The true cost of a manual expense claim in a UK business is between £6 and £15 of finance time, before you count the policy errors, the missing VAT and the receipts that never arrive. AI-assisted expense automation gets the per-claim cost under £1 and removes the spreadsheet that runs your expense process from the picture entirely. Here is the work that pays back, the work that does not, and what it actually costs.
A typical UK SME finance team handles between 30 and 300 expense claims a month. Take a 25-staff business doing 80 claims a month. Each claim travels through this loop: the employee photographs the receipt, sends it by email or files it in a shared drive, the finance person types it into the company expense spreadsheet (date, vendor, amount, VAT, category, project, mileage if relevant), checks it against the expense policy, decides whether it needs an approver, chases the approver, and finally keys it into Xero or QuickBooks once approved.
Time-and-motion on that loop is between fifteen and forty minutes per claim, depending on policy complexity and how cleanly the receipt arrives. At a finance person's £25 to £35 per hour fully-loaded cost, that is £6 to £15 per claim of pure finance time, before counting any other overhead. Eighty claims a month is two to four working days of finance time gone on expense admin, every month.
Then there is the silent cost. Claims that arrive without the receipt are typed in anyway and never get audited. VAT is keyed in or omitted by the employee, and the finance team rarely has the time to verify it, so VAT recovery quietly leaks. Approvals get chased over Teams and never get pinned to the claim. The expense spreadsheet itself becomes a brittle artefact that breaks when two people edit it, when the year rolls over, or when the rules change.
The honest reframe: this is not a fancy automation problem. It is a workflow whose unit economics no longer make sense.
Open the spreadsheet that runs expenses in any UK SME and you will see a version of this: columns for date, employee name, vendor, gross, VAT, net, category, project or client code, approver, status, paid date, and a free-text notes column where the unusual cases live. Frequently a second tab for mileage, a third tab for credit-card reconciliation, and a fourth that tries to be a year-to-date summary that never quite balances.
The spreadsheet works for about eighteen months. It breaks when one of these happens:
Once the spreadsheet has broken in two of those ways, the cost of keeping it running starts to exceed the cost of replacing it.
The market is noisy because four different categories of vendor sell into this space. Here is what they actually do and where each fits:
| Option | Typical cost | Best for | Avoid if |
|---|---|---|---|
| Stay on the spreadsheet, add Zapier flows | £30 to £150 / month | Under 30 claims a month, simple policy | Multi-entity, multi-approver, VAT complexity |
| Pleo or Soldo prepaid-card-plus-app | £5 to £12 per user / month plus card interchange | Field teams, simple approval policy, modern accounting stack | You need claim-back-from-bank-statement, complex VAT recovery |
| Concur, SAP Expense, Workday Spend | £15 to £30 per user / month plus £8k+ implementation | 250+ employees, enterprise approval chains, audit-heavy sector | Under 100 staff. The implementation cost alone is the build cost of custom |
| Dext, AutoEntry, Lightyear receipt capture | £20 to £80 / month, more for high-volume tiers | Accountancy practices reading their clients' receipts, small businesses with simple flow | You need the policy, approval and reporting parts, not just receipt-to-Xero |
| Custom build with AI receipt extraction | £4,000 to £9,000 fixed quote, three to four weeks | Multi-entity, unusual approval policy, restricted-fund tracking, the exact workflow you actually run | Your workflow already fits Pleo or Soldo and the SaaS price is acceptable |
An AI that auto-approves expenses without a human in the loop. Even for low-value claims under £50. Auto-approval looks like a saving until the first time it lets through a personal Amazon order or a duplicate claim, and then the audit cost dwarfs the time you saved. The right pattern is AI extracts and routes, a human approves, every time. The time saving is in the typing, not the approval.
An AI that classifies expense categories without checking your chart of accounts. Generic AI receipt-readers categorise an Uber as Travel. Yours might want it as Client Hospitality, Project 4502 - Travel, or Excluded - Personal. The model has no way to know unless your specific chart-of-accounts logic is wired in. Skip the SaaS that promises "AI category prediction" without seeing your books.
An AI receipt scanner sold as the whole solution. Dext and AutoEntry read receipts well, and if that is all you need, use them. But people often buy them expecting them to handle the policy check, the approval routing and the reconciliation, then discover those parts are still manual. If the receipt is only twenty percent of the friction in your process, you have not solved the problem.
These are illustrative shapes drawn from UK SME workflow patterns. The fixed-quote bands are the real ranges we quote for builds of this shape; the workflows described are the patterns that fit each context.
Profile: finance person spending around two days a fortnight on expenses. Receipts arriving by email and WhatsApp. Mileage tracked separately and never matching the rest. The shape that fits: a phone-app receipt capture (employees photograph, AI extracts), a Xero-integrated approval workflow routed by site manager, and a mileage tracker tied to the project codes. Typical band: £5,000 to £8,000 fixed quote, three to four weeks. Expected outcome: finance time on expense admin drops to under a day a fortnight, against build cost paying back inside six months on time saving alone.
Profile: half the spend in USD on tooling. Founder approving on Slack, finance double-keying gross then FX-adjusted. The shape that fits: a card-linked claim flow integrated with Wise and Xero, currency captured at source, exchange rate fixed on approval, posted to Xero with correct multi-currency entries. Typical band: £4,000 to £6,000 fixed quote, three to four weeks. Expected outcome: founder out of the approval bottleneck, finance off double-entry.
Profile: each site running its own informal expense flow. Vet nurses claiming reimbursable items, lab fees passing through, drug-supplier returns. The shape that fits: a site-routed approval workflow with a clinical-supplies category that auto-flags items above a per-site threshold, integrated with the practice-management software for the project codes. Typical band: £6,000 to £10,000 fixed quote, four to six weeks. Expected outcome: a single group-wide view of expense spend across all sites, with per-site rollup for management.
Profile: materials receipts, fuel receipts, subcontractor day-rate claims, plus the CIS deduction consequence on the subcontractor side. The shape that fits: a receipt capture that separates employee expenses from CIS subcontractor records, posts to Sage correctly tagged, and surfaces VAT recovery on diesel that is commonly missed at this scale. Typical band: £7,000 to £11,000 fixed quote, four to six weeks. Expected outcome: VAT recovery cleaner, finance audit trail tighter for CIS verification.
Profile: charity restricted-fund tracking adds a layer to every claim: which fund the spend draws from, audit trail for the trustees, compliance for the SORP report. The shape that fits: a claim flow with mandatory fund tagging, a restricted-fund balance visible to the operations manager, and an export tailored to the audit format an external accountant typically requests. Typical band: £5,000 to £8,000 fixed quote, three to four weeks. Expected outcome: year-end restricted-fund reconciliation in hours rather than the typical week of work.
The team that does the build is a team of AI agents directed by twenty years of software delivery experience. The agents bring the volume of the build. Senior engineering judgement decides what should be built and what should not. That combination is why a project a traditional agency would quote at three months ships from us in three to four weeks at a fraction of the cost.
The shape is consistent across every engagement. We scope it together in a 30-minute call. We send a written fixed quote. You approve. We ship in cycles, you see real working software in week one. We deploy on your environment so you own the code, the data and the deployment. If we part ways, the automation keeps running.
See the full delivery model on AI implementation for one-off projects, or tech partnership if you want the build plus the ongoing run-and-improve.
A focused automation, taking the receipt-and-claim workflow off one finance person, runs £4,000 to £9,000 as a fixed-quote project, shipping in three to four weeks. A larger build that handles multi-entity, multi-currency and FX, or that integrates with Xero plus your CRM and your bank feeds, runs £9,000 to £18,000. Off-the-shelf SaaS like Pleo or Soldo costs £5 to £12 per user per month plus card interchange. Custom is worth it when the SaaS does not fit your approval chain, your VAT recovery process or your multi-entity reporting.
It removes the receipt-photo, the spreadsheet keying, the policy check, the approval routing, the bank reconciliation and the VAT extraction from the manual process. The employee photographs the receipt on their phone; AI reads the date, vendor, amount, VAT, expense category and any policy flags; the claim auto-routes to the right approver based on amount and team; once approved, the claim posts directly to Xero or your accounting system, the bank reconciliation lines up, and the VAT recovery is captured. The finance person reviews exceptions and edge cases instead of typing.
Use Concur if you are large enough that the £15-£30 per user per month plus implementation cost is a rounding error and you need enterprise approval chains. Use Pleo if you want the prepaid-card-plus-app model and your approval policy is simple. Build custom when your approval workflow has unusual edges (multi-entity, charity restricted funds, project-coded expenses, complex VAT recovery) that the SaaS forces compromises on. The break-even is usually around 30 to 50 finance-touched expense claims a month.
A focused automation ships in three to four weeks from kickoff. The biggest single factor is access to your accounting system. Xero is fast, QuickBooks is fast, Sage is slower because of older API patterns, and any in-house accounting system depends entirely on how its API was built. The AI piece is a small part of the timeline, the integration is the work.
Receipts captured by the automation are stored digitally, which HMRC has accepted as the record of original supply since 2013. The automation captures VAT amount, VAT rate and supplier VAT number where available, and posts them in the format your accounting system expects for VAT return purposes. You still need the supplier VAT number for any claim over £250, so the automation flags missing VAT numbers for human handling rather than guessing.
Yes. The automation removes the typing, the chasing and the categorisation, not the judgement. The finance person now spends their time on the exceptions, the suspicious claims, the policy refinement and the work that actually requires a human view of the business. Typically, a finance team that was spending half their week on expense admin reclaims most of that time for higher-value work.
Related guides: Automate invoice processing · Automate data entry
Email us with rough volume and current setup, get a written fixed quote within two working days, free.
Email oliver@digitalsignet.com