Cash Flow Part 1

Episode 6: Cash Flow Part 1

Procure to pay: the money going out

Procure to pay isn't an AP automation problem. Most invoices need investigation that OCR can't do, and Otto follows the thread from order to payment so every check you write is one you can explain.

Procure to Pay · 10 min

Opens December 1, 2026 at 9:00 AM ET

Read the transcript

At the end of the day, it's about money in, money out, right?

How do you think about this famous the procure to pay workflow before AI and after AI?

Yeah. This is a tough one because the toughest people to work with at a contractor is the AP department. It really, really is. And it's not that they're bad people. They just have an impossible job. Yeah.

They just have an impossible job. So, that whole rant that I went on about why POS and subcontracts are all fooled before the project even starts is part of why they tend to be a little bit crustier than the rest of the departments. The reality is that, you know, none of the POS are right and the subcontracts are empty. Change orders come in, they're missing. There's only two ways to make money in construction. Cut what you're spending or bill more. That's it.

Simple. It's really hard to cut what you're spending when you don't know what you've already committed to spend. So when people come to us in the old days, the AP bots or number one seller, right? Highly complex though. There's a million edge cases and the AP person doesn't want to use a new accounting system. That's the biggest difference before and after. A lot of people have bought these like standalone OCR packages to read invoices and then import them into their accounting system.

That only solves 20% of the invoices. 80% of the invoices require investigation. They have to find out what was bought. Why is the purchase amount different? Why is the price different? Why are the units different? It's not that you're doing this work because you want to. You're doing to it because you're blind to what's actually being purchased. So when we think about procure to pay, how much money is leaving your bank account is where how I want to talk about it because you're writing checks and you're paying these people.

And if you're a GC, maybe you have a little bit of protection with pay when paid, which is illegal in a lot of states. So you can't do that. But in the case of material supplies, you're buying them on credit card. You're you're committing to that spend whether there's a purchase order or not. If somebody at your company says I need a,000 widgets and they deliver a,000 widgets and there is proof of delivery, you owe them that money. That's it. Orders or purchase orders be damned.

You buy it, you got to pay it. So the idea here is people have misdiagnosed the procure pay workflow as an AP automation problem.

That's the biggest issue that it's not an AP automation problem. There's that's a portion of it, but it's the golden thread from order, which we talked about previously, to the check being cut or the AC being submitted. That's the real that's the real magic of the AI component of this, capturing that owner's or that order authorization, order acknowledgement through the ticketing system or through, you know, a packing slip or what have you. Having Otto be able to communicate with vendors via email, SMS, having them understand, hey, I got this invoice.

What is this invoice? Instead of Jenny or Johnny being forced to work 12 extra hours a week calling vendors, calling superintendent, using Teams or Slack to find out what was bought, why it was bought, how it was bought. That's the problem with AP. And it sure as hell doesn't get fixed by OCR. It gets fixed by an autonomous platform like Otto finding out for the AP person why it was bought. You cannot fix a missing cost code on a job, a missing commitment or PO number when you're trying to enter the invoice into the accounting system.

You can't fix that. And that's what people think that bots fixed. It doesn't. So they're like, "Well, the bot's not saving me that much time." No, it's not because 80% of your invoices are totally screwed. And that's the issue. So the idea here is that the collection of this information, the confidence factor of where this was purchased, why this was purchased. Think about it. If the PO is written by a machine to itemize all the things that were actually bought, a lot of invoicing would just be done automatically.

If the schedule of values for a subcontractor was kept up to date, G703s would be easy to reconcile, but you don't have time to do that. AI does. And that's the difference with procure to pay being it's not about the end of that cycle, it's about the entire cycle. And that's where AI becomes your purchasing agent, becomes your shop, becomes your inventory management system, becomes your subcontractor management system. Where we failed before in the bot company was after all that great orchestration, we try to shove it in a 50-year-old accounting system or a 30-year-old PM system.

And then we were rated on how quickly that 50-year-old accounting system ran. And they could type, you know, because we were using computer use that AP invoice goes in. But the AP invoice that we received is of a different shape and size than what can be accepted in that system and is missing 80% of what it needs to actually entered into that system. So that's why we replace that system now. That's why that's why ERP is dead. Long live AI ERP because all of that work you don't need to, you know, use computer ease anymore.

Otto has an amazing accounting module. It knows construction. It will do 60 to 70% of the work for you. And then you just run AP in Otto. You don't run it in a 30 50-y old system because it's a warm blanket because that's what's breaking. And I'm going to explain this for the IT folks listening to this. I'm going to explain it a little differently. When we take an invoice, we don't artificially convert the invoice into rows and columns, which is a ma a major issue with accounting systems.

We also don't care what the code is on the tax drop down that you were implemented 50 years ago in CMiC. We don't care what that code is because we just can read. And so the idea here is that the work doesn't need to be in a shape and size that is accepted by an accounting system. The work can stay in its original format as an invoice. Not to get too much into the inside baseball, but again for the IT folks, it's stored in something called JSON, which is a representation of the invoice and data.

It's not forced to be transformed yet. In its original form, it can find the original data. It can communicate with the person who sent it to you in that original data format. That's a very important thing. If Otto is going to email a vendor about what this invoice is, it should be reading the original invoice, not what you shoved into the accounting system. Oh, I don't have a field for that, so I've lost that. That's the key to procure to pay. The idea here is you should not be writing checks obviously unless you know.

But we are writing checks right now where we don't know or we're making some really strong assumptions. We've had clients get defrauded. That is a very high rate right now. AP invoices getting processed for fake vendors. That's going to increase over time with AI. You need the AI. You don't need just a AP system or an AP bot. You need an AI enforcer around your purchasing for if you're writing checks, money going out. And we really I think we're titling this cash flow because the cash flow equation is money out, money in, control the money out using AI to validate what the check should be.

And that's what changes the most. Your folks deal with exceptions, which is 30% of the time, not the things that can be resolved by asking a superintendent or asking a vendor. You deal with the real exceptions. And those real exceptions are where a lot of fee gets eroded. And that's what it gives you. It gives you hours back. And so, think about it. It's ra, you know, in plain English, it's giving you those real exceptions. It's not giving you exceptions because Vista is trying to put it into a closed period and it's raising a hand.

Of course, it supports periods, but it's not about data entry exceptions because who cares about data entry except it's about purchasing exceptions and that's what changes after AI. We hope that this series was genuinely helpful. If it sparked new ideas, if it gave you an inspiration to pursue new AI projects in the business, reach out to us. We're here to help. So we thank you for your time and we thank you for watching.