PredictAP

AI invoice coding for real estate accounts payable

Fintech

Payments

Artificial Intelligence

Hi Product Hunt, I'm David, co-founder of PredictAP. This started because I kept seeing the same thing at real estate finance teams. They'd bought AP automation, implemented Yardi PayScan or Nexus, gotten routing and approvals working well, and their AP specialists were still spending most of the day typing. The automation had handled everything around the hard part and left the hard part alone. The hard part is coding. Every invoice is a stack of small decisions. Which property, which entity, which GL account, which cost center, and how to split a line across all of them. None of those are complicated on their own. They're hard because they depend on how your specific organization does things, and that knowledge usually lives in one person who's been doing it for eleven years and is the reason month-end closes on time. There are good general purpose AP platforms out there, and if you're a manufacturer or an agency, one of them is probably the right answer. Real estate is different in a way that doesn't show up in a demo. You're not coding to a flat chart of accounts. You're coding to a property, inside an entity, against a GL account, split across cost centers, and the same landscaping invoice gets handled one way at a stabilized office asset and another way at a mixed-use property where part of it is recoverable. A platform built to work everywhere has to stay shallow enough to work everywhere. We only had to get real estate right. So we built PredictAP to make the coding decisions, not just read the invoice. You forward an invoice to one address. It gets read against your own posted history, coded at the line level, and comes back as an approval-ready record in about 45 seconds, synced into the AP platform you already have. On how our thinking changed. We started out treating this like a document understanding problem, and that was wrong. Reading the invoice turned out to be the easy half. The real work was learning how a given customer codes, and what we eventually realized is that the training data was already sitting there. Every firm has years of posted invoices in their ERP. That's a labeled record of exactly how they do it. Once we stopped trying to build something that understood real estate accounting in general and started building something that learns one organization's conventions, accuracy went from interesting to actually deployable. We were also wrong early on about how much automation people wanted. Teams don't want a black box, and they shouldn't. Anything unusual routes to a person. I think that's a big part of why teams trust it. It's honest about what it isn't sure of. Some questions we get a lot: How is this different from OCR? OCR pulls text off the page and hands you a half-filled record. PredictAP makes the actual coding decisions and gives you a complete invoice that's ready for approval. Why not just use a general AP automation platform? Plenty of teams do, and they work fine for routing and approvals. The gap shows up in coding. Industry-agnostic tools don't model the property, entity, GL, cost center structure that real estate accounting runs on, so someone on your team ends up finishing the job by hand. That's the part we built for. Does it replace Yardi PayScan or Nexus? No. You need an existing AP automation platform. We sit upstream of it. We're the piece that makes the platform you already paid for pay off.