Nothing in this post has been built. It’s how I’d approach a problem a grocer’s finance lead described to me.
Where this came from
I ran a discovery checklist with the finance lead at a four-store natural grocer in Nashville. This isn’t a shop that needs AI explained to it. He had already built Claude projects and a reporting pipeline out of NetSuite. The stores run ECRS Catapult for point of sale, online orders, scales, auto-ordering and shrink worksheets.
His answers named a few pains. The one I’d build for first: when a delivery shows up short or damaged, the store asks the vendor for a credit. Following up on it is cumbersome, so some credits never arrive.
Where Catapult stops
Catapult knows what was ordered and what was sold. Chasing a vendor for a missing credit is still done by a person, by hand.
How I’d build it
- Log the credit when it’s promised. From the receiving record, the AP inbox or a short form at the back door. Vendor, item, amount, date, who asked.
- Watch for it to post. Check the vendor’s statements and invoices for the credit. When it shows up, close it.
- Draft the follow-up. When a credit is late, the email to the vendor is written and waiting. A person approves it before it goes.
- Track the deadline. If a vendor has a claim window, the agent flags the credit before it closes.
- Report the number. Each month: what’s outstanding, by vendor, in dollars, and what came back.
What stays with a person
Anything sent to a vendor. The agent logs, matches and drafts. A person decides. You can loosen that later, one vendor at a time.
What I took from it
Small businesses don’t need AI introduced. They need the gap closed between a tool existing and people using it. A credit-recovery agent wouldn’t replace Catapult or NetSuite. It would take over the follow-up that’s done by hand today.
The same pattern came up as expense reconciliation in a conversation with a brewery. I wrote that one up in Expense reconciliation that isn’t manual.