There is a story doing the rounds that HMRC has handed 28,000 staff a Microsoft Copilot. That one is about admin speed. The deal HMRC signed on 14 May matters far more to the owner-managed clients I look after, and almost nobody is discussing it in those terms.
TL;DR: HMRC's £175m, ten-year Quantexa contract buys entity-resolution technology that maps the connections between a person and all their companies, not a faster way to read one return. For owner-managed groups, the bar moves from each return standing up alone to the whole network reading as commercially coherent. The fix is better documentation, not panic.
What Did HMRC Actually Buy?
On 14 May 2026, HMRC signed a £175 million, ten-year partnership with Quantexa, a London-founded data firm (Quantexa, 2026). The platform does two things in plain terms. Entity resolution links scattered records - companies, people, transactions - that point to the same underlying party even when names and references do not match. Graph analytics then traces how those resolved entities relate to each other across ownership layers.
HMRC's stated target is the £46.8 billion tax gap for 2023-24, 5.3% of total tax liabilities (HMRC, 2025), with the government committing to recover an extra £10 billion a year by 2030 (TNW, 2026). It is also a sovereignty move. The UK has spent more than £900 million with the US firm Palantir across government (TNW, 2026), and this contract keeps British tax data with a British supplier.
Is This Just the HMRC Copilot Story Again?
No, and the gap between the two is the whole point. The Copilot rollout gave roughly 28,000 HMRC staff a general assistant that saves about 26 minutes per officer per day on admin tasks (The Register, 2026). That makes an officer quicker at the desk. It does not change what the officer can see.
Quantexa sits in a different category. Its specialism is joining HMRC's internal data to external sources and surfacing the hidden structure between them. The old enforcement model was a stretched officer sampling a fraction of returns in isolation. The model this contract builds assumes the regulator can hold the whole network in view at once. I wrote recently about an AI close agent on my side of the desk, and this is the same architectural idea pointed back at the files I prepare.
What Entity Resolution Actually Sees in an Owner-Managed Group
Take a typical client of mine. A trading company, a property company, a holding company, one director who owns all three, and a spouse on the share register. None of the individual moves is wrong on its own. The intercompany loan between trade and property is documented. The director's loan account nets down each year against a consultancy invoice from a connected party. A residential property sits one step removed in a separate entity.
Read return by return, each one stands up. Read as a graph, the pattern is exactly what entity-resolution software is built to surface: circular value, connected-party pricing with no stated commercial rationale, balances that move in lockstep. The transactions have not changed. What has changed is that the regulator can now see them together rather than one return at a time. This is the same direction of travel as the close-company reporting consultation. HMRC wants the connections, by disclosure where it can get it and by inference where it cannot.
Should My Clients Be Worried?
Not if their structures are commercially real and documented. The honest version of this matters here, because the scaremongering version is already everywhere. Quantexa's chief executive Vishal Marria has been explicit that the system is built to pursue "professional criminal networks rather than tripping up honest taxpayers caught in paperwork errors", and that "everything the platform does is fully auditable" (Accountio, 2026).
He has also said that in government use "AI cannot operate as a black box", and that HMRC data "would remain inside HMRC's own environment and would never leave it" (TNW, 2026). Every automated flag still goes to a person. So this is a triage amplifier, not robo-assessment. It widens the net of what gets looked at; it does not decide the answer. The real risk is being one of the many more files that now gets a second look, with thin contemporaneous notes to explain a structure that is fine but reads as odd from the outside.
What I Am Changing in My Files From This Week
Three things, none of them dramatic. Intercompany balances get a written commercial rationale at the point they are created, not reconstructed at year end: why the loan exists, the terms, the repayment expectation. Director's loan account entries stop being bare journal lines; if a DLA nets against a connected-party charge, the file records why that charge is set at that amount.
Connected-party pricing gets a one-paragraph basis note. That is the same discipline the ICAEW credible-basis test and the new sanctionable-conduct regime already push me toward. None of this is new structuring or defensive paperwork for its own sake. It is contemporaneous documentation of decisions I was making anyway, so that when the network is read as a whole, the story on the file matches the story in the numbers.
Frequently Asked Questions
What is the HMRC Quantexa deal and what will it actually do?
HMRC has signed a £175 million, ten-year partnership with the British data firm Quantexa, announced 14 May 2026 (Quantexa, 2026). The platform uses entity resolution to link records that refer to the same person or company across fragmented datasets, and graph analytics to map the relationships between them. In practice it gives HMRC a connected view of a taxpayer and all their associated entities, rather than a return-by-return one.
Can HMRC now see connections across all my companies and related parties?
The technology is designed to surface exactly those connections - intercompany loans, common control, connected-party transactions, ownership held one step removed. It does not give HMRC new legal powers; it gives HMRC a faster way to spot patterns that were always there in the data. If your group structure is commercial and your transactions are documented, that visibility is not a problem. The exposure is undocumented arrangements that look anomalous when viewed as a network.
How should I document intercompany loans and director's loan accounts now?
Record the commercial rationale when the transaction happens, not at year end. For an intercompany loan, note why it exists, the terms, and the repayment expectation. For a director's loan account that nets against a connected-party invoice, record the basis for that invoice amount. The aim is a contemporaneous file note that explains the structure to a reviewer who is seeing it as part of a wider network.
Does this mean an AI will decide my tax enquiry?
No. Every automated flag still goes to a human officer, and Quantexa's chief executive has stated the platform is fully auditable and that HMRC data never leaves HMRC's environment (TNW, 2026). The system widens and prioritises what gets reviewed. It does not assess tax or replace the enquiry process. Treat it as a triage tool that makes a second look more likely, not a machine that reaches verdicts.
If you run an owner-managed group and want a second pair of eyes on how your intercompany and connected-party file notes would read to an outsider, get in touch - I am working through exactly this for my own clients and happy to talk it over.