Most invoice automation pitches start with OCR accuracy and end with a SaaS subscription you renew without checking. The honest test is different: how many invoices still need a human eye after the tool has run, and what does that cost you per month. This is the guide that answers that, with the price bands, the comparison and the cases where a custom build beats Concur and Tipalti on three-year cost.
The UK SME AP function typically has the same shape. A clerk opens emails, downloads PDFs, types line items into Xero or Sage, codes the nominal, matches to a purchase order if one exists, gets the approver to nod over it on Teams, files the PDF in a SharePoint folder somebody set up three CFOs ago. On a hundred invoices a month it is annoying. On a thousand it is a full-time job, plus overtime in the last week of the month.
The arithmetic that matters: the average UK SME AP function processes 200 to 2,000 supplier invoices per month, at a fully-loaded cost of £8 to £15 per invoice when done by hand. Automate it properly and that drops to £1 to £3 per invoice including the exception queue. On 800 invoices a month, you are looking at £80,000 a year of recovered capacity. The investment to get there is somewhere between £4,000 and £12,000, plus a few hundred pounds a month of running cost. That is the maths nobody puts on the slide deck.
Before any AP automation conversation, there is an Excel workbook. It typically appears in every sector. It usually has these columns:
This workbook breaks at predictable points. More than four people touching it. Multiple branches or entities sharing one VAT registration. Any month with more than 300 invoices. Cross-period accruals. The auditor asking for a PDF of invoice 0427 from August. The senior who left taking the formulas in their head with them. We are not building software to replace the workbook. We are building it because the workbook has already broken.
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. The payback curve is steepest where invoice volume is high, supplier formats are varied, or the rules for posting an invoice are domain-specific.
Profile: a 12-partner practice doing outsourced bookkeeping for around 180 clients, spending roughly 35 hours/week on invoice posting alone. Dext catches 70 percent of it. The remaining 30 percent (handwritten receipts, unusual layouts, multi-line builders' invoices) sits with a junior. The shape that fits: a focused automation that handles the residual and pushes a verified batch into Xero per client, with branch logic and CIS where relevant. Typical band: £6,000 to £10,000 fixed quote. Expected outcome: recovered junior time pays it back inside four months.
Profile: a regional builders' merchant taking around 1,500 supplier invoices/month against open POs, needing three-way matching (PO, GRN, invoice) that Sage 200 does grudgingly and Excel does badly. The shape that fits: a custom build that reads the invoice, finds the PO by supplier reference, validates against goods-received notes, flags pricing or quantity discrepancies for the buyer, posts the cleanly-matched 80 percent straight through. Typical band: £10,000 to £15,000 fixed quote. Expected outcome: a full AP role of capacity recovered.
Profile: CIS deductions, retention, application-for-payment cycles, payment notices, pay-less notices. This is the work that defeats every off-the-shelf tool. The shape that fits: a construction-specific AP automation that reads the subcontractor's application, validates CIS status with HMRC, applies the correct deduction, holds retention against the contract terms and matches against valuations. Typical band: £10,000 to £18,000 fixed quote. Expected outcome: the error rate on manual CIS is high enough that the HMRC-penalty saving alone often justifies the build.
Profile: a 6-home care group with supplier invoices arriving to a head-office inbox that need splitting across the right home for cost analysis, with restricted-fund tracking on anything CCG-funded. The shape that fits: an automation that reads, classifies by home, posts to Sage with the right cost code, surfaces variance against the home's budget. Typical band: £8,000 to £12,000 fixed quote. Expected outcome: the registered manager finally sees real-time spend per home instead of a six-week-late report.
Profile: contract-and-temp recruitment running on two-sided invoice flows. Umbrella companies invoicing the agency, the agency invoicing the end client, with margin calculated per timesheet. The shape that fits: an AP automation that captures the umbrella invoice, reconciles to the contractor timesheet, posts to Bullhorn or the back office, surfaces margin slippage on the spot. Typical band: £8,000 to £14,000 fixed quote. Expected outcome: recovers 1 to 3 percent of contract margin that currently leaks.
Four routes get you to invoice automation. Each is right for a different size and shape of business. The honest comparison is below. No vendor pitches, no asterisks.
| Route | When it fits | Real cost |
|---|---|---|
| DIY (Zapier, Power Automate, Make) | Sub-100 invoices/month, simple supplier list, in-house tech-confident person, willing to maintain it. Breaks fast at scale. | £20 to £100/mo + your time |
| Off-the-shelf OCR SaaS (Dext, AutoEntry, Lightyear) | Up to 500 invoices/month per entity, mostly standard suppliers, accounting in Xero or QuickBooks. Best entry point. Stops at OCR + push. | £20 to £100/mo per entity |
| Enterprise AP SaaS (Concur, Tipalti) | 500+ invoices/month, multi-entity, global payments, regulated industry. Real implementation cost, real ongoing licence. | £15,000 to £50,000/yr |
| Low-code (Power Automate + AI Builder, Microsoft Dynamics) | You are already a Microsoft 365 shop with a Power Platform licence and an IT lead. Real capability, slow build. | £8,000 to £25,000 build + licences |
| Custom build (us, or an equivalent shop) | You have a specific rule set the SaaS does not handle (CIS, three-way match, multi-home cost coding, complex VAT). 200+ invoices/month. You want to own the IP. | £4,000 to £18,000 fixed quote |
We turn down more AP automation work than we take on. The patterns:
"Replace our entire finance team with AI." No. Your finance team makes judgement calls that an AI gets wrong in expensive ways. The honest path is automating the data-entry layer and freeing the senior people to do the work they trained for. If the pitch is headcount reduction, we walk.
"Auto-pay every invoice that matches a PO." Tempting, dangerous. The fraud surface on auto-pay is real (supplier-impersonation, invoice-amendment attacks, bank-detail changes). A human approval gate on any payment over a defined threshold is non-negotiable. The automation gets the invoice ready for approval in two seconds; the approval still belongs to a person.
"One AP tool to handle every entity in our group." If your group has materially different sectors (one entity is construction, one is hospitality, one is retail), one tool will fit none of them. Build per-entity or per-sector. The overlap is rarely worth the compromise.
Every project is scoped and fixed-quoted before any work starts. The bands below are the real ranges we quote for builds of this shape.
| What it is | Timeline | Fixed quote |
|---|---|---|
| Small focused AP automation (one workflow, one accounting system) | 2 to 4 weeks | £3,000 to £8,000 |
| Invoice automation with classification and posting | 3 to 6 weeks | £4,000 to £12,000 |
| AP assistant with approval flow and exception queue | 4 to 8 weeks | £6,000 to £15,000 |
| Supplier portal with three-way match and payments | 6 to 12 weeks | £12,000 to £40,000 |
| Ongoing tech partnership (we run, monitor, improve) | Monthly, no lock-in | £450 to £1,500 / month |
The shape of an engagement is on the AI implementation page. The ongoing version, where we keep your automation running and adapt it as your supplier mix changes, is on the tech partnership page. The full engagement shape and previous delivery experience sit on projects and the bigger custom builds on app builds.
Invoice processing is rarely the only AP pain. The two guides most clients ask for in the same conversation:
A focused invoice automation that reads supplier invoices, posts them into Xero, QuickBooks or Sage and routes anomalies to a human runs £4,000 to £12,000 as a fixed-quote build. An ongoing tech partnership where we run, monitor and improve it is £450 to £1,500 per month. Off-the-shelf SaaS like Dext or AutoEntry sits at £20 to £100 per month per entity but stops at OCR plus routing; once you want approval flows, supplier portals or three-way matching, you are either paying enterprise SaaS prices (Tipalti, Concur) or building.
For a small practice or a sub-100 invoices/month business, Dext or AutoEntry is usually the right answer. They handle OCR, line-item extraction and the Xero push competently. You hit the wall when you need supplier-specific logic (multi-PO matching, contract-rate validation, project coding, branch splits), real approval workflows, or anything custom on the AP side. At that point you are paying £400 to £1,200/month for a tool that still leaves a person doing the awkward bits. A scoped custom build is often cheaper over three years.
Three to six weeks from kickoff for a focused build (read invoice, classify, post to accounting system, exception queue). Six to ten weeks for a multi-system flow involving PO matching, approval routing and a supplier portal. The biggest delay is never the AI. It is getting API access to your accounting platform, your ERP and your bank feed. If we have credentials in week one, we ship fast.
Yes, sometimes. Modern document AI gets line-level accuracy in the high 90s on well-structured invoices and lower on handwritten or photographed receipts. The honest design is not perfect accuracy. It is a confidence threshold: anything below a defined score routes to an AP clerk with the suspect fields highlighted. Anything above posts straight through. The volume on the human queue drops by 70 to 90 percent compared to manual entry, and the residual is the work that needs human judgement anyway.
Yes, and this is one of the highest-payback cases we see. CIS deductions, retention, application-for-payment cycles, dayworks, variations: all of it is rule-based once a domain expert defines the rules. A custom AP flow for a construction firm will validate the subcontractor's CIS status against HMRC, apply the correct deduction, hold retention, and reconcile against the application for payment automatically. The off-the-shelf tools do not do this.
All three. The accounting platform is where invoices live, so the automation has to integrate cleanly. Xero has the friendliest API and ships the fastest. Sage 200 and Sage 50 work well but the API surface is older. QuickBooks is straightforward. We do not replace your ledger; we feed it cleaner, faster, with a tighter audit trail.
Tell us the supplier mix, the volume, and the accounting system. We will tell you honestly whether off-the-shelf is enough, and if a custom build is the right call, scope and quote it before any work starts.
Email oliver@digitalsignet.com