In plenty of UK businesses, someone spends part of every week copying supplier invoices into the stock or accounts system, line by line. It is slow, dull and easy to get wrong. Owners reasonably ask why, when AI tools can now read documents, this is still happening.
The short answer is that reading an invoice was never the hard part. The hard part is that there is no single way in.
Every supplier sends something different
Picture Example Cash and Carry Limited, an invented wholesaler with around 30 regular suppliers. One sends a PDF with the totals at the top and the line items underneath. Another sends a photo of a paper invoice taken on a phone. A third sends a spreadsheet with product descriptions but no product codes. A fourth changes its layout every few months without telling anyone.
None of these is wrong. Each supplier uses whatever its own system produces. But your stock system expects one format, with your product codes, your units and your prices in the right fields. Someone has to bridge the gap between 30 layouts and one system, and at the moment that someone is a member of staff with two windows open.
That is where the errors come from. A wrong price or a case quantity entered as a single unit slips through, and nobody notices until margins look odd or stock counts stop adding up.
Why the AI tool on your desk does not fix it
Microsoft Copilot can read an invoice you upload and tell you what is on it. That is useful for a one-off question. It does not then type those lines into your stock system for you, match the supplier’s product names to your codes, or flag that a price has changed since last month.
So the general assistant moves the problem rather than solving it. Someone still has to upload each invoice, read the answer and enter the result. The retyping becomes copying and pasting, which is a little faster but has the same weak points.
What a proper fix looks like
The fix is one route in. Every invoice, whatever its format, arrives in the same place and goes through the same steps:
- It is read, whether it is a PDF, a photo or a spreadsheet.
- Each supplier’s layout is mapped to your own fields, including your product codes.
- The result is written into your stock or accounts system.
- Anything the system is unsure of goes to a person to check before it is posted.
That last step matters. Purpose-built invoice tools, such as the invoice model in Microsoft’s AI Builder, give a confidence score for each field they read. A sensible setup posts the confident lines automatically and holds back the uncertain ones for a human to look at. Your staff move from typing everything to checking the exceptions.
The AI is one part of this, and not the biggest one. Most of the work is in the plumbing: deciding where invoices land, building the product code mapping for each supplier, connecting to your systems, and agreeing what counts as “unsure”. Get that route right first and the AI has a reliable job to do. Skip it and you have a clever reader with nowhere to put its answers.
A free check you can do this week
Before talking to anyone about tools, count how many different invoice layouts your team typed by hand last week. Note roughly how long it took and how many corrections were made afterwards.
That number tells you the size of the job. If it is three suppliers and an hour a week, a fix may not be worth building yet. If it is dozens of layouts and a day or two of someone’s time, every week, the case is much clearer, and you will have the figures to judge any proposal against.
The order matters
It is tempting to start with the AI tool, because that is the part being advertised. In practice the route in comes first: one place for invoices to arrive, one mapping to your own codes, one clear rule for what a person checks. Add the AI to that and it has something useful to do.
If supplier invoices are eating your team’s week, our Data Foundations service is where we would start, by building that single route into your systems.