Invoice Automation for UK Businesses: Build or Buy?

Every invoice that arrives at your business costs you money before you have paid a penny of it. Someone opens the email, reads the PDF, types the numbers into your accounting system, checks it against a purchase order, and files it. UK estimates put that at £8–12 per invoice once you count the staff time, at 10–20 minutes a go.

At 50 invoices a month, that is a nuisance. At 500, it is a salary.

So you start looking at automation, and within ten minutes you are drowning. Every result is a software company telling you their tool is the answer. None of them tell you the thing you actually want to know: whether you should buy one of these tools at all, or have something built that fits how your business already works.

This article answers that. No product pitch — we build custom systems and implement off-the-shelf ones, and the honest answer is that most businesses should start by buying.

What "invoice automation" actually means

Strip away the marketing and there are four separate jobs. Most tools do some of them. Confusion about which is which is where money gets wasted.

Extraction is the part everyone advertises, and it is the part that is basically finished. Good tools now claim over 99% accuracy on data extraction. That number is real for clean, typical invoices.

It is also the reason so many projects disappoint. Extraction was never the expensive part. Matching and posting are where the work is, because those two depend on your suppliers, your processes and your systems — none of which a software vendor has ever seen.

When you should just buy a tool

Buy, and stop reading, if most of this sounds like you:

• Your invoices are ordinary — PDFs from established suppliers, in one or two currencies
• You use Xero, Sage or QuickBooks in a fairly standard way
• You do not do complex three-way matching against purchase orders and delivery notes
• Under roughly 500 invoices a month

For that business, an off-the-shelf tool will do the job this month, for a predictable fee, with no build cost. Pricing sits around £125/month for 100 documents and £279/month for 500 at the small end. Set against £8–12 of staff time per invoice, it pays for itself quickly and you are done.

Anyone who tells you to commission a custom build for this is selling you something. Buy the tool.

When buying stops working

Off-the-shelf tools are built for the average business. The cost shows up when you are not average. Watch for:

Your invoices are not standard. Handwritten delivery notes, scanned faxes, supplier statements instead of invoices, or line items that need splitting across cost centres or projects. Generic extraction models are trained on ordinary invoices, and yours are the exception.

The matching rules are yours alone. If "does this invoice look right?" depends on knowledge that lives in someone's head — this supplier always overships by 5%, that one bills carriage separately — no tool arrives knowing that. Someone has to encode it.

The posting step doesn't fit. This is the one that quietly kills projects. A tool that "integrates with Xero" pushes a standard record into standard fields. If you use tracking categories in a particular way, run a bespoke chart of accounts, or need the invoice to land in an ERP alongside stock movements, the standard integration will not do it. That connective work never appears in the sales demo.

The per-document fee overtakes the build. At high volume, subscription pricing crosses over. Do the arithmetic over three years, not one month.

You need it to do more than invoices. Once documents are being read reliably, the same capability handles purchase orders, delivery notes, contracts and expense claims. Tools priced per invoice do not extend cheaply.

The comparison, honestly

That last row matters more than people expect. A custom system is an asset you own and a thing you are responsible for. If nobody in your business owns it, a tool with a support line is genuinely the better answer.

The part that actually decides it

Here is what a decade of integration work teaches you, and what no vendor page says: the automation is rarely the hard part. Your data is.

Industry estimates put data preparation at 60–80% of project budgets, with integration another 40–60%. Those figures overlap and vary, but the direction is consistent and matches what we see: the modelling is cheap, the plumbing is not.

Before you spend anything, check whether these are true:

1. Supplier records are consistent. If "ABC Ltd", "A.B.C. Limited" and "ABC" are three entries in your system, matching will fail no matter what you buy.
2. Someone can state the approval rule in one sentence. If nobody can, you are not automating a process — you are inventing one, and that is a different job.
3. Your accounting system's API is switched on and someone has the credentials. Sounds trivial. Routinely takes two weeks.
4. You know your real invoice volume. Not the estimate. The number.
5. You have a month of real invoices to test against — including the ugly ones you would rather not show anybody. Especially those.

If three or more of those are shaky, fix them first. This is the same failure pattern we wrote about in why most AI automation projects fail before they start: the pilot stalls six weeks in, and the cause is never the AI.

A sensible order to do this in

1. Count. Invoices per month, and roughly how many are awkward.
2. Multiply. Volume × £8–12 is what manual processing costs you now. That is your budget ceiling.
3. Buy the cheapest tool that plausibly fits and run it for a month on real invoices.
4. Write down what it could not do. Be specific: which documents, which step, how often.
5. Decide from that list. If it is short, you are finished — keep the tool. If it is long and all in matching and posting, that list is the specification for a build, and you got it for the price of one month's subscription.

Most businesses stop at step 3, which is the correct outcome. The ones who go further do it knowing exactly what they are paying for, rather than guessing.

That fourth step is worth more than it looks. A build brief written from a month of real failures beats one written from a wishlist, every time.

If you are at step 4 with a list you cannot solve, that is the conversation we are useful for — workflow automation and the custom software that makes it fit your systems is what we build. And if the tool is doing the job, we will tell you to keep it.

Have a look at what you are processing manually, and count it. That number usually makes the decision for you.