Workflow
The Automation I Didn't Build
Someone checking supplier invoices line by line will get bored, and bored people miss things. The obvious answer is to automate it. But automation only catches half of what goes wrong on an invoice, and it's not the half that costs you. Here's what we did instead.
A client of mine had a person whose job included checking supplier invoices. Line by line. Look, tick, look, tick. So much money every year in supplier spend moving through that process.
Any intelligent person doing that is going to get bored eventually. When you get bored, you miss things.
The idea I walked in with
Claude doing a sweep every hour, reading the invoices, checking pricing against what was agreed, flagging anything that looked off, chasing the supplier when it didn't line up.
Then we sat down as a group of senior leaders and talked it through.
And we didn't build it.
Two ways an invoice goes wrong
Only one of them is a pricing problem.
The first one's easy enough. The price per item is wrong. A system can catch that, you've got agreed pricing, you compare, you flag the difference.
The second one is the goods themselves. They deliver ten extra units and they charge you for ten extra units. Every line on that invoice is correct. The accounts person has no way of picking it up, because on paper it's perfect.
And an automated reconciliation can't really pick it up either. It could go back and check what was ordered, sure. But things get changed in flight. Sometimes those extra units were wanted, someone on site rang up and asked for them.
❝ The only people who know what actually landed are the people who were standing there when it landed.
Where the checking went
The blokes on site check deliveries off as they arrive. By the time it reaches accounts, the pricing lines up and site has confirmed what turned up, and it can be approved.
We didn't automate it. We took a lot of the manual work out of it, so there's far less room for error, and enough control left that someone can still make a change when they need to.
I don't walk into a business trying to sell a solution. It's not "here's an automation to fix your problem." Sometimes you get in there, look at where the errors are actually being made, and it turns out the answer is a process change and someone with a clipboard.
The idea I walked in with wasn't a bad one. It just would have caught half the problem and made everyone feel like the whole thing was covered.
