Cash Flow Management Machine Learning: Automating Invoicing and Payments

Megan - 00:00:37: Today, my guest is Carlos Vega. Carlos is the CEO and co-founder of Tesorio. The company was founded in 2013 to address the cashflow disconnect faced by finance teams. Under his leadership, Tesorio's AI-powered platform aggregates real-time data, delivering actionable insights and automating workflows that optimize collections, cash application, and forecasting. Carlos's career spans roles such as co-founder and director at Elemento Factoring and investment banking associate at Lazard. Carlos holds a BA in economics from the University of Pennsylvania and an MBA from the Wharton School of Business. He resides in Panama with his family. Carlos, thank you very much for being my guest on today's episode of CFO Weekly.

Carlos - 00:01:53: Excellent. Thank you for having me.

Megan - 00:01:54: Today, we'll be talking about how AI is transforming the invoicing and payment collection process. And Carlos, I'm excited about the opportunity to learn about you and your thoughts and experiences with this topic. So let's jump right in.

Carlos - 00:02:07: Excellent.

Megan - 00:02:08: First, with your extensive background in finance and your journey from investment banking to founding Tesorio, what inspired you to focus on cash flow performance and the use of AI in financial processes?

Carlos - 00:02:21: Yeah, it's a great question. So going back, I worked about a decade in finance, started off working for the CFO of General Motors Pension Fund, which back when I was in college and right after college was a big fund, $160 billion fund. I then worked in the reinsurance space. And then I was at Lazard doing M&A. While I was at Lazard doing M&A, I had a really interesting idea with a coworker, which was let's start a factoring business, which is an accounts receivable financing product, right? And we said, what we can do, is help people with our knowledge from M&A and Lazard and all of that graduate from a short-term factoring solution to a longer-term. Banking relationship and line of credit. And what was super interesting was seeing how most people actually said, no, thanks. We'll take your money and see you. And for us, it was a little bit disappointing because we thought we're going to be able to help smaller businesses graduate to a more formal relationship. But what ended up happening is that it felt like payday lending for business. And that's what got me curious about this, right? Saying, hey, do you really understand what this means for your cash flow? Or do you understand cash flow? Do you know what we're talking about here? And as much as we tried, it was difficult. So we turned around actually. So when I decided to go to business school and see if we can come up with a solution that would make it easy for companies to make their cash flow predictable and understand it. And this was not to age myself, but I'm 10 years out from business school and been working on this company for about eight, nine years. And AI wasn't as hot as it is right now. It wasn't as readily available. So a lot the AI concepts have come in more recently. We have been using machine learning for some years. But the point is that a lot of financial data goes into understanding your cash flow. And you can use different technologies to have a more granular look, a more bottoms up look at things. And I'd say without getting into more details that we'll probably cover later, that's where I'd say the biggest part was AI and machine learning can help us look at granular data much faster and much greater detail to come up with a more, more accurate and predictable cash flow.

Megan - 00:04:32: And can you share a memorable moment or success story from your time at Tesorio that highlights the impact of your platform on clients' financial management?

Carlos - 00:04:41: Yeah, sure. I'll share three that I really love. One is from one of my favorite customers, a guy named Steven Odell. He used to run collections at Slack. He's now Contentful and owns all of Quote to Cash, which is a really interesting thing. But when he was at Slack, he was able to have some of the best metrics we've seen in the industry on the percent agings over 60 days, well below 3%, and was able to go all the way through to the $27 billion Salesforce acquisition with just two people running collections. A couple of other quick ones like that, a company called UiPath. They were able to go from $100 million to $600 million in ARR and not hire anyone in this space, in this part of their financial operation. And then another company called Redis was able to see some very serious financial gains within just two quarters of implementing the product. Those are a couple that come to mind.

Megan - 00:05:32: Yeah, those are great stories. So can you share a little bit about Tesorio, like a little bit about what it is that you guys do, who your target client is?

Carlos - 00:05:44: Yeah, excellent. Yeah, shame on me. I introduced myself and not the company, so I'm just at my risk. Yeah, so what we do, we call it connected financial operations, right? And essentially, what we're trying to automate is all the run the business parts that a lot of different people in the financial team take care of, right? And this is a mix of finance folks, a mix of accounting folks. And so imagine everything after you hit closed one, actually. I mentioned him earlier, Steven Odell said, closed one is not done because revenue ain't real till you get paid. And so after closed one, a lot of things have to happen to make sure you're generating the right invoice, delivering the invoice either through the portal or via email, following up with your customer to make sure that they pay you on time, making sure you receive the payment, making sure you reconcile the payment, and then looking at, okay, based on how much came in, how much am I able to pay out? All of that stuff, they're like baton handoffs that go from, one person to the next. And sometimes there's people outside of the finance team that have to get involved. All of that is what a lot of folks call financial operations. And that's separate from what you typically think of accounting, which is compliance, audit, risk. All of those areas are slightly just different from these workflows. And so what we're trying to do is automate the financial operations. And if that can be strung together through integrations to the right tools, through integrations to the right processes, integrations to the right financial rails, you can actually get a really strong view that's real time and updated of your cash flow. And so if we can automate all the financial operations workflows, the tactical workflows, you can get a pretty automated 13 week cash flow view, which helps you make decisions more intelligently. So that's what we do. In summary, it's connecting all the financial operations to get you a clear cash flow.

Megan - 00:07:36: And you mentioned AI gaining momentum. We all know that's happening. So how is it that you see AI fundamentally changing the invoicing and payment collection landscape? And what specific pain points in traditional processes is it addressing?

Carlos - 00:07:52: Just to use a few examples, right? More generally, I would say there's three areas that we think about it. A lot of people say data is the new oil. But what I counter to that is like, yeah, but petroleum is pretty useless. So what are you going to do with the data? And I think AI allows you more generally to do a couple of things really well. Number one is clean up existing data, right? And so in this process that we're talking about, think about deal is closed one. A sales rep goes into Salesforce really quickly and adds in the amount, adds in the address, adds in the purchase order number, adds in the contact, and et cetera, et cetera, et cetera. We're finding that 50 to 70% of payment delays are actually due to that data, that origin point of the data being input actually being incorrect. And that's an own goal, right? It's a self-inflicted wound. So can you use AI, for example, to look at a PDF of a contract that was signed, which had teams of lawyers on both sides, thousands of dollars spent on both sides to make sure every T was crossed, every I was dotted, and extract information from that to either, one, directly create the invoice, or two, validate that the information that was put on the invoice is accurate and is matching. That's a really cool use of AI. I won't go into the other examples because I know there's a lot of questions you want to go to, but the two other use cases, one is looking at primary source data. So in that example, something like order form is a primary source, right? It hasn't been. It wasn't a human typing something in, having potential mistakes. And then the third example is using granular data, which we talked about earlier. Imagine you're forecasting, right? And typically you just look at sales and you apply DSO and you're good to go. Well, what if you could look at every single invoice and every single payment made by that customer on all the past invoices and train a model on that information? And now you're looking at all the open invoices for that customer, predicting when they're going to get paid, but doing that across every single one of your customers, right? So that's what I mean by when I say, first off, cleaning up existing data, second, using primary source for new data, and third, using granular data to make cash flow more predictable and streamline the invoicing process.

Megan - 00:10:02: And I'm sure you just touched on this a bit, but what role does real-time data play in improving cash flow management? And how is it that Tesorio leverages AI to provide actionable insights for finance teams?

Carlos - 00:10:15: Yeah, that's a really good point. I'd say there's a couple of different areas. I'll use just one example, right? We've all dealt with portals. And actually, one of the big procurement portal providers is one of our customers. And they're great tools in streamlining the process for the corporation that's using them to manage their budget, manage the purchase order process, manage the payments process in a lot of cases. On the vendor side, right, so if you're the one that has to go in and input the invoice, that can become burdensome. If all of your customers, we're finding somewhere around 20% to 30% of a typical company's customer base is using some type of portal. The benefit, though, is if your invoice gets received and approved, you're going to get paid like clockwork on the net terms, right? So, again, there's pros and cons of everything. However, there's a difference between the invoice date, that gets approved. And the right to bill date, which is something that doesn't often get measured. And so that's an example of a place where real-time information can be really helpful for cash flow. For example, let's say you submit an invoice into one of the portals and it has a dispute or it gets rejected or something like that occurs, right? Or there's some issue with the line item. The clock on the net terms doesn't start until that invoice is perfect, right? So what if you can use AI, for example, to process all the notifications coming from across all the different portals that you're getting into your shared inbox and automatically mark if something was approved, automatically update your forecast or predicted pay date based on the net terms plus the approval date, right? Or if something was disputed, you can summarize that email into a tweet size summary, put it on the note of the invoice in your product and in the ERP. And then you know who that invoice was assigned to. So you immediately create a task for the person assigned to that invoice to go in and correct it. That's a really efficient use of real time information to make sure you're making your cash flow move more efficiently. Hopefully that helps. Happy to expand on other examples if you think others would be more relevant.

Megan - 00:12:22: Yeah, no, that's great. I'm curious to know, is this a product that just sits on top of an ERP? How does it work as far as implementing this solution?

Carlos - 00:12:32: Yeah, so it does sit on top of an ERP. So we're plugging in. I mentioned earlier financial operations. We're getting asked to plug in not just ERPs, but to billing systems. Some companies have a homegrown invoicing, invoice generation system. CRMs is a very common ask. Slack, email systems, like whether it's Gmail or Outlook 365, and more and more systems getting requested. Everything I just mentioned, we already do, by the way, but there are many others that get requested. Now we're seeing an interesting trend. One of our teammates calls it, she says collections is a team sport. And we're seeing chief customer officers come and approach us and wanting us to, one, they're the ones interested in our product because in this market, farming from your existing customers is a lot easier than hunting for new logos. So that relationship at the point of payment is something that's bubbling up now to the chief customer officer organization. And so they're asking for, hey, can you connect to Zendesk? Can you connect to Gainsight? Can you connect to our customer intercom? A lot of folks like that. So it's interesting to see the number of systems we're being asked to connect to. But yes, we do not replace an ERP. We work very closely with the ERP.

Megan - 00:13:41: And I'm just going to piggyback a little bit on your answer to that question and skip ahead a few. But given the importance of customer relationships and finance, how does AI ensure more personalized approach to payment collection while still maintaining efficiency? How does it improve the customer experience?

Carlos - 00:14:01: Yeah, so for the customer, I think it boils down to customer segmentation. Which is, again, it's very ironic that until pretty recently, this part of the customer experience was viewed as just some back office process, right? But if you had your marketing team send the same exact message to every single person that visited your website or to every single customer during customer marketing, they'd probably lose their job, right? And then here, that's how most of these processes were handled before, right? It was just a step in the ERP or some other system that said, 15 days late, hey, you're late. 30 days late, you're very late. 60 days late, hey, you're going to get shut down. 180 days late, you're getting shut down. Thank you, bye, right? However, we've all got different customers. There's long-tail customers. There's customers that are consistently on time. There are customers that are consistently late. There's high-priority customers and things like that. So using AI, you can automate the customer segmentation. You can also automate the messages or personalization of the messages that go out to each customer. Which, if you're on the receiving end, right, and someone's reaching out to you, it's with context. Or if you're a good payer, no one should be following up. If you happen to be, I don't know, someone was on vacation, you happen to be seven, 10 days late, but you've always paid every invoice consistently on time. How are you going to feel if someone's like, hey, you're late, pay me? I think you're going to say, hey, I've always paid you on time. Like, why are you nagging me like this? So that's a big part of the customer relationship. Another part, though, that's really exciting, tying to the previous question, we're seeing close to 40% of our users are actually in account management and sales. And so if you're using smart workflows to orchestrate the communication across the teams in your own company, right, for the end customer, having your account manager be the one that reaches out and says, hey, I noticed we haven't received your payment. Is everything okay? Right? It was a very different experience than having someone kind of calling to break your knees with a lead pipe, if you will, right? And so those are, I think, very big influences of not just purely AI, but AI embedded into smart workflows that can improve the customer relationship.

Megan - 00:16:14: And you may have actually just answered this one, but in what ways is it that this platform enhances cross-functional collaboration within organizations? You mentioned collections being a team sport. So can you expand on that a little?

Carlos - 00:16:28: Yeah, totally. So funny examples we hear, right? Recently, someone was saying like, look, a CCO who reached out recently interested in our product. When they do, this is fairly new for us to be very clear. It's something we've talked about for a long time that this is an important step in the customer experience journey. Actually, something we say internally, it's that it's the last first impression you make, right? Because sales, I do sales a lot. Of course, you can imagine put on, even though we're fully remote, put on a polo shirt, make sure everyone's looking sharp, all our presentations are great and all that sort of stuff. And customer success is showering your customers with love. And then all of a sudden finance shows up and beating people up. It's not a great and consistent story, right? So making sure that stays aligned is really critical. And so one of the things we're seeing, one of the stories I heard recently was a customer wasn't paying. Someone from the finance or accounting team followed up a little bit more stern with the customer. And then the account manager called up and said, do you know why they're not paying? Because I'm in the middle of increasing their contract by 50%. They're signing a new deal. They're going to pay us more money. That's why they haven't paid that old smaller invoice. And that was a lack of just basic collaboration or coordination. And so that's an important way, right? Not to get into the details of it, but we don't charge per user. And the point is so that people in those other teams can join the product and benefit from those insights, right? Another thing that's really important on this cross-functional collaboration, just a product pillar for us is meet users where they work, right? So yes, you asked about the ERP earlier and our product is, I think it's beautiful, but you know, everyone's entitled to their opinion. It's nice to work in, but account managers and sales folks live in Salesforce, for example, right? So one of the things we're doing is we're pushing Tesorio information into the account or the opportunity in Salesforce. So those other teams don't have to leave, but they have financial visibility into this health of the customer. So that's another way that, that it helps people collaborate. And the last thing is, for example, this is a conditional situation where based on a certain condition, the account can actually be automatically handed over to the account manager or the sales rep based on different rules that the users can easily configure to have them reach out and follow up, right? In order to address any delays on cash.

Megan - 00:18:48: And as companies increasingly adopt AI for invoicing and collections or really anything in general, what is it that you consider the most significant challenge or barrier to implementation?

Carlos - 00:19:00: Yeah, that's a really important thing to keep in mind when developing software for this department. Something that's also important that we talk about a lot is what's the cost of being wrong? When you're going to use AI. For example, if you're going to write a marketing blog post and it doesn't sound quite right, you can always step right into Google Docs and fix it, right? Or Word or whatever tool you're using. If you're going to reconcile a payment or if you're going to send a note to someone to make sure they pay you, you only get one shot, right? Or yes, you could go and reverse payment reconciliation. But if you already assumed something was paid and you're operating under that assumption, well, you might have missed an opportunity to bring in some cash, right? Or if you send a note incorrectly to someone or if the note is not providing the right information, you could have an issue to deal with down the line. So one of the things that's really important and one of the challenges is how do you know when something is an exception that needs a human eye versus when it's something that you can just allow the automation to carry out. And so for us, what we do when we develop tools like this is keeping in mind that in finance, the cost of being wrong is typically high. We just, not to get a little Reagan on us, but trust but verify is what we follow, right? So anytime, for example, earlier, I mentioned automatically reading emails using AI to extract the status of an approval from a Coupa or an Ariba email that you received, right? When you hover on that, just with your mouse, we show you an image of the original email. In case you want to, double check it and read it with your own eyes and make sure that yes, this was in fact approved. And this is in fact a good prediction of the payment date or if they're, that's an example, right? So giving you the ability to always look inside instead of it being a black box, I think is how I would summarize all of this.

Megan - 00:20:55: And with the rise of AI-driven solutions, how is it that you anticipate the relationship between finance teams and technology evolving? And what new skills are going to be needed in the coming years?

Carlos - 00:21:07: Yeah, it's really cool to see this actually happening before our eyes. More and more, we're seeing in large companies for a while, there's been finance operations or FinOps has been something very associated with cloud spend and things along those lines, right? With AppTO and other companies who is a customer, by the way, kind of spearheading that. But now we're seeing this new role, financial operations emerging more and more. And this is this. These are not folks from IT. These are CPAs, MBAs, undergrads in finance and accounting, former controllers, former VPs of finance, former senior revenue managers. It's exciting to see this role come up more and more. As people who are experts in both systems and what finance folks have to deliver for their companies to drive efficiencies. And that's, I think it's something we're going to see more and more. It's like a CFO secret weapon almost because we've already covered a few systems, right? Talking about procurement, ERP, CRMs, customer success tools, email, but there are so many more. There's so many categories and all of them have information that can be used to understand the health of the company and your cashflow. And so more and more, I'd see that as something that's critical, mainly because AI underlying it all, AI has just a bunch of models or very complex models that we don't really understand necessarily, but it's still a model, right? And every model, as we know, I think we all know the GIGO rule, right? Garbage in, garbage out. And so if your data isn't cleaned, if your data isn't transformed properly, if the business logic isn't correct, whatever AI tools or whatever automation tools you have can't do their job. And so you need expertise in the middle there to make sure that all of that's flowing properly. I don't know if I'd go as far as to get technically into the data pipeline, but it is getting involved in the data pipeline to some extent to make sure that these processes work efficiently and the tools work efficiently. So that's where I see a role emerging already. Shutterstock, another customer, for example, recently, they have a whole department. The person who signed the contract was a vice president of financial operations, which is really exciting to see. So this is definitely a trend that's emerging, starting to happen. I think we'll see that happen more and more.

Megan - 00:23:27: And how is it that you envision the future of invoicing and payment collection in the next three to five years as AI technology continues to evolve? Where are you guys placing your R&D dollars?

Carlos - 00:23:39: It's a good one. If you abstract away, it's kind of funny that AR and AP still exist, right? Because you asked about ERP earlier. In our accounting system, we have everything. It's obviously all electronic data. And on the AP side, whoever our customers are and we're sending an invoice to. They're going to take and turn whatever we send them into data as well, right? And so if you just really zoom out, it's kind of funny that in today's world, what we're doing is we're exporting from my system to a PDF that I can then email to you that you're going to take that PDF and import it into your system and make it electronic again. And then when you send me the payment, you're going to send everything over email or other data, not electronically. And so I'm not making a case here for blockchain or any other distributed ledger or anything like that. But I think as integrations become much more streamlined, which is something that AI, I think, will enable, you could end up in a place where it's no longer AR and AP going back and forth, but it's companies setting their terms and the conditions under which they're willing to change their terms. And there being optimization across those conditions and those terms. So basically set another way, there is no more AR and AP, and it's actually companies doing business and just communicating without the system to PDF, to email, to PDF, to system. It's just system to system.

Megan - 00:25:05: Yeah, that's really interesting. And yeah, 10 years ago, I thought blockchain was going to make a bigger impact than it seems to have. And yeah, I just wonder why it hasn't made the big impact everyone was expecting.

Carlos - 00:25:19: I have a theory on that.

Megan - 00:25:20: I would love to hear it.

Carlos - 00:25:21: We can borrow from the EDI experience. Let's go back to the 70s, right? Like EDI was supposed to solve all of this as well. But what ended up happening is you need tools like SPS Commerce and others because everyone made their own EDI and everyone wanted to own their own blockchain.

Megan - 00:25:34: Yeah.

Carlos - 00:25:35: And so I think that's what happened. Like in order for the promise of the blockchain was a shared ledger, right? But if everyone owns their own, if everyone wants to own the shared ledger, is it really shared anymore? And so how do they all talk across each other? I think that's a difficult part.

Megan - 00:25:48: Yeah.

Carlos - 00:25:49: So it's like a chicken or the egg problem. I think build.com did a pretty cool job, right? With their own internal network, for example, and everything there to get paid and to send payment. But maybe it's going to come down to open source technology or using AI to drive integrations really efficiently, right? Like you can't, I don't think this is five, 10 years out. I think this is in the two to five year range. You could easily imagine, right? If you look at API docs today for any company, it tells you exactly how to speak to the API in human language. You can use an LLM potentially soon, I hope, to read an API doc and create at least an initial version of an integration to almost any system. And once you can do that, then you can start to get to a place where systems can directly communicate. I don't think that's that far away. It sounds a little crazy, but I don't think it's crazy.

Megan - 00:26:39: Yeah, no, I would agree. I think it's closer than anyone can imagine.

Carlos - 00:26:44: Yeah.

Megan - 00:26:44: And finally, what advice would you give to CFOs who are hesitant to implement AI solutions?

Carlos - 00:26:50: We talked about it earlier, but I would dig in and say like, hey, can you show me how I get to peek inside the black box? Is that something I would say is critical right now? The other thing is, this is just a personal thing, personal opinion, but I'd say be wary of companies masquerading as being AI companies when really what is the value that they're driving? Similar to, you just brought up blockchain, right? Like how many companies were out there that they were basically solutions in search problems, right? Instead of problems, with problems being solved by a specific solution. With AI, it's a lot of the same. Like what is at the end of the day, if you take a step back, what is the objective that I need to hit? And, is this tool going to help me achieve that objective? Does it use AI? Yes, great. Okay, there's a lot of things today that can be made more efficient with AI. Perfect. Is it, is someone just selling me like the dream of like, hey, come buy my AI tool because you need AI. Like, well. Help me understand, again, what objective I'm going to hit. So I guess I rambled a little bit, but I'd summarize it as those two things. Like, one is, can you help me understand and look inside what's happening just to make sure it's not a black box and I can trust but verify? And the second thing is, what objective are you going to help me achieve? Is your solution built to hit an objective or are you just trying to wow me with some cool AI wizardry? That's what I would say would be my advice.

Megan - 00:28:17: Great advice. And Carlos, thank you so much for being my guest today.

Carlos - 00:28:20: Excellent. Thank you for having me. I really enjoyed the conversation.

Megan - 00:28:23: Yeah, I really enjoyed speaking with you as well. And thanks for finding the time to be here with us today to share your experience and knowledge. And I wish you and Tesorio all the best.

Carlos - 00:28:32: Thank you so much. You too.

Megan - 00:28:33: And to our listeners, please tune in next week. And until then, take care.