Sylvestre Dupont, Parseur

AI Took Over His Market. His Bootstrapped SaaS Grew 60%

Episode

0:00 / Loading…

Listen on:

Like this episode?

Get real founder strategies for the AI era. Delivered weekly.

Free weekly newsletter · No spam

Episode Summary

Parseur's competitors have raised hundreds of millions of dollars and are pouring money into AI. ChatGPT can do the basic version of what Parseur does. And Sylvestre Dupont's entire bootstrapped SaaS is six people spread across six countries. His bet - that most businesses want something simple that works in minutes, not a complex enterprise solution - is what keeps them alive.

In this episode, Sylvestre explains how he rebuilt his entire product around AI while bootstrapped, why he ignored the "pick a niche" advice, and how a 10-year-old document parsing tool is still growing 60% year over year against competitors with 100x the resources.

Sylvestre Dupont is the co-founder and CEO of Parseur, a platform that automates data extraction from emails, PDFs, and spreadsheets.

The idea started with a side project in 2015. Sylvestre wanted to build a travel map that auto-updated from booking confirmations. He and his co-founder Sylvain, a senior Python developer he'd known for 25 years, saw a bigger opportunity: a general-purpose document parsing tool.

They put up a landing page, ran Google Ads, and collected 50 email signups. That felt like enough validation. So they spent the next year heads-down coding. Full features, payment system ready, zero marketing.

In December 2016, they launched on Product Hunt and Hacker News. Nothing happened. They emailed the 50 people from a year earlier. Two signed up and quit immediately.

So they started from scratch on the marketing side. They began answering questions on Quora, genuinely helping people with document automation problems. That's where their first real customers came from. They also dropped the price from $49 to $9 a month just to get anyone to try it.

What set them apart was simplicity. Competitors required users to write complex extraction rules by hand. Parseur let you visually highlight what you wanted. Setup took 10 minutes instead of two hours. That bootstrapped SaaS advantage - simple, self-serve, no sales call required - became the foundation of everything.

Growth came slowly through SEO and a Zapier integration that converted at 20 to 30 percent. For the first five years, it was just the two of them. No employees, no investors, no board.

Then AI changed the game. ChatGPT could do basic document parsing. VC-funded competitors like UiPath and ABBYY were spending hundreds of millions on AI. Sylvestre had to rebuild his entire product around machine learning - funding the transition from customer revenue, not investors.

His bootstrapped SaaS strategy for survival: don't try to out-feature the giants. Be the tool that any business can set up in minutes without talking to sales. Simplicity as a moat, not technology.

Today, Parseur generates seven-figure ARR with close to 1,000 paying customers in over 70 countries. A bootstrapped SaaS, still six people, still 100% founder-owned - and still growing.

Topics: Bootstrapping|Positioning & Differentiation|AI-Powered SaaS

Key Insight

Parseur co-founder Sylvestre Dupont built a bootstrapped SaaS to 7-figure ARR and 1,000 customers with just 6 people while VC-funded AI competitors spent hundreds of millions. His strategy - simplicity and self-serve setup in minutes instead of enterprise complexity and sales calls - lets a bootstrapped SaaS compete in a market being commoditized by AI.

Key Ideas

Parseur's bootstrapped SaaS advantage is simplicity: 10-minute self-serve setup vs enterprise tools that require sales calls and complex configuration
Dupont rebuilt from rule-based to AI-powered parsing while bootstrapped, funding the transition entirely from customer revenue
Competitors with hundreds of millions in funding serve enterprise - Parseur targets the businesses that want automation without complexity
ChatGPT can do basic document parsing, but can't handle end-to-end automation at scale with compliance and data quality controls
Going horizontal instead of vertical worked because SEO lets Parseur target infinite use cases individually

Key Lessons

🏢 A bootstrapped SaaS can beat enterprise complexity with simplicity: Parseur's 10-minute self-serve setup wins against competitors that require sales calls, even when those competitors have hundreds of millions in funding.
🧠 AI commoditizes features, not end-to-end solutions: ChatGPT can parse one PDF, but it can't handle pre-processing, routing, compliance, and integration at scale - that's where the real product value lives.
💰 You can fund an AI rebuild from revenue, not investors: Parseur rebuilt from rule-based to AI-powered entirely from customer revenue, keeping 100% ownership and avoiding dilution.
📉 A bootstrapped SaaS dies faster from bad marketing than a bad product: Sylvestre launched to crickets, dropped price 80%, and rebuilt his approach from scratch. The product was fine - the go-to-market was the problem.
🚀 Integration partnerships pre-qualify customers: Parseur's Zapier connector converted at 20-30% because those users were already automation buyers looking to connect tools.
🎯 Horizontal SaaS works when your positioning is use-case specific: Parseur is generic, but their SEO targets individual use cases - making them appear vertical to each customer segment.
🤝 Genuine community engagement beats marketing at the start: Answering real questions on Quora without being promotional built trust and attracted Parseur's earliest paying users.

Watch the Episode

Episode Q&A

How does a bootstrapped SaaS like Parseur compete against VC-funded AI companies?

Sylvestre focuses on simplicity as the core positioning. Enterprise competitors require talking to sales and complex custom setups. Parseur offers self-service setup in 10 minutes. Most businesses want something simple that works, not a complex enterprise solution.

Why isn't Sylvestre Dupont worried about ChatGPT replacing Parseur?

ChatGPT can extract data from a single PDF, but can't handle end-to-end automation at scale. Parseur handles document pre-processing, routing, data quality controls, compliance, and integration with downstream systems. The AI is just one piece of a much larger pipeline.

How did Parseur rebuild from rule-based to AI-powered while bootstrapped?

They funded the AI transition entirely from customer revenue. They focused on the full pipeline - pre-processing documents for optimal AI performance, caching to reduce costs, and using fast models over expensive ones to keep response times and costs under control.

What is Parseur's bootstrapped SaaS strategy against enterprise competitors?

Parseur combines three constraints: simple to use, scalable in volume, and compliant with data privacy. Enterprise competitors nail compliance and scale but are complex. Parseur bets that self-serve simplicity is what most businesses actually want.

How did Parseur succeed as a horizontal bootstrapped SaaS instead of picking a vertical niche?

It wasn't a conscious strategy initially. They built a generic parsing tool because they were developers who liked abstraction. The same tool processes utility bills and pigeon genealogy PDFs. They found horizontal works because SEO lets them target each use case individually.

How did Parseur get its first customers after launching to crickets?

They started answering questions on Quora about document automation, being genuinely helpful rather than promotional. They also slashed the price from $49 to $9 per month to remove friction. The first 50 paying customers came largely from Quora and early SEO content.

How did Parseur's Zapier integration drive growth with a 20-30% conversion rate?

Users arriving from Zapier were already automation-minded. Because Parseur's value depends on sending extracted data to other apps, Zapier users were perfectly pre-qualified. Zapier also promoted the connector through newsletters and blog posts.

How does Parseur manage AI costs as a bootstrapped company?

Server costs still exceed AI costs. They cache aggressively, avoid the most expensive models (which are also the slowest), and balance quality against response time. They optimize for fast, good-enough AI rather than the smartest, most expensive option.

How is Parseur adapting its SEO strategy for the AI search era?

They monitor how well they appear in AI search results and optimize for AI overviews. Sylvestre notes AI overviews mostly replace informational queries, not product searches - people looking for a tool still visit the website, so signups haven't dropped.

Book Recommendations

The Black Swan

by Nassim Nicholas Taleb

Links

Parseur: Website | LinkedIn | X
Sylvestre Dupont: LinkedIn | X
Omer Khan: LinkedIn | X

Full Transcript

Omer Khan [00:00:03]:
Welcome to the SaaS podcast. I'm your host, Omer Khan. AI has changed the playbook for building and growing SaaS. Every week I talk to founders who are writing the new one. My guest today is Sylvestre Dupont, who bootstrapped a SaaS to seven figures in ARR with a team of just six people and zero outside funding.

Omer Khan [00:00:22]:
But he spent an entire year building without talking to a single customer. Launch day came and went, nobody signed up. He even reduced the price of his product to $9 a month just so people would try it. Then well funded competitors started pouring hundreds of millions into AI doing exactly what his product does.

Omer Khan [00:00:42]:
In this interview, Sylvestre breaks down how he rebuilt his entire product around AI funded entirely through his customers revenue and how an unconventional channel that most founders overlook became his primary source of customers. Sylvestre, welcome to the show.

Sylvestre Dupont [00:01:00]:
Hi Omer, thanks for inviting me.

Omer Khan [00:01:02]:
My pleasure. So for people who don't know Parser, what does the product do, who's it for, and what's the main problem you're helping to solve?

Sylvestre Dupont [00:01:10]:
So parser is a B2B SaaS. And what we do is quite simple. We help automate data extractions from documents.

Sylvestre Dupont [00:01:16]:
So if you receive too many documents, too many PDFs, too many emails, too many spreadsheets, and you want that data on the documents to be somewhere else in your accounting software, in your CRM, in your just spreadsheet, you send them the document, you tell us what you need and we extract it for you.

Sylvestre Dupont [00:01:35]:
It's a very generic product. We are not vertical, we are not tied to an industry or particular use case. You can do it for many, many things. And what we like about it is that every company has documents.

Sylvestre Dupont [00:01:53]:
I don't know any single company that says, oh no, I don't need to copy and paste some data from documents I received. I have invoices, I have delivery notes, I have contracts. So the pool is gigantic, the use cases are infinite and that's quite fun to work with.

Omer Khan [00:02:11]:
So give us a sense of the size of the business. Where are you in terms of revenue, customers, size of team?

Sylvestre Dupont [00:02:17]:
We are quite an old company, so we started 10 years ago so we had the time to grow. Today we do seven figures. We remain quite small. Today we are only six people. We kept very small for very long. For the first five years we were just the two of us, my co founder and I.

Sylvestre Dupont [00:02:37]:
And then slowly we increased the team, starting with marketing and later development and support. But yeah, we keep very small. One particular thing about our team is that we are remote. Everybody remote. And very far spread apart. We are six people, we live in six countries across five time zones. That spreads 12 hours.

Sylvestre Dupont [00:03:00]:
So the latest recruit is from Singapore, so it's the most eastern one and the furthest west is the support in Miami in the US and how many.

Omer Khan [00:03:09]:
Customers do you currently have?

Sylvestre Dupont [00:03:11]:
Close to 1,000 now paying customers. We have many more, we have a free plan. So we have many more non paying.

Omer Khan [00:03:17]:
And the business is totally bootstrapped.

Sylvestre Dupont [00:03:20]:
Yes, we started fully bootstrapped, 50% shares to each share to its co founder and we kept it like this.

Omer Khan [00:03:28]:
Great. So let's start with like, let's go Back to like 2016, 2015, like where did the idea for this product come from?

Sylvestre Dupont [00:03:38]:
My co founder and I, we know each other for very long, like 25 years now and we are both technical minded people. He's a senior Python developer. I used to be in consulting at the time when we started, but I have a technical background and we started to want to create some things on our own.

Sylvestre Dupont [00:03:58]:
So trying to start a business thinking about different ideas and at the same time we were both traveling quite a lot at the time. And so I started to work on some travels. I started to work on a map to map out my travels to make them show them nicely.

Sylvestre Dupont [00:04:13]:
And I got bored by having to enter all the trips one by one on this map because we were traveling a lot that way. I made it trips. And so an idea came that oh, maybe we could have a tool. You just send the booking confirmations and you get your map updated.

Sylvestre Dupont [00:04:28]:
So we started to research the space and we saw that there were already some tools doing that, but none that were doing it the way we wanted. And we said, oh, maybe that's an idea. That was 2015. Okay. We said let's go. I think we did all the mistakes that everybody does when they start a new business.

Sylvestre Dupont [00:04:47]:
So we spent a full year hunkered down coding, zero marketing. We did just a little experiment at first to have a quickly done homepage and put some Google Ads to see if people would click sign up and then see if we would collect emails and we collected 50 emails.

Sylvestre Dupont [00:05:05]:
So we said, okay, we are product market fit, we're good.

Omer Khan [00:05:09]:
That's right.

Sylvestre Dupont [00:05:11]:
And then one year later, we did everything. We even had the payments ready on day one. We did everything. So one full year, full development, no marketing, no nothing but full features. And then we launched in December 2016 and we launched on product hunt. We launched on hack and use like everybody does.

Sylvestre Dupont [00:05:32]:
And nothing happened of course because we didn't do any marketing, we didn't talk to any potential customers. We did send the 50ish emails we had collected one year before. We did send them, hey look, we are ready. But it was one year later. So I think two people signed up and quit immediately.

Sylvestre Dupont [00:05:53]:
So it was a huge failure.

Omer Khan [00:05:55]:
This isn't the first time I've heard something like that. And a lot of new founders get into this situation. They're excited about the product or an idea, they start building it. It's more fun building the product than it is to go and talk to potential customers and think about marketing or sales.

Omer Khan [00:06:12]:
So you keep doubling down on that and ultimately you get to a point where you've invested so much time and money into the product that you're just looking for confirmation that you're doing the right thing.

Omer Khan [00:06:28]:
So it sounds like you did a very sort of similar thing where you were excited about the solution and then you're looking for a market that might want this. What would you say when you sort of look back at that?

Omer Khan [00:06:46]:
I mean, there's obviously a whole bunch of things that you could have done differently, but what was the biggest mistake that you made in that first year?

Sylvestre Dupont [00:06:54]:
Well, we should have found potential customers that wanted to talk to us that were excited by such a product and could have built it with. We could have built it together. I would have loved to have somebody that would have said, hey guys, I'm happy to pay X amount per month.

Sylvestre Dupont [00:07:09]:
And we would have built a version that works for them together and we would have been much faster, much more straight to the point. And that would have been good. But yeah, and in general, just start marketing is not enough to.

Sylvestre Dupont [00:07:24]:
It's not only just having a homepage and a sign up button, you need to have a strategy. You need to see how you're going to attract your customers, where you are going to find them, what is going to be your elevator pitch? What are you.

Sylvestre Dupont [00:07:37]:
Is the idea you have in mind, the way you explain it, something that speaks to them. So there's a lot of work there. I think as a technical founder, you just focus on the technical because that's what you know, that's what you are comfortable with. But you have to learn marketing.

Sylvestre Dupont [00:07:52]:
If you are launching a new product without somebody that is more product and marketing oriented, this is something you have to learn on your own and you better do it sooner than later.

Omer Khan [00:08:02]:
And you didn't know who your customer was either at that point, which makes things really hard to figure out. And it's very easy just to Say, oh, our customers are people who need data extraction from documents. But as you know, that's probably not the best way to figure out who your ICP is.

Omer Khan [00:08:23]:
So once you got to that point where you launched and it was kind of crickets, how did you then figure out who you were going to serve, who you were going to solve this problem for?

Sylvestre Dupont [00:08:34]:
So we did two things. The first thing was to start and publish some content on our website. Some blogs about what is document parsing, what is data extraction, how does that work? Some things a bit more technical, some things a bit more general of what we thought people would be interested in. But that was SEO play basically.

Sylvestre Dupont [00:08:55]:
But that takes lots of time. You don't overnight have some traffic. But we did start at the time and then in parallel we started to contribute to some forums. The one at the time that was the had the most conversation was Quora that I think now is totally disappeared. And we pestile Reddit.

Sylvestre Dupont [00:09:16]:
But yeah, and Quora is where we found our first customers. We started to. There were already some people asking about document automation, data extraction, et cetera. We started to answer that, we started to be active, we started to be helpful, not only just to blast our products, but just to honestly answer what could be the best solution.

Sylvestre Dupont [00:09:34]:
And this is where we started to see some signup and finally we could start and talk with people because we had the little chat widget already at the time and it was priceless to just interact with people that were drying the product.

Omer Khan [00:09:48]:
So what were you posting on Quora? There's a principle here. You weren't just posting a link to your website, right?

Sylvestre Dupont [00:09:56]:
No, we were making elaborate answers, answering exactly the customer's problem, mentioning our brand in passing along with other competitors, just to be fair, just not to be too pushy. And that worked, at least at the time. I don't know anymore.

Omer Khan [00:10:12]:
Yeah. Now I'm sure Even back in 2016, it seems like a long time ago, but I'm guessing Parser wasn't the only game in town. So what was it about your product that you think got people interested in compared to other potential solutions that were already out there?

Sylvestre Dupont [00:10:36]:
So we are talking about pre AI boom and everything. So at the time, products were quite complex to use. Basically what you had to do in all of the competing products was to have to create some rules. So you say, oh, this is my document.

Sylvestre Dupont [00:10:52]:
Look for a line that starts with invoice and then remove everything that is before the column and then go to the end of the line and you have to specify Those rules manually, one by one. It was super tedious.

Sylvestre Dupont [00:11:03]:
And if your document had another layout, we would have to do that all over again and decide some rules to choose with this rule set or that rule set. It was complex to set up and to use. We decided to keep it simple and stupid. So we decided to be visual.

Sylvestre Dupont [00:11:19]:
So we would show you your document, you would highlight the data you wanted, and we would do the heavy lifting behind the scenes to figure out where this bit was and how to make it work. And we make it so that you could have as many layouts as you wanted and we would automatically pick the right one.

Sylvestre Dupont [00:11:38]:
So for us, simplicity was what was our first selling point. You could basically get set up in maybe 10, 15 minutes rather than two hours.

Omer Khan [00:11:47]:
And how did you figure that out? Did you look at the products that were out there and say, these are complicated to use. How can we make this simpler? Or were you getting feedback from people complaining how hard it was?

Omer Khan [00:12:03]:
What was the process that got you to the point where you said, a visual type solution is going to be one of the best ways that we can differentiate our product right now?

Sylvestre Dupont [00:12:14]:
Well, that was quite early because I am. I'm quite analytic as a person. I love Excel spreadsheets and going deep into the data. So I needed to have a bit of an approach on how we would do that business. So I did look at all the competitors. I tried them all. I look into the pricing.

Sylvestre Dupont [00:12:33]:
I do look at everything. And for me, what came out of it was the complexity of the product. And my co founder, Sylvain, is also somebody that is very good at making simple solutions, at finding elegant answers to complex problems.

Sylvestre Dupont [00:12:47]:
And it appealed him as well to try to find the best way to have something visual and easy to present for our users.

Omer Khan [00:12:56]:
You mentioned you got your first customers through Quora. Did the first 10 customers come through there? And how much were you charging those early customers?

Sylvestre Dupont [00:13:08]:
So when we launched, the very, very first day we launched, we had our first pricing, I think was $49. But when we saw that nothing was happening, we slashed our pricing down to $9, just, you know, just to try. We just wanted feedback. And so that's how I think that contributed to having our first paying customers.

Sylvestre Dupont [00:13:28]:
And, yeah, I think. I wouldn't say the first 10 were from Kora, but I think a large share of the first 50 were from Kora. And by that time we kept publishing some articles, and by that time the SEO play started to work for us.

Sylvestre Dupont [00:13:42]:
It was a bit easier back then, and we started to attract most of our users from SEO.

Omer Khan [00:13:48]:
So aside from posting on Quora, having these sort of answers and doing the creating content and trying to make the SEO play, was there anything else you were doing to find customers?

Sylvestre Dupont [00:14:01]:
Well, we tried a bunch of things like posting on some communities we would find or, or even at this time, I think we already started do some ads, but nothing really worked. The first thing that after SEO, the first thing that worked for us was when we created a connector for Zapier.

Sylvestre Dupont [00:14:22]:
Zapier is an automation platform, if you know about it. And then because our tool is a bit useless on its own and it shines is that when you can send the data to your applications.

Sylvestre Dupont [00:14:33]:
And because it takes a lot of efforts to build integrations for each and every application, we decided to integrate with Zapier that already integrated with thousands of applications. And these guys, they, well, you could create a connector for them. So we did that.

Sylvestre Dupont [00:14:47]:
And then they would advertise about you, they would send a newsletter and say, oh, we have this new connector that arrived and they were even did a blog post at some point about mentioning us. And that was great because the guys that were on Zapier were super qualified. There was not that much traffic compared to SEO.

Sylvestre Dupont [00:15:08]:
But the conversion rate from the traffic we got from Zapier was incredible. We had like maybe 20 or 30% conversion rate. So that was very nice.

Omer Khan [00:15:17]:
So today you have over 1,000 customers. Are they still coming from SEO?

Sylvestre Dupont [00:15:24]:
Yeah, mostly it's maybe, I don't know, 95% today with the SEO. So what we did is that our first recruit was a marketing lady person that would dedicate it to publishing content. So that was in 2020. And we kept growing and growing and growing our surface because our use case are unfit.

Sylvestre Dupont [00:15:49]:
So we just have to understand what our customers do, write articles, try to find the long tail keywords, and publish and publish and publish. That works great.

Sylvestre Dupont [00:15:59]:
And very recently, like last year, we doubled down on it and we translated all those articles in at first three languages, but now we have, I think close to 10 languages to widen the net, basically.

Omer Khan [00:16:14]:
But Sylvestre, all the LinkedIn experts tell us that SEO is dead. So how is it working for you?

Sylvestre Dupont [00:16:21]:
It's still working today, but indeed there's been a drop in traffic because of the AI overviews. AI traffic. So now our place to try to do AI SEO Geo, I think, as they call it nowadays. And so we monitor how good we appear in the AI search results, we try to increase that and.

Sylvestre Dupont [00:16:47]:
But at the end of the day, it also depends on the type of traffic when your answer that gets replaced by an AI overview from Google. Most of the time the people were looking for an informative answer. They were looking for a definition. They were not looking for a product.

Sylvestre Dupont [00:17:03]:
If they look for products, they will not stay on the explanation from Google. They will visit the website. So even if our traffic has dropped, our numbers of sign up has not dropped that much, if at all, it's increasing.

Sylvestre Dupont [00:17:17]:
So you have to look at the end of the day of the result of your conversions, are you getting traffic.

Omer Khan [00:17:24]:
From ChatGPT as well?

Sylvestre Dupont [00:17:26]:
Yes, and growing and perplexity as well and more and more anthropic.

Omer Khan [00:17:32]:
So are you doing anything different when it comes to SEO versus aeo, geo, whatever the term is these days. But is there anything you're doing different?

Sylvestre Dupont [00:17:43]:
We try a few things. The first thing we tried is that on our blog post now you have some buttons to summarize the blog post with AI in the hope that then we will end up in some training data from those AIs later down the line to get our brand mentions.

Sylvestre Dupont [00:17:58]:
I have no idea whether that works or not, but the buttons are clicked. So some people enjoy having the ability to have a quick summary. And then we try to keep up to date with whatever we read or the people that research about AI do.

Sylvestre Dupont [00:18:13]:
And so a lot of them is about, for instance, putting some FAQs, lots of FAQs, structured data FAQ in your articles so that the AI has no trouble reading them, putting some bullet points at the top if possible.

Sylvestre Dupont [00:18:28]:
In a lot of ways, the, at least from what I see today, the recommendation that we read for being good at AI looks like the recommendation you would read 10 years ago for Google. So a bit of a dumber AI, but just trying to provide the easiest, simple solution to whatever the customers are looking for.

Omer Khan [00:18:49]:
Okay, great. So you have those initial customers. Finally things are, are looking on the app. What did you, what did you change about the product is then get to 20k, 30k, 40k in MRR? Did you start focusing on a particular niche or an icp? Did you double down on a specific problem?

Omer Khan [00:19:15]:
Like what was the, what were the changes that you made that helped you go from, you know, a handful of customers to, to maybe the first hundred or so?

Sylvestre Dupont [00:19:25]:
So a few things. The most important one was we actually made the product simpler. I'm coming back again at this, I'm sorry, but that's my thing.

Sylvestre Dupont [00:19:36]:
For instance, at first we would ask the customers that would sign up, they would ask them to create a Mailbox where they could send documents to and they would have to name the mailbox. Then we would ask them to create a template where they could visually highlight the data.

Sylvestre Dupont [00:19:49]:
Then they would have to name the fields, name the templates and etc. And there was a lot of, we would request from a lot of inputs from the user. Progressively we started to remove all of that and then AI came. So with AI it's much easier to have AI do things for you.

Sylvestre Dupont [00:20:10]:
And now we don't ask anything from the user. They sign up, they upload the document, we create a mailbox automatically, we find a name for it, we, we look at the documents. We use AI to try to identify what is the most likely data points that the user want to extract.

Sylvestre Dupont [00:20:27]:
We do it, we show it and we say, hey, that's what we think you want, not you can change it. But at least we removed all the friction that we saw over time.

Sylvestre Dupont [00:20:38]:
But that took months and years of maybe watching recordings or finding bugs, of looking for some niche corners and we have not done, we are still working on it today to remove friction.

Omer Khan [00:20:52]:
So you started out as this rule based parser which you explained earlier and today you are this AI powered parser data extraction tool. But you guys didn't just put like a wrapper onto like GPT4, right? You decided to take a different approach. Can you just explain what you did?

Sylvestre Dupont [00:21:16]:
Yeah. So today AI is very good at extracting the data. I mean you can send one PDF to ChatGPT and most of the time it would get you some data and you can extract it.

Sylvestre Dupont [00:21:27]:
But if you really want to automate your whole workflow, it's not just about sending a document to ChatGPT, it's having the full end to end, automatically fetching the documents, looking at them, routing them to the right mailbox, they need to be.

Sylvestre Dupont [00:21:46]:
There is a lot of work in pre processing the documents so that the AI can have the most chances of finding the right data. So for instance, if there are images or scans, we try to make them straight. We play with contrast if we have to, we resize the documents. Some PDFs are awful.

Sylvestre Dupont [00:22:04]:
I don't know if you ever entered the wonderful world of PDFs, but some of them are so complex and so different than usual that we have to have some codes that fix all the issues that can happen so that we present the best possible version to an AI and then the AI does it.

Sylvestre Dupont [00:22:22]:
And what the AI does, even if we have years of practice about finding the right flow and prompts and Tweaks and many things. At the end of the day it's an LLM. So we keep up to date with what happens. And then you have to export the data and then there again here you have a lot of.

Sylvestre Dupont [00:22:43]:
You have to transform the data, make sure the dates are all the same, the numbers, the locations. We have a lot of downstream post processing that happen before we can send it. So what we do today, basically if you have three documents on a folder, you don't need us.

Sylvestre Dupont [00:23:00]:
But when you need is what if you have 100,000 of the documents that you have to pass every year and you want an end to end flow?

Omer Khan [00:23:12]:
What's like your token usage or costs of processing this volume of documents? You're a bootstrapped business. We know that using AI to do anything like this can very easily get seriously out of control. So what's the situation for you? How are you managing those costs?

Sylvestre Dupont [00:23:39]:
To be honest, it's not that crazy. Our costs today are. Our server costs are more important than our AI costs. Maybe this year is going to change because we keep adding AI a bit everywhere, but it's under control. It's not a large part of our costs. And why is that?

Sylvestre Dupont [00:23:59]:
Is that again, we try to keep the most simple approach as possible, cache everything we can so that we don't have to ask twice.

Sylvestre Dupont [00:24:09]:
We don't use the smartest, best AI, most expensive ones as we see today, not because we don't want to spend the money, but because they are slow and they get limited and the people that users, they cannot wait for half an hour to get their 100 page document processed.

Sylvestre Dupont [00:24:27]:
So we try to find the balance of the best quality AI for the fastest response times. And that's not too much today.

Omer Khan [00:24:36]:
Well, I think you're in an interesting space because I sort of see you in the middle.

Omer Khan [00:24:42]:
On the one hand, you have other players in this space in sort of document processing, which as far as I know, I'm not an expert on this, but it's a pretty big opportunity and there's a lot of spend in that space and some of these companies are spending hundreds of millions of dollars while you're having to sort of be in this bootstrapped business with the constraints that come of running a beam bootstrap.

Omer Khan [00:25:06]:
At the same time, you've got these developments happening with ChatGPT and Claude Code and all of this stuff. So what would you. I want to talk about both of those in terms of how you think about your positioning and kind of competing against some of these bigger Players with a lot more money.

Omer Khan [00:25:29]:
But let's start with the other scenario where people are looking at something like Parser or other products and saying, I don't need that anymore, I can do whatever I want. With ChatGPT and Claude.

Sylvestre Dupont [00:25:42]:
Well, some, again, for some use case, it might be perfectly reasonable and I would actually encourage some users to do it. Especially again if the volume is low. Sorry, you may not need us. If you are happy with downloading an Excel and keeping it on your folder, you may not need us.

Sylvestre Dupont [00:26:02]:
But if you start doing your own rollout, your own solution based on those AI, so you try to ask them to write some codes and host the codes, I can guarantee you that it's going to work no problem.

Sylvestre Dupont [00:26:15]:
But you're going to have to maintain it quite often because the models change, the package code out of date and it may not end up being as cost effective as it is using a third party solution. We had quite a few customers still today that come to us and say, okay, we tried our own solutions.

Sylvestre Dupont [00:26:36]:
It was a nightmare to maintain. As the volume grew, it kept exploding. We went back to looking for SaaS cloud solution and you guys are doing it and it's great.

Sylvestre Dupont [00:26:47]:
Even us now we have to keep on the latest and greatest AI development because they change so fast that it's almost a full time job to make sure we have the best engine we can offer to our customers. So basically we do the R and D for our customers.

Omer Khan [00:27:04]:
Yeah, I think the whole idea of Vibe coding is, is so fundamentally it's just a completely different change to what we've been used to when it comes to software and this sort of the space.

Omer Khan [00:27:25]:
But I think there's a lot of realities of Vibe coding your own solution that a lot of people or businesses haven't fully thought through or understood the consequences of maintenance being one of them. Right.

Sylvestre Dupont [00:27:41]:
It's just as a developer, it's a lot of fun to Vibe code because you can go so fast, so quickly. But then if you start looking under the hood and see what the AI is actually producing, you can see some quite horrendous stuff. And I do that in a weekend because that's what I enjoy doing.

Sylvestre Dupont [00:28:01]:
So I have some side projects and I, I like to wipe code them, but it's fun for a typical toy project, it doesn't scale at some points it gets too much, too complex and the AI is lost and maybe it's going to improve, but today it's not ready.

Omer Khan [00:28:18]:
I saw a tweet somewhere, maybe it was a meme or something. It said something like software is dead. Vite coding is the way to go. And I've canceled my $10 a month calendly subscription and I now have my own solution which is 5% accurate with the scheduling and only costs me 150 bucks a month to maintain.

Omer Khan [00:28:47]:
So let's talk about the other side. So you're also in a space where there's some very well funded competitors who are probably investing a lot more into the product innovation, AI and so on. So where do you see yourself fitting in the market? If you're talking to a potential customer and they're saying why should I use you?

Omer Khan [00:29:08]:
Why is your product better or different than these other things? What's your answer to those people?

Sylvestre Dupont [00:29:15]:
So we try to mix three constraints that usually are hard to mix together. We want to be simple to use, we want to be scalable in terms of activity. You could go as much volume as you need.

Sylvestre Dupont [00:29:31]:
And we want to be compliant or trustworthy because our customers are sending us personal data and we know that we have to do everything we can for ensuring that we treat that data as privately as we can. For I think it's quite hard to combine all of them.

Sylvestre Dupont [00:29:56]:
So if I look at the bigger players, the VC funded players, they are compliant because that's a given for them and they can scale, they have the money for that. But usually their processes at their tool is very complex because they are serving very large companies. So they are used at making very complex custom built solutions.

Sylvestre Dupont [00:30:29]:
I think that deserves the purpose of automation. For me, automation should be simple as possible, self service. You shouldn't have to talk to sales to get access to the product. For me, if you understand business automation, you shouldn't have a human in the process.

Sylvestre Dupont [00:30:46]:
So for me, the big VC funded customers that have as their first step talked to us is a bit of a red flag. But even the ones that have some free service for what I can see, they could do better in having something that is actually simple to use.

Sylvestre Dupont [00:31:05]:
And this is where we hope we try to keep our focus and have something that can be set up in few minutes, keep it simple always.

Omer Khan [00:31:16]:
Let's talk a little bit about your target market. The conventional wisdom for any founder especially bootstrapped would be to pick a very specific problem, very specific niche market, very clear ICP and go and solve the problem for them first before you try to expand from that.

Omer Khan [00:31:44]:
And you didn't do that, you went horizontal I think from day one with the product. And then when we look at the landscape today, with everything that's happening and AI and the changes, we often hear, horizontal products are toast.

Omer Khan [00:32:00]:
It's all about if you're going to succeed, it's going to be more about vertical SaaS and those kind of plays. So first, maybe tell me why you decided to go horizontal from the start.

Sylvestre Dupont [00:32:13]:
It was not a decision we make consciously at first. We just said, oh, let's build a parsing tool that sounds fun and it's generic and we are developers, so we're going to abstract everything. So yes, maybe we should have niched at first and then expand it. But that didn't happen. But that worked for us, so why not?

Sylvestre Dupont [00:32:37]:
And there is the. It's so thrilling when that works. It's so thrilling. We had some guys that in one day they got to pass 10,000 utility bills. And you talk to them and they're very happy. They get all the utility bills.

Sylvestre Dupont [00:32:52]:
And then you jump to the next conversation and it's the guy that is a pigeon race, a breeder, and he has a genealogy of pigeons raised pigeons, and he has PDF with papa pigeon, mama pigeon and baby pigeon, and he wants to extract that from a PDF and that works too.

Sylvestre Dupont [00:33:11]:
And the exact same tool with the exact same config. Because we created it in such a generic and simple way, it works the same. And there is a beauty in that, right? To build a product that is so lean and hopefully so well thought out that we don't care about whichever data we receive.

Sylvestre Dupont [00:33:31]:
We just care about what you need to extract from it and then we give it back to you.

Omer Khan [00:33:36]:
But does that make it harder to market the product and reach potential customers?

Sylvestre Dupont [00:33:41]:
Yeah, I think. I mean, I've never built a vertical SaaS, so I cannot have firsthand experience on how the difference is really. But yes, I definitely. It's not like on our homepage, we can say the solution number one to extract the data from your pigeon genealogy tree. So we have to find some wide language.

Sylvestre Dupont [00:34:03]:
But again, most of our traffic is SEO, and seopay is great for that because you can write about how many use cases as you want and deep dive on it and talk to your customers and make sure you use their vocabulary.

Sylvestre Dupont [00:34:16]:
So there's a bit of work there, but you can go very wide with SEO, there is no limit.

Omer Khan [00:34:22]:
So that makes sense. And it sounds like what you've set up with SEO and this funnel and how you acquire customers is working for you.

Omer Khan [00:34:36]:
Do you worry that maybe now, with the way that we're talking about AI tools, that you can have new players come in who are more vertical focused and doing a much better job at finding a solution for the pigeon guy.

Omer Khan [00:34:55]:
Or do you think that that isn't necessarily a concern because of just the nature of what this document parsing and extraction is about?

Sylvestre Dupont [00:35:06]:
I think in general, it's a very crowded space, players in and out all the time. So maybe with AI we have a few more players, but I don't think it has tremendously changed the game because again, it's not just about the AI. It's also about everything around privacy, data compliance and scaling and these things.

Sylvestre Dupont [00:35:27]:
They are a whole different game. But at the same time, it's true that with everything that happens in AI and things moving so fast, we are trying to think about where we should go in five years and in 10 years.

Sylvestre Dupont [00:35:43]:
And I think we start now to have some ideas or some directions that we would not have picked a few years ago before AI, because. Yeah, because it is definitely shifting our direction. So now the way we want to go is very early, so I won't be able to share about it here.

Sylvestre Dupont [00:36:06]:
But yes, it's to be able to provide more values to the users beyond just data extraction, just in case it gets too commoditized.

Omer Khan [00:36:15]:
Yeah. And do you think that how much of a risk you think there is that it does get commoditized?

Sylvestre Dupont [00:36:23]:
I don't know. It depends on the timeline. Maybe 20 years. Of course, we are gone. I hope in five to 10 years we're still around.

Omer Khan [00:36:31]:
I mean, there's no shortage of these PDFs out there, is there?

Sylvestre Dupont [00:36:35]:
No. And I think we also live a bit in a bubble. If you ask some small business owners, many of them don't use AI today. They don't even know what. Maybe ChatGPT. Yes, they would know. They heard of it, but many have never used it. You should start to talk about Claude or Anthropic or Codex or whatever.

Sylvestre Dupont [00:36:54]:
No. So there is also a lot of inertia that maybe we. Things will take a long time to change.

Omer Khan [00:37:00]:
And I've tried extracting data from PDFs with ChatGPT and a lot of the times only to realize a day later that it didn't actually extract the right data.

Sylvestre Dupont [00:37:13]:
Yes, yes, data quality and controls are something fun with AI.

Omer Khan [00:37:19]:
All right, we should wrap up, so let's get onto the lightning round. I've got seven quickfire questions for you. You ready?

Sylvestre Dupont [00:37:25]:
Yep. Let's go.

Omer Khan [00:37:27]:
What's common startup advice that founders often get that you think is wrong for.

Sylvestre Dupont [00:37:33]:
The ones like us that are spoiled and bootstrapped I would say to say no often to keep your focus, know what's important for you for your growth, and say no to everything else.

Omer Khan [00:37:44]:
What book would you recommend to our audience and why?

Sylvestre Dupont [00:37:46]:
The one I've read most recently that left a lasting impression on me was Black Swan by Nassim Nicholas Taleb. About the biggest risk in your life are the ones you don't know about.

Omer Khan [00:37:59]:
What's something you're good at now as a founder that you were terrible at in sort of the first year of building your business?

Sylvestre Dupont [00:38:08]:
Just be driven by customer concerns. Just don't do what I think is right, but listen and do what they want.

Omer Khan [00:38:16]:
What's your favorite personal productivity tool or habit?

Sylvestre Dupont [00:38:20]:
Now that we are growing as a team, I try to prioritize everything that my team depends on me. So I try to unblock them first before I do what the things I have to do later so that I don't think they are on productivity.

Omer Khan [00:38:34]:
What's a new or crazy business idea you'd love to pursue if you had the time and money?

Sylvestre Dupont [00:38:38]:
Well, if I have all the time, I would retire today. I would try to create a vibe coding tool that can keep things simple and not create ugly codes of thousands of line.

Omer Khan [00:38:49]:
What's an interesting or fun fact about you that most people don't know?

Sylvestre Dupont [00:38:53]:
Well, I like to travel. I travel quite a bit and I like to take a picture of me jumping when I travel. So I have a website called Jumping Traveler and every time I travel somewhere, I post a picture of me on there.

Omer Khan [00:39:07]:
What's one of your most important passions outside of your work?

Sylvestre Dupont [00:39:10]:
Well, travel is one of them and everything related to it. Slow travel, preferably ferries, cruise trains. Love him.

Omer Khan [00:39:19]:
All right, great. So, Sylvestre, thank you so much for joining me. It's been a pleasure. If people want to check out Parser, they can go to parser.com so it's p a r s e u r.com and if folks want to get in touch with you, what's the best way for them to do that?

Sylvestre Dupont [00:39:39]:
Twitter is fine. Tribe is my handle.

Omer Khan [00:39:43]:
Okay. We'll include a link in the show notes. Thank you. It's been a pleasure and I wish you and the team the best of success.

Sylvestre Dupont [00:39:48]:
Thank you very much. Omer, thanks for inviting me.

Omer Khan [00:39:51]:
My pleasure. Cheers.

Related Episodes

Hewitt Tomlin, TeamBuildr

How a Vertical SaaS Hit $10M ARR With Flat Pricing

Hewitt Tomlin and his co-founder James started TeamBuildr as a social app for college athletes. A single conversation with a campus strength coach changed everything. The coach didn't want a social experience for his players. He wanted better tools for his actual job - building and distributing training programs. That pivot turned TeamBuildr into a vertical SaaS company built around one job function: strength and conditioning. Hewitt worked a full-time job for the first three years while moonlighting on TeamBuildr, growing revenue from $20K to $100K. When he went full-time, revenue jumped from $100K to $500K and then to $1M. Today TeamBuildr is a $10M ARR business with 45 employees. They've never raised funding and still use the same operating agreement from 2012. About half their customers are high schools, and they deliberately charge pro teams the same price as everyone else. Hewitt's reasoning is pragmatic. As a bootstrapped company, they couldn't afford to build different products for different verticals. By keeping pricing flat, they got NFL and college logos as social proof that drove volume in the high school market - the segment that actually moves the needle for growth. On AI, Hewitt is taking the opposite approach from his main competitor Volt, which is building AI-generated workouts to replace coaches. Hewitt believes AI should enhance the coaching profession, not replace it. He's focused on using AI internally to help his team work better and building AI features only when coaches actually ask for them - which they haven't yet. In this conversation, Hewitt shares how he built relationships with early customers instead of just handing out logins, why he thinks founders who plateau at $500K have a product-market fit problem, and how building for a job function instead of a vertical unlocked every segment from high schools to pro sports.

Adam Fard, UX Pilot

How a Bootstrapped SaaS Hit $5.3M ARR in Under 2 Years

Adam Fard is the founder of UX Pilot, an AI platform that helps product design teams create and ship great user experiences faster. In 2023, Adam was running a successful UX agency when ChatGPT and LLMs started taking off. He began experimenting with ways to apply AI to his team's design processes and built a Figma plugin that helped users work through UX frameworks and activities. Then during a user interview, someone asked a simple question: "I have all these ideas on my canvas, but can I turn them into something visual? Can I create a wireframe?" That question stuck with him. He started looking around to see if any tools could actually generate wireframes from text input. He found a few products claiming to do it. But when he tested them, he realized they were faking it. They were just swapping existing templates and personalizing the copy. None of them could truly generate a layout from scratch. There was a technical reason for that. Creating wireframes with AI was genuinely hard. So Adam started working on it himself. He explored fine-tuning LLMs, hired AI researchers, and tested component-based approaches. He spent four or five months iterating. Slowly, things started working. The outputs became stable enough to use. He added Figma integration so designers could bring wireframes into their existing workflow. Within six or seven months of that original user question, UX Pilot hit $10K MRR. But growth created a new problem. Adam hired too slowly. At $30K MRR, he kept questioning whether this was the ceiling. He added one engineer, waited, added another, waited again. Looking back, he says he should have hired five people at once instead of dragging out the process. Adam built a bootstrapped SaaS that now generates over $5 million in ARR with a team of 30 and over 15,000 paying subscribers. He proved that a bootstrapped SaaS can compete with well-funded competitors by focusing narrowly on one hard problem - AI wireframe generation for professional design teams - and shipping a code-first product that enterprise teams actually wanted.

Flo Crivello, Lindy

How Repositioning This AI SaaS Unlocked 7-Figure Growth

Flo Crivello is the founder and CEO of Lindy, an AI SaaS platform that lets anyone build AI agents to automate workflows without code. In 2020, Flo Crivello was running TeamFlow, a virtual office startup that raised over $50 million. But when people returned to offices, growth flatlined. With no path forward, Flo pivoted to build Lindy, an AI SaaS platform for building AI agents. The idea came from his sales team asking if AI could automatically update Salesforce. Flo kept climbing the "ladder of abstraction" until he realized he was building an AI SaaS agent platform. In March 2023, he launched with a demo video that generated 70,000 waitlist signups. But the AI SaaS product was terrible. It would send emails that literally said "the user wants me to send an email to 50 software engineers." Users were surprisingly forgiving because they understood they were early adopters. Flo's breakthrough came from repositioning. His AI SaaS started as "AI employee" - too futuristic for the broken product. He repositioned as "Zapier of AI," making the AI SaaS accessible by positioning against something familiar. Within months, Lindy hit product-market fit and grew to high 7-figures. This episode covers the brutal reality of pivoting an AI SaaS, why familiar positioning beats visionary messaging for early adoption, and how to know when your AI SaaS has reached product-market fit.

←All Episodes