HMRC Is Using AI to Audit Tax Returns | Receiptflow
HMRC Is Using AI to Audit Tax Returns. Is Your Clients' Paperwork Ready?
Tanvir Alam•Sep 14, 2026•7 min read•Receipt Management
HMRC is now using AI to review tax returns at scale, so complete, categorised, retrievable digital records are your clients' first line of defence when a return is flagged.
HMRC now has an AI system reading your clients' tax returns. In May 2026 it signed a £175 million, ten-year contract with the British data firm Quantexa to connect its fragmented data, flag errors and fraud, and narrow a tax gap estimated at £46.8 billion. Audits used to be slow, manual, and partly random. That is no longer the model. Every return your practice files is now cross-referenced by software that does not tire and does not stop looking.
For your practice, this is not abstract policy. It is a question about the folder of receipt photos sitting in your inbox right now. When an algorithm reviews returns at scale and surfaces the inconsistencies, clean, categorised, digital records stop being good housekeeping. They become your clients' first line of defence.
What HMRC's AI tax audit contract actually does
The contract was announced on 14 May 2026 and runs to March 2034. It hands Quantexa responsibility for modernising HMRC's data infrastructure and layering AI on top of it.
Quantexa's platform pulls structured and unstructured data from HMRC's own systems and outside sources, then builds a connected view of each taxpayer. The aim is to spot patterns across billions of data points that a human reviewer would never see.
That covers large-scale VAT and payroll fraud. It also covers ordinary mistakes in ordinary returns. HMRC has said the system is meant to correct unintentional errors, not only catch deliberate ones.
AI flags the return, a person runs the enquiry
One detail matters for how you brief clients. HMRC and Quantexa have both said AI supports human decisions rather than replacing them. Automated decisions are meant to stay transparent, auditable, and explainable, with an officer signing off. Quantexa's chief executive, Vishal Marria, told the BBC that in government environments, AI cannot operate as a black box, and that decisions need to be transparent, auditable, and explainable, particularly in areas affecting citizens directly.
So no machine is issuing penalties. What has changed is the quality of the triage that decides which returns get a closer look. More returns get flagged, and the ones that do are flagged on evidence.
Making Tax Digital is what makes it work
The timing is not a coincidence. From April 2026, sole traders and landlords earning over £50,000 have to keep digital records and file quarterly under Making Tax Digital for Income Tax. The volume of structured digital data reaching HMRC has jumped, and Quantexa's platform exists to process exactly that. The more digital the tax system gets, the more an AI layer can do with it.
What an AI-driven HMRC enquiry looks like for your clients
Take a sole trader who claims a run of supplier costs but has kept only a partial set of receipts. Under the old model, that gap might never have surfaced unless the return was picked for review. Under an AI system, the mismatch between declared expenses and the supporting trail is exactly the kind of inconsistency the platform is built to find.
When the enquiry letter arrives, the question is no longer whether you can explain the gap. It is whether you can produce the records, dated and itemised, that day. A practice with every receipt captured, categorised, and stored digitally answers in an afternoon. A practice working from photos in a messaging thread and a box of paper spends a fortnight rebuilding the position, usually at its own cost.
Why the records matter more than the mistake
HMRC's penalty regime already treats an inaccuracy differently depending on the behaviour behind it. An honest error that the client took reasonable care to avoid can carry no penalty. The same error with no supporting records, treated as careless, can carry a penalty of up to 30% of the tax at stake. If it looks deliberate, that figure rises sharply.
The dividing line is often the quality of the evidence. A client who can produce a complete, dated record trail looks careful. A client reconstructing figures from memory after the letter arrives looks careless, whatever the truth. Clean records are not just faster to produce. They change how the enquiry is categorised.
The £46.8 billion reason for the investment
HMRC is not spending £175 million without a return in mind. The tax gap, the difference between tax owed and tax collected, sits at around £46.8 billion. Closing even a few per cent of that covers the contract many times over.
For a sense of how well this approach already pays, HMRC's existing Connect system brought in an extra £4.6 billion in 2024/25 by cross-referencing datasets, a Freedom of Information request confirmed it flagged 540,000 cases of undeclared tax that year. Quantexa is the next and much larger step down the same road.
Clean digital records are the defence
Here is the reframe for your practice. Receipt capture used to be an admin chore you fitted around the real work. Now it is the evidence base that protects clients when a return is flagged.
Digital receipt records for an HMRC enquiry need three things. They have to be complete, they have to be accurate, and they have to be retrievable on demand. A pile of paper fails all three.
This is where automated receipt extraction earns its place. Receiptflow reads the supplier, date, VAT, and net and gross totals from every receipt as it arrives, then passes clean data into your accounting software. No typing, no chasing, no gaps for an algorithm to find later.
> Capture every client receipt the moment it lands and turn audit season into a non-event. See how Receiptflow keeps your records HMRC-ready.
Which of your clients are most exposed
Some client profiles draw more algorithmic attention than others:
Cash-heavy trades such as construction, hospitality, and hair and beauty, where expense claims lean on receipts rather than bank data.
Recent Making Tax Digital entrants filing quarterly for the first time, where early returns set a baseline the system measures against.
Clients with round-number or fast-growing expense claims that do not track a matching supplier trail.
Clients who changed accountant mid-year, where the record trail has a visible handover gap.
None of these are problems on their own. They become problems when the supporting records are incomplete. Knowing which of your clients sit in these groups tells you where to tighten capture first.
How to get your clients audit-ready in 2026
You do not need to change everything at once. Three moves close most of the exposure.
1. Move every client off paper and photos
Paper receipts fade and go missing, and they carry no data you can search. Photos in a messaging app are not a record system. Move clients to one capture method where every receipt is read, dated, and stored as it arrives, with no manual re-entry. Manual re-entry is where gaps and errors start.
2. Categorise at the point of capture
An enquiry tests whether your numbers reconcile. Records that are captured but left uncategorised still leave you rebuilding the position under time pressure. Categorising at capture means the data is enquiry-ready as it lands, not three months later when you can least afford the work.
3. Keep a retrievable trail for every figure
If a client's return is flagged, you want to produce the source document for any line in minutes. A digital system with searchable, dated records turns a stressful enquiry into a routine request. That retrievability is the biggest single difference between a practice that dreads HMRC contact and one that takes it in its stride.
What this means for your practice
The Quantexa contract runs for ten years. It is not a pilot. AI-assisted review of tax returns is the baseline now, and it will sharpen as more digital data arrives through Making Tax Digital.
The practices that come out ahead are the ones that make clean, digital, categorised records standard for every client now, rather than after the first flagged return. Getting there is the straightforward part. Sort capture and categorisation once, and your practice is ready for whatever the algorithm asks.
> Ready to make every client audit-ready? Start with Receiptflow and capture receipts the moment they arrive.
FAQs
Common Questions with Clear Answers
Is HMRC really using AI to audit tax returns?
Yes. In May 2026 HMRC signed a £175 million, ten-year contract with the British firm Quantexa to deploy AI across its compliance and data work. The system reviews returns at scale to detect fraud and correct errors, with human officers signing off automated decisions.
Will AI replace HMRC tax inspectors?
No. HMRC has said AI supports human decision-making rather than replacing it. Automated decisions must stay transparent, auditable, and explainable, and an officer approves the outcome. The AI decides which returns need a closer look; people run the enquiry.
What records do my clients need for an AI-driven HMRC enquiry?
Complete, accurate, and retrievable digital records for every figure on the return. That means every receipt and invoice captured, categorised, dated, and searchable, so you can produce the source document for any line on demand.
How does receipt automation help with HMRC compliance?
It removes the gaps and errors an AI system is built to find. Automated capture reads each receipt as it arrives, extracts the key fields, and stores clean data in your accounting software, giving you a complete digital audit trail without manual entry.
Does Making Tax Digital make AI audits more likely?
It makes them more effective. Since April 2026, sole traders and landlords earning over £50,000 must keep digital records and file quarterly. That gives HMRC far more structured data, which is exactly what Quantexa's platform is designed to analyse.
