[00:00:00] Speaker A: AI is of course very good at starting certain projects and with most complex problems, the issue when you start working on them is that you don't know who you talk to. For most people, AI is still just this chat box that you ask the question like it's a Google search replacement. But it's much more than that. And my answer is always the same, which is that focus is absolutely key. You can do things that were unthinkable before and people are not aware of this and they haven't properly updated their vision.
[00:00:30] Speaker B: Welcome to The Entrepreneur's Logbook Podcast. I'm your host Zach Minard. You can find me on social at Zack B. In each episode I bring on experts from various industries for you to learn about their strategies and insights driving externally business growth. Today we're joined by Alberto Jimeno, co founder and CEO of InvoFox, a document parsing and data extraction API that helps software companies turn unstructured documents like invoices, mortgage application and insurance claim. Alberto, funnily enough, is a Spanish mathematician and computer scientist who spent nearly a decade building companies with his same founding team in Madrid before stumbling onto the document parsing problem while he was consulting for a client who just could not find a tool that actually worked for once. He launched Infofox in 2022, went through Y Cobbler's Summer's 2022 batch, and has since raised $11.5 million, relocated a company from Madrid to San Francisco and grown into the fox to serve 150 plus companies across the U.S. india, Europe and Latin America processing north of 100 million documents a year. Alberto, it's great to have you on the show. Welcome aboard. Walt.
[00:01:34] Speaker A: Thank you, thank you. It's really, really nice to be here.
[00:01:38] Speaker B: Well, one of the things I always like to ask to any guest to come on this show, Alberto, is it is a business, it is an entrepreneurship podcast and to put it simply, I mean you've done quite a few things. I was looking through your LinkedIn and we talked earlier and you had like a book rental services and stuff like that that you did earlier in your entrepreneur journey, which I always thought it was like interesting. Like my girlfriend's in like university. I was like I should try to like sell books to like college students, stuff like that. Like I definitely make a lot of money doing it. So it, it seems that you've had your, your venture in that space a little bit. But one, one of the things I'd love to, to understand from you and hear from you is if you had to like restart, I guess your entrepreneurial journey or even like Invo Fox as of today. What's maybe the one thing that you feel you would do differently that a lot of entrepreneurs get wrong or similar there?
[00:02:24] Speaker A: For sure. And this is a question that I get asked often by younger founders that are in the early stages of their journeys. And my answer is always the same, which is that focus is absolutely key. And we have made a lot of mistakes, for sure with many of the previous businesses that we built before inbofox, but definitely with Invofox as well. And whenever growth has stalled or things have not gone very well for us, it was because of the lack of focus.
And that sounds easy enough to address, right? Just don't get distracted. But it really isn't because being focused means saying no to a lot of things and doing that every day. So saying no to easy revenue, you know, a potentially large client or a great logo that reaches out and they ask for something that is, you know, adjacent but not quite what you do. Or maybe you get some feature requests from some of your most important clients.
So being able to distinguish when something makes sense and you should do it's not saying no to everything, of course, but being true to your vision and to the mission of the company and being able to build a strategic roadmap and sticking to it, that's the difference between success and failure in my experience. And I would say that especially in the early stages of any business building, it's really important to start extremely narrow because that allows you to be extremely sharp and it's easier to become excellent at just one thing. And it's also easier to build momentum that way. And then you can always expand later. Once you have critical mass, you have certain revenues, you have a team you can, you know, leverage and delegate some things to, so it's easier to grow them. The, the, the approach at some point, if you do it too soon, it will kill your company. I've seen it and that we've been grow, we've been close to, to, to that a number of times.
So yeah, that is my, my, my take.
[00:04:32] Speaker B: I love what you mentioned about like, focus though. And I feel like the way that you're saying is not just like focusing, like being productive. It's more so about focusing on the things that actually move the needles, like things that are actually like, driving the business forward because there's so many things in the day that you could be focusing on but that are not going to actually do, like, anything. I mean, a good example, maybe I have a feature request at dark mode on your software.
It sounds cool in theory, but is it actually driving the needle for to actually revenue get more class or anything?
Not really. And I'm assuming you've probably learned quite a few things going through all these motions over the years.
[00:05:09] Speaker A: For sure. The thing is that when you get sidetracked, it's not only that your energy dissipates and you invest it in something that is not the most conducive thing to you growing your business, which of course is an important concern, but it's also that complexity scales and it does so very fast. Entropy is a thing and you have to fight it actively. So if you start doing things that are outside of your core competency, you will start making mistakes that compile things. Like if you're in the software industry like we are, you will add a lot of complexity to your code base. You will generate technical debt and that will make you slower in the long run. And you will have to pay that technical debt at some point. Or if you decide that you're going to explore this other use case or this other ICP because it looks cool and sexy and it's always, you know, fun to do a new thing, even if it seems adjacent, then you start doing that. And maybe you hire one or two people to do that and bring in people into the team to do something that is not mature with clear processes, that is a recipe for chaos. Bringing people into an organization is super complex because it's not just we go from, I don't know, 10 to 11, it's that human relations work as kind of a factorial function which everybody relates to everybody else.
So your, the, the, the complexity of the organization, of the processes of everything that you have to do goes through the roof. So you have to be very selective. At least that is my, my experience.
[00:06:42] Speaker B: Yeah, like focusing on the right things, but also focusing on the things that are actually in your core competency instead of just trying to do everything at once. So it's good to be a jack of all trade, don't get me wrong. But you should be spending your time on things that or your area of genius.
[00:06:57] Speaker A: You can be a jack of all trades and still be focused. For example, in our case. So we do this very niche, very specific problem which is transforming unstructured data from documents into structured data. That is something you can leverage in your business process.
And you can bring a lot of different abilities to the successful execution of, of this, you know, successfully solving this problem. You can bring mark expertise in marketing in. I don't know in sales, in product, in technology you can do like an exciting different campaign with user generated content. You can do a lot of things but they should be all focused on the same target.
And I would say that it's, it's good to have a clear goal like whatever stage your company is, if you just started, maybe it's getting your first client or maybe get into the first million in revenue, getting to 10 million in revenue, whatever that is. I think it's important to have that very clear to communicate it very clearly to all of the of the team and ask yourself the question which is, is what I'm doing right now the thing that you know the shortest path, that goal that I set for the company and if it isn't the don't
[00:08:16] Speaker B: do it, focus on something else. What are the things that that's the shortest path. I love that I want to shift gears like a little bit because I know we talked a little bit about like Invo Fox and everything like that, but for someone that's like never heard of Invo Fox or they're not too sure like what the concept is. I'd love for you to like somewhat explain to the audience what actually happens when you have a document, for example, one of your clients document that comes in the door. Like how do you guys turn something, for example, like, like a mortgage application. We'll use that as an example as a file for though for sure.
[00:08:46] Speaker A: Maybe for the audience. Let me start by better explaining the problem that we solved because sure, we are talking parsing and data extraction solution. That is indeed the solution. That is not the problem. No. Right. So what is the problem? The problem is that a lot of the data and the context on which business operates is what we call unstructured documents. Imagine a PDF, you know, that is not something that you can load onto your database and power a workflow on. It's a PDF or the same with a spreadsheet, right? So what we do is that we bridge that gap. So for example, if you, I don't know, receive invoices from your vendors, you have to pay those vendors. Maybe you want to manage your inventory. So we go from the PDF invoice to a fully structured information that you can go into your database. Or following with the example that you just mentioned, Zach, when we work with companies in the mortgage industry, mortgage applications are quite interesting because it's a use case that was actually not possible to solve, at least not to the extent that it is today, just a few years ago, because it has a lot of different challenges that kind of compound this. You know, we're talking about many a few thousand pages in each individual application. There are different documents, they come all together in the same file. So you have to make sense of all of that.
It's quite challenging. So instead of having people going through it by hand, or maybe having some sort of semi automation that then has to be supervised by people again, we can leverage AI and do a lot of really interesting things to fully streamline and automate these processes.
And you were asking about the technologies and basically what happens under the hood. Right. So in our case, what we do is that we are what we like to call technology agnostic. So whatever is best in the market. And as new technologies and solutions come out, we will incorporate them into our pipelines, into a way of solving this problem and bring them to our clients. So right now we basically combine a mix of very different technologies, from traditional OCR for sure, to more advanced layout, aware OCR that is basically able to understand the topology or the distribution of information in any document.
Also we have our own machine learning models that we build and train in house. And of course, all of the different LLMs and VLMs are now available in the market, both open source and closed source. And the key thing, the key component and where we bring value is to understand how to best combine these technologies.
Which tools you should use for each individual client and use case, what the goal for each of those tools should be, in what order you should use them, and that combination that is a series of very complex individual decisions. That combination brings out what we call a pipeline, basically a definitive execution from a specific use case. And those pipelines are unique for each of our clients. We very much optimize that for.
And the thing about the optimization is that it's a very hard problem because you're balancing several competing requirements from reliability.
Many of our clients process millions of documents with us, and in doing so we become part of their critical business paths. So, you know, having downtime and basically not being able to deliver in key moments that that, that can't happen. So you have to figure out how to solve that.
Then there are of course also requirements related to latency, and that is quite. That changes a lot across use cases.
In some use cases you need a response in just a few seconds because it's a synchronous process and someone is actively waiting for the response, while in others it's perfectly fine to have an asynchronous, more detailed process that can sometimes take several minutes. And that's Completely fine.
Then of course you also have requirements for accuracy which, you know, everyone wants the best possible results. But there's, you know, kind of sub problems here, which is that first, how do you measure accuracy in such a noisy and complex environment?
And even if when you learn how to measure it, which again is actually much more challenging than one would imagine, then how do you optimize?
So that's actually quite complex. And when any client decides to work with us, what they actually buy is that we take all of that complexity, it goes away and they just plug into this very simple API and it just works.
I would say to anyone listening to this that they can try it for free on our
[email protected] but if this is a problem that they care about, that is core to their business, if they be compliant of inbuilt folks, they can actually see how we optimize this and we follow the complexity.
[00:13:57] Speaker B: Yeah, no, thought so. You're taking the complexity out of the equation for a lot of these companies need to be dealing with all these documents. It's pain just trying to manage everything.
[00:14:09] Speaker A: Let me give an illustrative example of what I mean by that. So imagine that.
Let's continue with the mortgage example.
So you are running a company that's operating in the mortgage space. What's the business of the mortgage companies? It's to first of all decide to who should or should not get their loan approved.
And basically they make money when they approve those loans. That's the business. Right. You have to basically be able to filter out those that are too risky. So you don't expect that the return of investment is going to be worth the risk. Right.
So. And there are a lot of moving pieces from coordinating different actors to stay compliant to getting, you know, getting the clients and you know, there's a lot of different pieces to this business and actually going through the documents. That is not the business. That is a task within the business, a very complex one, a daunting task. But, but something that if you just have solved, you're going to be much better at your business because that is not your core competence. The core competency is everything else. Your understanding of the industry, of the ecosystem, of the client.
[00:15:19] Speaker B: Yeah. And I feel like if you compare to, if you remember like an API, like an AI that's obviously built on like a lot of data, that knows how to process that data effectively. When you're thinking about accuracy, I mean, I'm obviously not a mortgage expert by like any means, but you're probably thinking that there is like a human element in evolving. Like before there was like all like these tools that could streamline this entire process and that minimizes the accuracy of things. If you make a bad decision, you tell this person, oh, this person's a great fit for mortgage. And after all it's like the complete opposite that can like screw up with your company. I mean you give mortgage to people that end up being a risk. That's like another thing I'd have in mind.
[00:15:58] Speaker A: For sure that's the case and well, it goes even deeper than that. So right now in the vast majority of companies operating in the mortgage industry, but also in many other industries, it's just a good example because it's a very document heavy process.
What's happening in those. It's that it's a human driven process. It's in many cases hundreds of people, sometimes locally, in many cases overseas, but that are basically triaging and going through those documents manually. Sometimes they have some tools to increase their productivity, but at the end of the day the information is going to be reviewed by a person. When you bring the technology like ours, you can actually automate most of, not all of the work. But we're also able to tell you which parts can and should be automated. And that changes the nature of the job because your ability, you will make money if you give that mortgage. So if you're going to approve a mortgage, the sooner that you know, the more likely the client is going to work with you, the more money that you will be making. So this is actually something that an operational excellence grows the business by default
[00:17:12] Speaker B: just by, just by being fast. I mean that's, if you get your application fast, it's like, oh, first person give me an answer on if I can get a mortgage.
[00:17:20] Speaker A: I'll go with them. Imagine you're trying to get your mortgage. So you apply to maybe like five mortgage provider banks or financial institutions of some sort. And four take a week to respond while one of them responds in four hours. Hey, if you're in a rush, you might even pay a premium, pay a higher interest rate to just get the.
[00:17:42] Speaker B: Yeah, yeah. It ends up being like a business advantage for like a lot of companies. When you sit that way. And I know the mortgage space is like one like example. I'm sure there's obviously a lot of like different like use cases on. One of the things I'd be curious is I feel like for example, if we think about more like software companies, like a lot of them are going to be thinking like, oh, like extracting data from Docker, like probably something that we could just build in house, you know, with AI, we could just vibe code the entire thing and probably, probably would be fairly easy to do. And I know you built a couple of companies.
I'd kind of be curious to hear, like, why you think people get that kind of wrong, you know, how easy or hard it kind of actually is, for sure.
[00:18:23] Speaker A: Well, I have a general response and then a more specific one to the nature of our business. The general response goes to the beginning of the conversation. Focus is gain.
So if you basically chase every shiny thing and try to build every piece of the puzzle by your own, hey, why don't you build your own payment processing platform, don't use Stripe, they're going to take 1% of your revenue. It's the same guiding principle, right? And then specifically about the document processing space, it's very easy to underestimate the complexity. And what I mean by that is that first of all, most teams building in house are not even measuring performance correctly because it's very hard to do.
And if you're not measuring correctly, it's very hard to know whether your solution is even that good.
And that takes me to something that I think we do quite well at, Immovable Fox, which is how to start a relation with a potential client, which is that we often start with a free proof of concept. We build an ImboFox optimized pipeline for them the same way I was describing before.
And that way they can directly test it against their own documents and data set and it gives them a real benchmark to make an informed decision.
The thing here is that the hidden cost of all of these potential in house projects is that maintaining software is quite complex and quite expensive.
Building the first version is just the beginning and, you know, just preventing drift as new models come out like new AI models and older models get deprecated, sometimes with a lot of notice, sometimes with not so much notice. And these providers have downtime, they have reliability issues, so you have to figure out how to orchestrate them and have fallbacks and all of that, and they change their pricing. So this is, these are just some of the concerns. There are many more. But basically you would be taking on all of this complexity. It does exist, we're just shielding you from it.
So basically you have to manage all of this change all of this complexity and you have to make sure that you do it without losing accuracy because you don't want to see degradation on your service without hurting reliability and without letting costs explode. You know, if we see a new model that is much better, but is 10 times more expensive kind of unit economics to power that, you need to know that.
So my general answer is that I wouldn't say that nobody should build in house, but I think it should be a conscious decision and everybody should be aware of what they're taking on. Right. It can make sense in some cases, but it is not a one time project. It doesn't stay solved. It becomes an ongoing engineering responsibility.
[00:21:10] Speaker B: Yeah, I can speak from experience because I'm not a technical founder by any means. We're not a tech company or anything like that. But we've recently started like building out like more integration in house tools like streamline some of our processes. And I actually spoke to someone else on a podcast like that helps people like build SaaS and everything. And that was like one of the things I was telling him was like, yeah, like I think we're about to be done with this project. I mean I like I've been doing this for like 60 days. Like I think we're like 90% there. And then it ends up being like another like month before it's like finalized and we've used it like now works very well. But yes, like you're so correct that it's all like a set and like done. There's a lot of like maintenance you're gonna have to do here and there. It's like, oh, this is not working anymore. The structure of this website change, we have to go change that. And I feel again like as you mentioned, like a lot of like people are like thinking about a product like oh, it's a D Z and then we have a product, we can use it like there's no other like time commitment involved. But it's, it's more like a maintenance game that forever will need to be like maintained. And I guess I'd be kind of like curious for you to like explain like why do you think that's some sort of like underappreciated cost of founders and operators when they're evaluating whether they should be building this themselves or if they should be just buying like a software or work of the company, for example?
[00:22:27] Speaker A: For sure, yeah, the build versus buy decision has always been an important one. But of course with the new paradigm and how technology and basically fluffy coding agents are working now, it becomes, and it's always there. You're always wondering, hey, is this something that is worth paying for or is this something that I feel considering, you know, building in house and taking.
So I would say that even though that problem has existed always, there used to be kind of a natural guardrail to it because only relatively small number of people were able to actually build software.
And those same people, assuming they could dedicate enough time and resources, were generally also able to maintain software that they built.
But that card drive is going away. So now almost everyone can create software because the barrier to doing that has, you know, dramatically decreased.
But building and maintaining are completely different skills. And in maintaining, especially as usage grows and you know, you put on more and more requirements of this piece of software as you should be, business is doing well. Right. You're doing it because you think you're going to use a lot, then you need to understand the real bottlenecks, the weak points of the solution. And you need to be able to anticipate how the problem will evolve so you can design your solution so it is able to accept the new developments, new technologies that should be incorporated at some point. Yeah, and you need to be able to do that while keeping everything under, under control. Right.
And that deeper understanding of systems architecture, technical data that hasn't scaled at the same pace as the ability to build software.
AI is of course very good at starting certain projects. It's getting better at understanding code bases, but maintaining is much more complex. Especially this is particularly true for systems that change behavior depending on the data. So in our case, if two documents come through the pipeline, you will get different results because they contain different data. And maybe you want to execute different pipelines for those documents because of a very specific parameter that you're controlling for. Right. So actually testing and being able to guarantee quality and preventive drift and all of the different concerns that we've already discussed on the conversation, that is very, very hard.
So again, I go back to my previous point, which is that I think building in house is absolutely worth it for some team.
The closer that this is to your core competence and your core value generation, I think the more serious you should take that concern.
But it is not the best universal approach for.
[00:25:24] Speaker B: Yeah, no, you should be doing a little bit of research before you dive in the deep end. To put it simply. And I feel like another component, like some companies, when they've been doing it, serving like a good amount of like people for a certain amount of time, they also get to collect more data and be able to make more improvements based on the different use cases and everything. So when you're building it for yourself, of course you're trying to like, make it specific to like you, but if you bring in different Use cases, different things. You have to, like, build new things because you don't have the other functionalities for certain type of, like, piece of the puzzle, etc.
[00:25:59] Speaker A: And with most complex problems, the issue when you start working on them is that you don't know what you don't know. So we have processed hundreds of millions of documents. So I won't say that we've seen it all because, you know, there's always something, but we've seen most potential problems and most challenges and we've built for them.
You will have to go through that all on your own if you decide to.
[00:26:24] Speaker B: Yeah. And last thing that I want to just go over a little bit because I was kind of curious. You talked about AI a little bit earlier and you just moved to San Francisco, I think it was like two years ago, from Madrid to growing Fox, and you've kind of talked about some sort of gap between how Silicon Valley perceives AI's current capability versus how the rest of the world sees it.
I'd love to hear from you what's maybe a misconception. Not you run into most often when you're talking about people about what AI can actually do for their document workflow. Because, I mean, good example is some people, they hear AI like, oh, chatgpt, amazing. Helps me do my grocery every single week. And then you got other people that's like taking over their entire company. It's doing like everything for them. So I'd love to hear your perspective on that.
[00:27:13] Speaker A: For sure. Yeah. I've been here in San Francisco for a couple of years, although I was coming here quite often before that.
And I'm originally Spanish from, from Madrid. So I lived the first 30 years of, of my life over there in Madrid. I think I have a.
At least for these two places. Right. A fairly good point of view.
And I would say that San Francisco is the bubble in the sense that it's completely different from everywhere else.
You get the feeling when you're here that everybody is completely what they say, AI build. Right. It's all AI wave. And they see first the LLMs and then the robots are going to solve everything for everybody, which is great, and I think we need that. But that is not the same across the world in every other place. Right. And most other places have a, first of all, much lesser exposition to this technology and a different approach to it.
Most companies in my experience are still very early. Many have only dipped their toes in the water, and some haven't done anything meaningful with AI at all yet.
I think Coding agents are a good example.
Probably that is the widest known, fastest growth use case for AI right now. That is coding agents like cloud code and Codex and so on and so forth.
And basically if you're here, you have the impression that everybody uses it and is orchestrating 10 or 20 agents at the same time to do whatever it is that they need to do. But then you go outside of this bubble and many companies, even many tech companies have not implemented this yet, or they're just starting to test it, or they're using some of the older versions of this approach, like GitHub, copilot dubbing.
It's a different paradigm. So it's not true that this has permeated everywhere, all of the economy yet.
And that is again for coding, which I think is the more mature use case that we imagine for editing apps. This is apps for craps. There's a lot of room for growth and opportunity for everybody.
And in our specific industry in document processing, I think we're seeing a similar phenomena in which.
So we started before ChatGPT, like right before actually we did the YC Summer 22, the batch and it came out right after the launch and we were building machine learning models and we were building this very sophisticated high performance solutions and then all of these new wave of technology came and we've incorporated that.
And every day I see that many people that could benefit from this technology because it now is able to solve problems that were impossible to solve or close to impossible without an insane investment of resources. Like again the use case that we've discussed about mortgages, but many others, insurance, finance, health care, any document heavy process, especially the more complex ones, you can do things that were unthinkable before and people are not aware of this and they haven't properly updated their vision. And actually one of the things that I invest time in, and that is one of my core functions at Infofolks I think is kind of make people in key industries aware of this or that. You know, a tool like IMO Fox can actually help them realize that new paradigm and bring their solutions and their processes to a much more advanced stage than they currently are in.
[00:31:13] Speaker B: Yeah, and I feel like you mentioned like Silicon Valley for example, like AI pill, like there's like this bubble again like where everyone's like using AI, everyone's launching like this new startup and everything like that. But outside of that, like there are still like some industries that cannot, I mean just for example for compliance, regulation, like they're, they're not able to get too much into, like, the AI side of things. I mean, I know, for example, financial, healthcare, they're like slowly rolling some out. From my understanding, they're either creating like, partnerships with these major AI companies or they have like, their own model that they're building just to make sure that everything is kept, like, private and everything. But it's like, yes, we're in this bubble where it's not like fully rolled out to everyone and we're looking at this like it's taken over the world. Like, if you're on X or like Twitter, for example, people are saying, like, not even working anymore. I got like 8 AI that replaced my entire company. Like, we're just stuck in this bubble. Like, most people are like, I'll just use this for my grocery list. Like, just something like that. So I always find that interesting and I kind of wanted to your perspective because I know that you're literally at the center of this bubble.
[00:32:14] Speaker A: So for most people, AI is still just this chat box, not your ask question, like it's a Google search replacement. But it's much more than that.
[00:32:24] Speaker B: No, that's fair. Well, Alberto, I really appreciate you coming on the podcast here and in terms, like, anyone that wants to, you know, reach, learn more about Invo Fox, reach out to you. What are kind of some of the the best ways to find you here for sure?
[00:32:38] Speaker A: Yeah, yeah. Well, I'm very active on LinkedIn, so reach out over there. I'm also on X and of course you can always check out our website and I, I will receive any message or a contact that you send through that chat.
[00:32:49] Speaker B: Perfect. Well, we're going to put Invofox in the show notes description if anyone wants to check that out. And your LinkedIn too, if anyone wants to reach out to Alberto here, learn more about what he's doing on how he's using all these data and API to make everyone's life easier here and then to listeners, if you've enjoyed this episode, don't forget to like, subscribe, comment, review all that fun stuff and then, yeah, until then, keep pushing and we'll see you in the next one.
