If your finance team regularly spends the first days of every month correcting how departments classified their expenses, the expense categories themselves may be the root cause.
Most mid-market finance teams have expense rules. What they often lack is a shared, trusted set of categories that maps cleanly to their statutory reporting, handles VAT correctly, and gives employees enough clarity to self-categorise without guessing. When that structure is missing, rules create a different kind of rework: chasing corrections that stem from a taxonomy nobody agreed on in the first place.
VAT treatment in particular can depend on your specific circumstances, so if in doubt, consult a professional accountant or HMRC directly.
Why most categorisation rework starts with the taxonomy, not the rules
)
Without a shared taxonomy, rules have nothing consistent to enforce, and departments default to their own classification logic. Only 2.6% of expense claims were approved immediately without follow-up or corrections, according to the Capture Expense 2025 Expense Trends Report. A further 27% were not approved until 30 days or more after submission.
Vague descriptions can prevent correct nominal code assignment. Missing or mismatched receipts may leave the actual category uncertain. Late submissions force accrual estimates, and incomplete details break the audit trail. The result: no enforcement mechanism until the close process.
The ICAEW toolkit (Institute of Chartered Accountants in England and Wales) identifies the root cause as structural: ambiguity about who owns the classification process. Senior management assumes it's handled. The people doing the day-to-day work don't have the authority or the tools to enforce consistency. The result is each department invents its own approach, and the finance team pieces the mess together at month-end whilst thinking there must be a better way to do this. And there is.
Building expense categories that hold under pressure
How do you make categories granular enough for your controller but simple enough for the marketing manager filing a receipt at the end of a long week? A practical expense taxonomy has to serve both audiences at once: your statutory reporting requirements and the employees who need to classify what they've spent without guessing wrong.
Consider separating taxonomy design from policy design. Define what categories exist first, then attach spend limits and approval workflows. When you start with the rules, you end up retrofitting categories around policies, and the taxonomy stops reflecting what you actually need to report.
Why every category is a compliance decision
Expense categorisation is a statutory obligation, and the consequences of getting it wrong are more concrete than most teams expect. Say a new hire codes a £12,000 annual software subscription to office supplies instead of cost of sales. It's a small mistake, easily made. But it flows through to the wrong profit and loss line, distorts your gross margin for two quarters, and when your auditors spot it, you're restating numbers the board has already seen.
The Companies Act requires accounting records that disclose your company's financial position with reasonable accuracy at any time. HMRC can impose penalties for inadequate records, as set out in its record-keeping rules. If HMRC determines you didn't exercise reasonable care, they can look back six tax years to recover underpaid tax, National Insurance, plus penalties and interest.
If you're applying Financial Reporting Standard 102 (FRS 102), as many mid-market companies do, employee expenses map across three P&L lines depending on their function:
Cost of sales: direct project costs, billable travel, and direct labour
Distribution costs: sales team travel, marketing expenses, and customer-facing costs
Administrative expenses: office costs, subscriptions, non-billable travel, training, and professional fees
How you categorise each expense determines which line it hits. Your team isn't getting this wrong because they're careless. They're getting it wrong because the taxonomy makes the right answer unclear.
Operational expense categories
The categories need to be specific enough that your controller can run management reports from them, but clear enough that someone outside finance can pick the right one without second-guessing.
The following table offers a starting point for mapping categories to FRS 102 reporting lines and VAT treatment. Adjust it to your chart of accounts.
| Category | Typical FRS 102 line | VAT note |
|---|---|---|
| Travel (transport) | Per function (cost of sales / distribution / admin) | Standard rated; keep VAT receipts |
| Travel (accommodation) | Per function | Standard rated |
| Mileage (own vehicle, at advisory rates) | Per function | Different treatment from actual travel costs; separate category required |
| Meals and subsistence (employee only) | Administrative expenses | Standard rated; must be employee-only to reclaim |
| Staff entertainment | Administrative expenses | 100% VAT recovery on staff-only events |
| Client or prospect entertainment | Distribution / administrative | Blocked: zero VAT recovery on non-employees |
| Office supplies | Administrative expenses | Standard rated |
| Software subscriptions | Cost of sales OR administrative | Sub-classify: direct delivery cost vs overhead |
| Professional development | Administrative expenses | Standard rated |
| Professional fees | Administrative expenses | Standard rated; check partial exemption if mixed use |
| Phone and communications | Administrative expenses | Apportion based on business vs personal use |
| Potential CapEx pending review | Holding category | Route for controller sign-off before posting |
| Miscellaneous (controlled) | Varies | Mandatory free-text description; quarterly review |
You need each category to map to exactly one nominal ledger code in your accounting system. When two categories share a code, your reconciliation breaks.
When your spend management platform can import your chart of accounts directly, every export matches your ledger structure without manual mapping. That gives you cleaner records, faster reconciliation, and fewer month-end corrections.
Staff and client entertainment
Staff and client entertainment must always be two separate categories. VAT Notice 700/65 allows 100% VAT recovery on staff-only events but blocks any recovery on entertaining UK non-employees. Merging these categories means you'll either claim VAT you shouldn't or fail to claim VAT you're entitled to.
Say your sales manager expenses a dinner with a prospective client and two colleagues. Without separate categories, this lands in a generic entertainment line, and your finance team can't split the VAT recovery without going back for the attendee breakdown.
When events include both staff and external guests, expense submissions should capture total attendees, number of employees, and number of non-employees. HMRC's compliance guidelines (GFC8) require VAT to be apportioned between reclaimable staff portions and blocked non-employee portions. Capturing this data at the point of submission will save your team from trying to reconstruct it later.
Partial exemption and digital records
If your business makes both taxable and exempt supplies, VAT Notice 706 requires you to attribute input tax across three categories:
Input tax directly attributable to taxable supplies is fully recoverable
Input tax directly attributable to exempt supplies is not recoverable
Residual input tax is recoverable only in proportion to taxable use
Every miscategorised expense distorts this calculation. The Chartered Institute of Taxation (CIOT) warns that businesses tend to default to treating uncertain expenses as exempt. If your team does that across £50,000 of mixed-use input tax over a year, the under-recovery adds up fast. That's cash you were entitled to reclaim, quietly leaking out of every VAT return.
Making the taxable/exempt split clear at the point of categorisation saves your team from untangling it at month-end.
VAT Notice 700/22 requires records to be kept digitally. A single incorrect coding rule applied across multiple transactions can flow into every VAT return until you identify and correct it. Quarter after quarter. What starts as an isolated error can become a recurring compliance failure that compounds over time.
Each transaction also needs context beyond the category itself. Cost centre, project or client code, supplier, and approval workflow data give you detailed reports across multi-department organisations and save time pulling data for audits or board reporting.
Keeping your expense categories accurate as you scale
Getting the taxonomy right is the hard part. Keeping it right is the ongoing one. Even a well-designed category structure degrades without automation to enforce consistent coding and a named owner to maintain it as the business changes.
Rule-based automation
Rule-based automation within your spend management platform handles known patterns. If a supplier is always coded to the same nominal account, a rule fires every time and produces the same result. You get deterministic, auditable categorisation without needing training data.
Machine learning
But what about the transactions that don't match any rule? A new supplier, an ambiguous one-off, a non-PO invoice from a vendor your team hasn't seen before.
Machine learning (ML) models analyse historical patterns to recommend the appropriate general ledger (GL) code for unfamiliar suppliers, ambiguous spend, and transactions where no rule exists. This is especially useful for invoice processing where there's no purchase order to match against.
The ICAEW's position is that AI outputs should be treated as a starting point requiring human review. In practice, that means keeping an auditable rule layer alongside ML predictions.
One thing worth acknowledging: ML suggestions are only as good as your historical coding. If your team has been classifying inconsistently for years, the model learns those inconsistencies. Cleaning up your taxonomy first gives ML a better foundation to work from.
The combination works best when both approaches apply consistently across every spend type. Card transactions, expense claims, and supplier invoices should all code against the same taxonomy, in the same system.
In practice, re:cap, a European fintech platform, found that applying machine learning to its own bank transaction categorisation improved accuracy from 92% to 98.8% across hundreds of thousands of daily transactions.
Spendesk applies a similar hybrid model. Rules handle known coding patterns while intelligent categorisation suggests GL codes for new transactions. According to Spendesk, accountants using the platform see up to 80% fewer manual data entries, with a 90% acceptance rate on AI-suggested coding. In practice, that means fewer Slack messages chasing GL codes and less time toggling between spreadsheets during close.
Ownership and review cadence
Categories that worked when you had 80 employees may not hold up at 200. A named owner in your finance team should field the inevitable "where do I code this?" queries and keep the structure current as the business changes.
Review exceptions monthly to catch recurring miscodings early
Review the miscellaneous category quarterly for patterns that warrant a new named category
Update the full taxonomy annually whenever business models, regulations, or vendor relationships change
A close checklist catches categorisation gaps before they compound. Maintaining a change log with version control means prior-period reporting stays consistent when your auditors need to trace a reclassification back to a documented decision.
For company-wide adoption, consider training managers before employees. Managers are your first line of enforcement, and if key executives back the policy, adoption follows. Adoption also sticks faster when managers can configure their own approval workflows and spending controls directly. Setting a single go-live date sends a clearer signal than running old and new systems in parallel. Easier said than done, but the teams that commit to a clean cut-over tend to see adoption stick.
Where this leaves your finance team
)
A well-structured taxonomy does three things: it gives employees a clear path to self-categorise correctly, it maps directly to your statutory reporting and VAT requirements, and it gives your automation tools a stable foundation to code against. Without it, every other investment in process improvement or tooling is working against an unreliable base.
If you're evaluating how to bring your expense categories, automation rules, and accounting integrations into a single workflow, accounting automation is a practical place to start.
Frequently asked questions about expense categorisation
What is expense categorisation?
Expense categorisation is the process of classifying business costs into defined, consistent categories for financial reporting, tax compliance, management accounting, and audit purposes. In the UK, your categories need to satisfy FRS 102 reporting requirements, HMRC's tax record-keeping rules, and the accounting record obligations set out in the Companies Act 2006.
Why must staff entertainment and client entertainment be separate categories?
VAT Notice 700/65 allows VAT recovery on employee-only events but blocks recovery on entertaining non-employees. A single entertainment category makes correct VAT treatment impossible without manual investigation of every claim.
How should you categorise employee expenses for HMRC?
HMRC expects expenses to be classified into categories that align with your tax return and support your corporation tax deduction claims. At minimum, you'll need to separate allowable expenses (office costs, travel, staff costs, professional fees) from disallowed ones (client entertainment, personal expenditure), and keep the documentation to prove each classification. Categories should also flag items with specific VAT treatment, such as motoring expenses and mixed entertainment, so nothing falls through the cracks at return time.
What is the difference between rule-based and machine learning expense categorisation?
Rule-based categorisation applies deterministic logic: if a defined condition is met, a defined action fires. It's fully auditable, but it can't handle new transactions it hasn't seen before. Machine learning categorisation learns from historical patterns to handle exceptions. The hybrid approach is increasingly common, and the ICAEW recommends that AI outputs require human oversight.
Which expense categories attract the most HMRC scrutiny?
The HMRC toolkit identifies four highest-risk areas: capital vs. revenue misclassification, private and personal expenditure mixed with business costs, business entertaining and gifts, and directors' loan accounts.
Curious how Spendesk works?
Try an interactive demo to see spend control and approvals end-to-end.
Get a free tour)
)
)
)
)
)
)