HMRC's AI Recovered £10B. Small Business Is 62% of Why.

What is HMRC's AI actually doing to small business records?

HMRC's AI is not a future threat. It is operational right now, and small businesses are its primary target. HMRC's own tax gap report puts small business non-compliance at 62% of the total UK tax gap, which ran to £36 billion in 2022/23. The £10 billion recovery figure reflects what automated systems are already pulling back. If you run payroll, file VAT, or take card payments, your data is already in the system.

The core tool is HMRC's Connect platform, which processes over 55 billion data points annually. It pulls from bank records, card processor reports, Companies House filings, Land Registry, social media, and online marketplaces. It does not wait for a tip-off. It runs continuous pattern matching against your reported figures.

What specific patterns is the system flagging?

Connect and its downstream models are not looking for obvious fraud. They are looking for statistical anomalies: figures that fall outside what similar businesses in your sector, region, and size band report. A few categories get flagged consistently.

Revenue suppression. If your card terminal receipts, reported through Payment Service Providers under HMRC's PSP reporting rules, do not reconcile with your VAT returns, that gap gets flagged automatically. Restaurants, tradespeople, and retailers with mixed cash and card income are particularly exposed.

Lifestyle inconsistency. Connect cross-references your reported income against property purchases, vehicle registrations, and even publicly visible business activity. A sole trader reporting £28,000 income who buys a £400,000 property will be queued for review.

Sector benchmarking. HMRC publishes business tax performance benchmarks for gross margins, expense ratios, and wage-to-revenue ratios by sector. If your numbers sit more than one or two standard deviations outside the norm, the system scores you as higher risk. No human required.

Director loan and payroll anomalies. Irregular director drawings, salary structures that shift aggressively around tax thresholds, and pension contributions that spike in a single year are all pattern-matched against prior years and sector peers.

"The question isn't whether your books will be reviewed by a machine. It's whether your records are clean enough to pass without triggering a human investigation."

How good is HMRC's AI actually getting?

Good enough that the trajectory matters more than the current state. HMRC has committed to significant AI investment as part of its Making Tax Digital programme, which mandates quarterly digital reporting for sole traders and landlords earning over £50,000 from April 2026, dropping to £30,000 from April 2027. That is not an administrative convenience. It is a data pipeline that feeds the risk models in near real-time rather than annually.

Connecting quarterly submissions to card processor data, open banking feeds, and Companies House means the lag between an anomaly appearing and HMRC scoring it as a risk drops from 18 months to weeks. Businesses that have relied on annual filing timelines as a buffer no longer have one.

What does a machine-proof set of books actually look like?

This is not about being perfect. It is about being explainable. HMRC's systems flag anomalies; human investigators then decide whether to open an enquiry. If your records can explain the anomaly before anyone asks, enquiries close fast or never open.

Reconciliation at source. Every revenue stream, card, cash, invoice, marketplace, should reconcile to a single figure before it hits your accounts. Gaps between what a platform paid you and what you reported are the single most common trigger. Tools like Xero, QuickBooks, and FreeAgent pull direct bank feeds that create this audit trail automatically.

Expense categorisation consistency. Reclassifying expenses between years, putting personal costs through the business sporadically, or using vague descriptions like "sundries" across large line items all score poorly. Consistent, specific categorisation with receipts attached at point of entry is what passes machine review.

Documented director decisions. If director pay, dividends, or loan accounts change materially year on year, board minutes or a written rationale should exist. This is standard governance. Most small businesses skip it. The ones who get enquiries usually wish they hadn't.

Benchmark awareness. Know what your sector's gross margin and expense ratios look like. If you run a 38% gross margin in a sector that benchmarks at 62%, document why: discounting strategy, high input costs, a particular client mix. Do not leave the model to guess.

| Risk Factor | What HMRC's System Sees | What You Need | |---|---|---| | Card vs VAT mismatch | PSP data vs return | Monthly reconciliation file | | Lifestyle vs income gap | Property, vehicles, travel | Income documented across all sources | | Sector margin deviation | Benchmarked gross margin | Written narrative for outliers | | Director loan movement | Year-on-year balance shifts | Board minutes, loan agreements | | Expense reclassification | Category changes between years | Consistent chart of accounts |

Does using AI tools in your own bookkeeping actually help?

Yes, for a specific reason. AI-assisted bookkeeping tools do not just save time. They enforce the kind of consistency that machine audits reward. When a tool like Dext or Hubdoc captures receipts at point of purchase, attaches them to transactions, and categorises them against a fixed chart of accounts, you are producing records that look the same to an auditor as they do to an automated risk model: clean, consistent, timestamped, and reconciled.

The irony is that the same AI infrastructure making HMRC more effective at finding problems is available to small businesses to stop those problems existing in the first place. Most SMB operators are not using it that way yet.

What we'd actually do

• Run a PSP reconciliation audit now. Pull every payment processor report for the last two full tax years and reconcile it line by line against what you filed. If there are gaps, document the explanation before HMRC asks for one.
• Map your numbers against HMRC's sector benchmarks. If your gross margin or expense ratio sits outside the normal range, write a one-page internal memo explaining why. Date it. Keep it with your records.
• Get your bookkeeping on a connected platform before MTD deadlines hit. Xero, QuickBooks, or FreeAgent with bank feeds active is not optional much longer. Set it up now while you can control the transition, not when a quarterly filing is due.