowing entity. But the borrowing entity is rarely the same legal person whose financial history, trading record or connected-party exposure actually matters to the decision. A property SPV set up for a single development might have no trading history at all — its creditworthiness is really a function of the developer group standing behind it, the fund providing mezzanine capital, and the family office that ultimately controls both. Matching “Project Alpha Developments Ltd” in the loan application to the operating group, the connected SPVs from previous facilities, and the natural persons who control all of them is an entity resolution exercise: reconciling names, addresses and registration numbers that are spelled differently across registries, filed years apart, and sometimes deliberately obscured. Get the match wrong and the credit model is scoring the wrong risk — either understating exposure to a group that’s already stretched across several facilities, or overstating it by treating genuinely unconnected entities as one. For a deeper look at how multi-layered structures get mapped once the entities are correctly matched, see Multi-Layered Corporate Structures: A Risk-Scoring Framework for Escalation Decisions.

Ownership structure sets the actual exposure limit, not just the sanctions answer

UK banks don’t set lending limits per legal entity in isolation. Under the PRA’s Large Exposures framework, a bank must identify “groups of connected clients” — borrowers that have to be treated as a single risk because one controls the others, or because financial distress in one would be expected to spread to the rest. The PRA’s supervisory statement on identifying these groups (SS3/25, effective from 1 January 2026) sets out how firms should apply the control test and an economic-interdependency test when grouping clients for this purpose. In practice, that means the same ownership-chain tracing used to find a UBO for AML purposes is also the evidence a bank needs to know whether a new loan pushes it over its internal exposure limit to a group it already lends to under a different corporate name. Resolve the ownership chain once, correctly, and it answers both questions — compliance and capital.

Data aggregation standards expect the entity to be identified correctly in the first place

The Basel Committee’s risk data aggregation principles (BCBS 239) set an expectation that banks can produce accurate, complete risk data by legal entity, not just by product or portfolio. A 2018 ECB thematic review of banks’ implementation found this harder in practice than the principle implies, in large part because legal entity identification itself — who is this counterparty, and who stands behind it — is inconsistent across systems. A Legal Entity Identifier helps once it exists, but LEI coverage is thin outside large corporates; most of OakNorth’s mid-market borrowers, and nearly all of the SPVs and family holding vehicles layered above them, won’t have one. That gap is exactly where registry-sourced UBO resolution does work an LEI can’t: it supplies the entity identification that aggregation, exposure reporting and credit risk modelling all quietly assume is already correct.

Ongoing monitoring has to survive the life of the facility, not just the day it was signed

A UBO check performed at origination is a snapshot. A commercial loan runs for years. In that time, a borrower’s holding company can be sold, a trust can be restructured, a connected SPV can be struck off, or a shareholder can cross a control threshold that wasn’t crossed at drawdown. None of that shows up in a covenant compliance certificate, which tests financial ratios, not ownership. A bank that only re-checks UBOs on a fixed annual calendar can carry a facility for eleven months against a borrower whose actual controlling party changed in month two — the same blind spot that shows up in periodic-review-only compliance programs generally, covered in more detail in UBO Verification Workflow: From Initial Check to Ongoing Monitoring. For a lender whose entire pitch to borrowers is continuous, data-driven monitoring of their financial health, leaving ownership monitoring on a once-a-year cycle is the one piece of the risk picture still running on a different clock to everything else.

The same ownership question recurs at every stage of the credit lifecycle:

The Solution

OakNorth deployed Zavia.ai to bring AI-powered UBO resolution into alignment with its AI-powered credit process.

How Zavia.ai works differently for OakNorth:

• Speed-matched compliance: Zavia.ai resolves UBO structures in minutes, not days. For a bank that promises borrowers a credit decision in days, compliance can no longer be the bottleneck. The AI-powered ownership resolution runs in parallel with OakNorth’s credit analysis, so the compliance output is ready when the credit committee meets.
• Property development structure resolution: Property developers are OakNorth’s bread and butter — and the most structurally complex borrowers. A single development project might sit inside an SPV, owned by a holding company, with equity from a family trust and mezzanine from a fund. Zavia.ai’s AI maps these layered structures automatically, calculating effective ownership through every path to identify the natural persons who control the project. For how circular or looped ownership specifically gets detected within structures like this, see Circular Ownership Structures: How Companies Use Loops to Hide Beneficial Owners.
• FCA evidence standard: The FCA expects banks to trace UBOs to a natural person, use authoritative data sources, and maintain a complete evidence trail. Zavia.ai links every UBO determination directly to the originating government registry filing, providing the provenance trail that OakNorth’s regulatory supervisors require. When the FCA reviews a loan file, the ownership evidence is traceable to an official source — not a commercial database screenshot. That same resolved UBO is also what the sanctions and PEP screening runs against — see Sanctions Screening and Beneficial Ownership for why screening the entity alone misses a sanctioned owner sitting behind it.
• Groups-of-connected-clients mapping: Because Zavia.ai resolves control relationships, not just ownership percentages, the same output that identifies a UBO for AML purposes also flags when a new borrower shares a controlling party with an existing facility — the trigger for PRA large-exposure aggregation, surfaced automatically rather than relying on an analyst to recognise a name across unrelated loan files.
• Portfolio-level monitoring for covenant compliance: OakNorth monitors borrower health continuously using its Credit Intelligence platform. Zavia.ai adds an ownership dimension to that monitoring: if a borrower’s UBO changes, its parent company restructures, or a connected entity gets struck off, the alert feeds into OakNorth’s existing portfolio monitoring workflow alongside the financial signals.

Illustrative: Resolving a Property Developer’s Ownership Chain

The structure, names and figures below are an illustrative composite — built to show how the resolution mechanism works on the kind of layered borrower OakNorth lends to, not a disclosure of any specific client’s file.

1. Facility request comes in. A mid-market developer, “Thornfield Developments SPV Ltd” (illustrative), applies for a facility against a residential scheme. The credit engine begins scoring the project against its 270 industry sub-sector models.
2. Entity resolution runs in parallel. Zavia.ai matches the SPV’s registration details against UK Companies House and cross-references the registered office, directors and PSC filings against any entities already on OakNorth’s book — checking, first, whether this “new” borrower is secretly connected to an existing one.
3. Layer 1 — the SPV itself. Companies House PSC data shows a single corporate shareholder, Thornfield Holdings Ltd, at 100%. No natural person yet.
4. Layer 2 — the holding company. Thornfield Holdings Ltd’s own filing shows two shareholders: a UK operating group at 60%, and a Luxembourg-domiciled investment fund at 40% (the mezzanine provider).
5. Layer 3 — the operating group. The 60% UK stake traces to a family trust, with a named trustee and two beneficiaries. Trust deeds and the trustee’s declaration identify the settlor as the controlling natural person.
6. Layer 4 — the fund. The 40% Luxembourg fund stake is itself owned by multiple limited partners below the UBO threshold individually, with the fund’s general partner identified as exercising control for AML purposes rather than any single LP.
7. Effective ownership is calculated. Multiplying through the chain, the settlor’s effective interest in the SPV works out to roughly 41% (60% of the holding company’s UK stake, adjusted for the trust’s beneficial split) — comfortably over the 25% UBO threshold and the PRA’s control test, so they are confirmed as UBO.
8. Screening and output. The identified UBO is screened against sanctions and PEP lists, the fund’s general partner is checked separately, and the full chain — with every filing date and source registry — is packaged as the compliance output, ready for the credit committee alongside the risk score.

The Results

Why It Matters

OakNorth built its entire business on the principle that AI should handle the complex, data-intensive work so humans can focus on judgment. Its Credit Intelligence platform does that for credit risk. Zavia.ai does the same for ownership risk. The bank’s compliance analysts no longer spend hours tracing ownership chains through registries — they review AI-generated ownership maps and focus on the judgment calls: does this structure make sense, is the risk acceptable, what questions should we ask the borrower.

For a bank that has lent £12.5 billion without a single credit loss, the reputational cost of an AML compliance failure would be catastrophic. It also isn’t only a reputational question: under the FCA’s April 2026 findings, and the PRA’s refreshed SS3/25 expectations on connected-client grouping, both the AML evidence trail and the exposure-limit calculation now trace back to the same underlying ownership data. A gap in one is a gap in both. Zavia.ai ensures that OakNorth’s UBO process is as rigorous and defensible as its credit process, on both counts at once.

Bottom line:

Zavia.ai gives OakNorth AI-powered UBO resolution that matches the speed and rigor of its AI-powered credit engine — resolving complex property development structures in minutes, providing FCA-grade evidence trails, and integrating ownership monitoring with the bank’s existing portfolio intelligence.

Frequently Asked Questions

1. Why does a commercial lender need UBO resolution if it already runs AML checks on the borrowing entity?

Because the borrowing entity and the party actually bearing the risk are often not the same thing. A single-purpose SPV can pass an entity-level AML check with a clean record simply because it has no history — the risk sits with whoever controls it. UBO resolution is what connects the clean-looking shell to the real controlling party, their sanctions exposure, and any other facilities they already have on the book.

2. What is the UK’s ownership threshold for identifying a UBO, and does it change for lenders?

The standard UK threshold is 25% of shares or voting rights, the same test used for the Companies House PSC register. Lenders apply the same percentage test but layer a control test on top of it — a shareholder below 25% who holds veto rights or board appointment powers can still qualify as a UBO, and under the PRA’s large exposures rules, control (not just ownership share) is also what determines whether two borrowers must be treated as one connected group.

3. How does ownership resolution affect a bank’s lending limits, not just its AML file?

Under the PRA’s Large Exposures framework, banks must aggregate exposure across “groups of connected clients” — borrowers linked by control or financial interdependency — and hold capital against the group, not each entity separately. The PRA’s supervisory statement SS3/25 sets out how firms should identify these groups. The ownership chain traced for UBO purposes is the same evidence that answers whether a new facility pushes a lender over its internal limit to a group it already lends to under a different corporate name.

4. What happens when ownership data is wrong or stale at the point a loan is monitored, not just when it’s issued?

A covenant compliance certificate tests financial ratios; it does not test who controls the borrower today. A change of control, a parent company restructuring, or a connected SPV being struck off can happen mid-facility without appearing anywhere in routine financial monitoring. Continuous ownership monitoring closes that gap by re-checking registry filings and sanctions lists on an ongoing basis rather than waiting for an annual review cycle.

5. Why do property developers pose a harder UBO resolution problem than most borrowers?

A single development is often financed through a dedicated SPV with no independent trading history, sitting beneath a holding company, with equity from a family trust and mezzanine debt from an investment fund. Each layer can be registered in a different jurisdiction with different disclosure rules. Resolving the real controlling party means tracing effective ownership through every one of those layers and calculating the combined percentage, not just reading the top-line shareholder register.

6. What makes UBO evidence “FCA-grade” rather than just internally satisfactory?

The FCA’s own 2026 review of firms’ due diligence practice found that the firms marked down were the ones that could not produce evidence of what was actually checked and when. FCA-grade evidence means every UBO determination is traceable to a specific, dated filing from an authoritative source — a government registry — rather than an analyst’s notes or a commercial database export that cannot itself be traced back to a primary filing.

7. Can AI fully automate UBO resolution for commercial lending, or does a human still need to be involved?

The data-intensive parts — pulling registry filings, tracing multi-layer ownership, calculating effective percentages, screening against sanctions and PEP lists, and flagging ownership changes — are well suited to automation. The judgment calls are not: whether a structure makes commercial sense, whether to escalate an ambiguous nominee arrangement, and whether the risk is acceptable given everything else known about the borrower. The effective model pairs automated resolution with analyst judgment at those decision points, which is how Zavia.ai is deployed inside OakNorth’s workflow rather than replacing the credit committee.