AI Invoice Scanning ROI Calculator

Supplier invoices cost you twice. Once in the time somebody spends typing them into your system, and again in the lines that get billed above the price you agreed and paid without anyone noticing. This calculator puts a number on both from four things you already know, shows you which half is bigger, and prints the working so you can argue with any of it.

1. Invoice Automation Is Usually Sold On The Wrong Number

Almost every invoice automation pitch leads with hours saved. It's the easy number to model and the smaller one to collect. The prices on those invoices are where the money actually sits.

Typing an invoice in is real work and it does go away. But one invoice is a few minutes of admin and a few hundred in purchasing, and only one of those two scales with how much you buy. That's why this calculator asks for your ingredient spend as well as your invoice count.

Two things get left out of most invoice business cases:

• The prices nobody checks: across 6.43 million supplier invoice lines, suppliers billed 6.6% of them above the price the restaurant had agreed, against 3.9% below. The tilt runs against the buyer, and nearly half the lines that run over are more than 20% over.
• The work that stays behind: extraction removes the typing. It doesn't remove checking the delivery against the order, approving it, or chasing the supplier when something is wrong. A business case built on the whole task vanishing won't survive its first month.

Volume cuts the other way. A group taking 30 invoices a week has far less to gain from the admin than a group taking 300, but both are exposed on every line they buy. One half scales with paperwork. The other scales with spend.

2. Where The Money Actually Leaks

A wrong price on an invoice is almost never obvious. Most lines are correct, which is exactly why the incorrect ones get paid. It isn't spread evenly either. Overbilling lands hardest on fresh produce, seafood and meat, where supplier prices move fastest and an agreed rate goes out of date within days of being set. Those are the lines to check first, and they're the ones a person scanning a long invoice is least likely to catch, because each one is small and there are dozens of them.

3. The Maths, Step By Step

A. The typing. Invoices a week multiplied by the minutes it takes to type one in, divided by 60. That gives the hours a week your team spends keying invoices. Only the typing belongs here, because that's the only part extraction takes off you.

B. What an hour costs you. Annual salary plus employer costs, divided by your working hours a year. Put an hourly rate in directly if that's the number you hold. Employer costs are the part people leave out, and on a salary they move the answer by a tenth or more.

C. Money at risk. Your ingredient spend for a year multiplied by 6.6%, the share of lines billed above the agreed price, multiplied by how far over a line runs when it does. The 6.6% counts lines and we apply it to spend, so read the result as an order of magnitude rather than an invoice you can go and reclaim.

D. The total. The two halves added together. The split shows as a bar, because which half is bigger tells you where to put your attention. For most groups it isn't the admin.

4. Where The Number Usually Goes Wrong

Open the working before you quote the total. Every step prints with the arithmetic beside it, and all three assumptions are yours to change.
The causes split into two groups. Some change the answer. Others only make it look wrong.

Real causes, roughly in the order they turn out to be the answer:

• Counting the checking as well as the typing: put your whole invoice handling time in the box rather than the keying alone and the time half gets too big, then the business case falls over the first time somebody times it properly. Put in the typing only.

• Using all supplier spend, not just ingredients: the 6.6% comes from food and drink invoice lines. Packaging, cleaning and equipment behave differently and aren't in that base, so leaving them out keeps the figure conservative and easier to defend.
• Treating money at risk as money recovered: flagging a line isn't the same as getting the money back. The supplier has to accept the credit note, and some differences turn out to be a price rise you agreed to and never recorded. The figure is the size of what you aren't watching.

Things that look like a saving but aren't:

• Hours freed that nobody redeploys: if the person who typed the invoices stays on the same salary doing the same shifts, you've freed capacity rather than saved cash. Worth having, but don't put it in front of a finance director as a cost reduction.
• A credit note nobody raises: a flagged line that nobody acts on saves nothing at all. This number only turns into money if somebody works the exceptions each week.
• An accuracy percentage from a vendor: extraction accuracy gets quoted a lot and measured almost never. Ask how it was measured, on how many invoices, and whether a human fixed anything before it counted. We don't quote one here, for that reason.

5. The Typing Is The Half You Can See

Hours are easy to picture, which is why every invoice automation pitch leads with them. Prices are harder to picture and usually worth more. If you take one thing from this calculator, take the split. At most groups the invoices themselves carry the bigger number, and nobody is watching them. There's more on where that comes from in The Spread, our study of 6.43 million invoice lines.