Ep. 199 - What AI Taught One Founder About the Future of SaaS

Guest: Ivan Lee, Founder & CEO of Datasaur

We’re looking at what happens when AI changes the market faster than the old SaaS playbook can keep up.

Ivan Lee, founder and CEO of Datasaur, joins SaaS Backwards to share how his company navigated one of the most dramatic shifts in enterprise AI. Datasaur started as a data annotation platform before ChatGPT changed customer priorities, paused AI roadmaps, and forced the company to rethink its product, GTM strategy, and business model.

Ivan explains why out-of-the-box tools like ChatGPT Enterprise and Microsoft Copilot can be useful starting points, but often hit a ceiling for regulated enterprises that need private AI trained on their own data, workflows, and processes.

He also shares how Datasaur moved from a traditional SaaS model toward end-to-end AI solutions, what founders can learn from disrupted marketing channels, and why the future of SaaS may depend less on selling software access and more on solving the customer’s actual job to be done.

Key Takeaways:

• Why enterprise AI often breaks down when it lacks access to private data and internal workflows
• How ChatGPT disrupted Datasaur’s original AI roadmap and customer base
• Why old SaaS GTM channels stopped working in a crowded AI market
• How Datasaur rebuilt around private, secure AI for regulated industries
• What SaaS founders should measure when marketing “best practices” stop producing results

---

Stalled pipeline? Lost deals? Diagnose your GTM gaps with a free, actionable checkup. 

🔗 Get your free SaaS GTM Checkup 

Transcript

Podcast intro and stakes

Intro/Outro 0:06

Welcome to the Stats Backwards Podcast, where we reverse engineer the success of fast-growing Stats firms and explore strategies CMOs and CEOs are using to drive their businesses forward.

SPEAKER_01 0:24

Welcome to Stats Backwards, the podcast that reverse engineers Sax success for go-to-market and C-suite leaders in the software and AI space. I'm your host, Ken Lempett, and I'm president of Austin Lawrence Group, an advertising and demand generation agency for companies in the space. So what happens to an AI company when a wave of AI for everyone wipes out its customers, invalidates its roadmap, and kills its marketing channels all in about 18 months? Most companies don't survive that. But today's guests did, and what's on the other side is one of the more interesting positions in enterprise AI right now. Ivan Lee is the founder and CEO of DataSource, which builds private, secure, large language models for regulated enterprises, organizations in healthcare, legal, finance, and government that can't send their data to OpenAI or Anthropic. Ivan has been building AI since 2018, years before ChatGPT made it a household word, and he's had a ground floor view of how enterprises have moved from skepticism through experimentation to the realization that off-the-shelf AI simply isn't enough. When OpenAI and Anthropic both announced in the same week that

Meet Ivan and Datasaur

SPEAKER_01 1:41

they were standing up joint ventures with management consultants to deploy enterprise AI, that was the market catching up to what Ivan's been doing for the last couple of years. Welcome to SaaS Backwards, Ivan. But before we dig in, could you please tell us more about your background and your company, Datasaur?

SPEAKER_00 2:00

Yeah, thanks for having me, Ken. Just a little bit more about myself and the company. I come from a computer science background, studied at Stanford University. This is my second startup. I sold my first one to Yahoo, and I've done stints at both the smallest of companies as well as the largest. Prior to starting this company, I was over at Apple building AI and machine learning there for a couple of years. Now, as you alluded to, when we started this company, we were in the AI space, but AI had a completely different definition. We had started off with a data annotation platform that turned unstructured data into structured data. But in 2022, when ChatGPT launched, it fundamentally changed our field. And what we do now is private AI for regulated companies who are not comfortable sending their data to

Why enterprise AI hits limits

SPEAKER_00 2:50

OpenAI and Anthropic for compliance reasons or for competitive reasons. They don't want to be sharing their data, their workflows, and have that all consumed and used to train GPT 6.0.

SPEAKER_01 3:02

So a lot of companies hand their teams an enterprise license to Chat GPT or they get Microsoft Copilot. And that's what they call their AI strategy. And you've watched up close as these deployments hit a wall. I think you need to walk us through what that ceiling looks like. What can out-of-the-box enterprise AI actually do? And where does it break down in a way that matters?

SPEAKER_00 3:27

Yeah, absolutely. And in order to talk about that ceiling, it's important to understand the nature of the models themselves. Now, ChatGPT and all the subsequent models have been trained on everything possible that is publicly available. But that's the key point there. It seems incredibly smart and knowledgeable about everything that is public, but it turns out public-facing information only comprises 2% of all knowledge that humans have ever come up with as a society. The remaining 98% is Bible. It's within these industry firewalls behind your organization's security and hosted within your own infrastructure. And so what often happens with our clients and customers is that they use, you know, Copilot, Enterprise Chat GPT. It's a wonderful starting point, right? That's really where everyone needs to start. And they're using it to correct their emails or upload a document and answer questions about it. It's a great starting point, but they start feeling the limitations of what the bot is able to answer. Because at some point, it's not enough to be asking based on the context of what's out there on the public internet. You need to answer based on all of the legal contracts that have been signed with your organization. You need to sign based on the internal wiki and handbook for how customer success is handled specifically within your department. And so that is when they start realizing hey, this technology seems powerful and maybe I'm even doing more with Chat GPT at home, but I just can't get it to, I need to do more and build up more. And so when people are coming to DataStore, they have either a capabilities wall where they need it to be integrated with certain things, be able to control their browser, control their computer in order to reach those heights, or they've come to data workflows that are so sensitive that their security teams, their legal teams would never allow them to upload those documents. And that's where we come in, we build and

Private LLMs and workflows

SPEAKER_00 5:41

deploy these open weights models that are already quite powerful, but now they're within your organization's infrastructure. The data's never leaving your servers.

SPEAKER_01 5:59

Are those also probabilistic LLMs or are they deterministic?

SPEAKER_00 6:04

It's the same fundamental technology. And so everybody in 2025 heard about this model from China called DeepSeek. It turns out there's actually a dozen of models, some that are trained in China, some in Europe, some in the US. For example, in the US here, we have Google's Gemma model. So they're the same fundamental technology, but they're condensed and they are able to be deployed directly to these bare metal servers.

SPEAKER_01 6:33

Got it. So this is the opportunity for that either chat or agentic experience against my own data in a safe environment. Is that right?

SPEAKER_00 6:45

Exactly. It's not only just the data, but it's also trained on your workflow. Let's say, for instance, that you're an insurance company and you have a particular way of underwriting every single insurance claim. That is something that you would not want to partner with OpenAI and kind of give your company secret sauce over to them. So that's something where you might work with us, build and deploy a private AI that you own forever, that is never being shared with any of your competitors.

SPEAKER_01 7:13

Got it. So my methods remain

Datasaur origins in 2018

SPEAKER_01 7:16

private as well as my data. And that allows me to keep my competitive advantages based on that corporate history. I want to take you a little back in time. Let's go back to around the founding of DataStore in 2018. And you know, ChatGPT was, you know, in nobody's universe, right? Didn't really exist. And now everyone is an AI expert, right? Which is sort of an interesting phenomenon. But I think you got to tell us what your company looked like in those early years, the problems that you were solving, and how the market was responding to you before generative AI kind of changed everything.

SPEAKER_00 7:55

Absolutely. So for the first two and a half years, as I alluded to briefly earlier, we were a data annotation platform. So you would come in and take unstructured documents that could be a PDF, a lengthy Word document, really just anything that wasn't in a database or like spreadsheet format, and you would annotate it, extract the information that you need, and use that to train machine learning models. Now remember, back then, machine learning was already something that people believed could be quite powerful. I started this company believing that machine learning and specifically the domain of natural language processing, all AI related to text, documents, and audio, would be quite powerful. But even I didn't foresee that, you know, it NLP could be taken to this level. And now it's it's seemingly the only thing that people talk about in the tech industries. And so for its first two and a half years, we had a standard SaaS platform. It was the most comprehensive platform in the industry. We had competitors, we were out competing them, and we were steadily growing. And it felt like we were on our pathways to a standard Series

ChatGPT shock and churn

SPEAKER_00 8:59

A and Series B.

SPEAKER_01 9:00

Yeah, that's that's where you were heading, right? Exactly. So then ChatGPT comes along, supposed to be a gift for firms like yours, right? Should have been an enabler in a lot of ways, but it actually puts you into a kind of a survival mode. As you told me when we did our prep, over an 18-month period, customers paused, churned, and the category shifted like right beneath you. Can you take us through that experience, why it happened, what that was like on the inside, what it cost you, and kind of set up what you did next?

SPEAKER_00 9:35

Yeah. So, like you said, a lot of people were telling me, wow, you guys were doing AI before. It was cool. This must be great for you. But under the surface, what was happening is that every single AI project that we were supporting for the last two to three years was suddenly put on hold. Every single client of ours was re-examining their AI roadmap, being told to pause all existing efforts and check out this new Chat GPT technology. And so we were churning customers as quickly as we were bringing them in. And it was a highly volatile time for us as an organization overall. And we weren't sure when this was going to end, what it was going to look like on the other side. And so we ended up creating a second platform specifically for LLMs. And we were hoping we could recapture a lot of the same success and drive that forward. But that also never reached commercial viability. Uh, what happened at that point was that everything was already moving so quickly that the data scientists and engineers who should have been our paying users didn't have the bandwidth to try something new. They also didn't have the political capital internally to try and convince their executive stakeholders that they should be paying five or six figures for a year-long license to any platform. And so if you look at the last, you know, two to three years, all these dev tools, a vast majority of them start with open source. That was the only thing that our customer base was willing to really invest in. And we had taken the completely wrong kind of commercial and business model approach to selling that second product.

SPEAKER_01 11:21

So your second product was an LLM platform, right? And that's a big shift

Pivot lessons and open source

SPEAKER_01 11:29

from where you were. But you did keep it proprietary. What was the lesson in that shift for you? You know, you sort of anticipated this question a little bit, but what was the shift that you had to make? And what's a lesson for founders about, and this is I think the hard part, recognizing a shift as it's happening to you.

SPEAKER_00 11:49

Frankly, we were just taught that lesson a little earlier than everyone else. We're here in 2026 and we've just been through what people are calling a SaaS apocalypse, right? I think for us, we had to learn the lesson the hard way that SaaS, as we know it, may not be around for very much longer. With the decreased cost to building software, what we had was very attractive to a lot of our customers. A lot of people were on the free trial. They loved the concept. In fact, this past week, a company called Open Router just raised a massive unicorn round, doing exactly what our proprietary platform did. The only difference is they went the open source route. They made it available, got the huge customer adoption base first, and then upsold to the enterprises who wanted to deploy this at scale. And that's what I should have done. Instead, we tried to go with the old SaaS playbook, charge people $100, $500 a month, and build the user base that way. And people just, they didn't have the time to assess and commit, let alone the, again, the political capital to ask for the budget for any kind of product like that. And so what I realized is first, if you're building DevTools for the AI world, it should absolutely be open source. The second thing that actually forced us to kind of move forward and evolve as an organization is we didn't have people who are willing to pay for the platform, but they were willing for us to use our own platform and build an end-to-end solution for them. And this concept has been popular in spirit in Silicon Valley for some time. People call it the job to be done. Well, for us, now with the cheaper cost of software development and our own powerful platform that we had built out, we could go out there and solve the job to be done for that end customer. So instead of asking their engineers to use our platform, let's build this for you and we can charge the end customer, the business unit owner, three times the cost, five times the annual recurring cost for an end-to-end solution.

SPEAKER_01 13:56

The uh Palantir forward-deployed engineer idea, right? Exactly. Before we

Sponsor break

SPEAKER_01 14:03

continue, I just want to have a short word from one of our sponsors on the podcast. And we'll be right back with Ivan Lee, CEO and founder of Datastore.

SPEAKER_02 14:13

If you're building a SaaS company, here's some data that's certainly worth paying attention to. According to Kyle Poyard's research across 6,500 software companies, only about one in five ever reach 5 million in ARR. And just one in 10 make it to 10 million. Now, those are some pretty sobering numbers. If you've got funding in a solid product, but you're still missing revenue targets, the culprit is almost always somewhere in your go-to-market. Now, maybe you're losing too many deals to no decision, and many times pricing hasn't changed and it's opened the door to competitors. And often sales and marketing are hitting their activity KPIs, but that's where the good news ends. Now these are all solvable problems, but you have to know where to look. And that's exactly why we built the SaaS Doctor's Go-to-Market checkup. It's a free diagnostic where we assess 12 critical components of your growth engine, from positioning and pricing to your sales tech and metrics. We'll come back to you with a clear picture of what's holding you back and what to prioritize next. No 80-page decks that you'll never implement, just a sharp, actionable read on why you're stuck and what needs to change. So if your product should be growing faster than it is, check out the link in the show notes and let's talk. And now back to the podcast.

SPEAKER_01 15:35

So before the break, we were talking about this kind of shift

GTM collapse and survival revenue

SPEAKER_01 15:39

that happened underneath you. And there was more to it than what we talked about. There was actually something of a nightmare scenario if you're in demand generation. Your go-to-market basically stopped cold, right? Cold outreach stopped working. Google Ads was delivering zero leads for months, despite you know, a significant spend for where you were in the company, you know, and its growth. And you said to me that you felt your messaging no longer differentiated you from a large number of competitors. First of all, like how do you keep the lights on and find a path forward when the playbook that got you to a certain point stops working entirely?

SPEAKER_00 16:20

Absolutely. Just to elaborate on that, prior to ChatGPT coming in and disrupting the AI industry, our go-to-market strategy was well on its way, I would say. We were getting four to five qualified leads every month that would help us reach our quarterly goals. And I thought this is we're we're on the treadmill of SaaS growth and everything was going smoothly. And then not only did ChatGPT change the definition of AI and stop all those projects, suddenly our Google ads were producing zero results. Our cold outbound, because everybody was able to use Claude and ChatGPT to formulate their messaging, our cold outbound was no longer even being read by our target profiles. And then on top of that, everyone was pitching AI, right? And it was really hard for us to differentiate and share that we had something of value to offer. And so everything about marketing broke down. Now, to answer your latter question first, in terms of keeping the lights on, this forced us back to fundamentals. I'm a product manager. I was actually very resistant to the idea of doing anything that seemed unscalable. And in my mind, I'd been trained to believe that SaaS was the only scalable path forward. But when I was faced with this existential crisis and somebody was willing to offer us more money for a custom development project, something that I would have turned down a year earlier, suddenly we were desperate enough to accept. And it forced me to be more open-minded and say, okay, wherever the revenue is, we just need to, as you said, keep the lights on. And we needed to take that money and continue chasing revenue wherever we could find it. And that going back to kind of our humble roots as a startup and listening to customers and what they're willing to pay for is what eventually put us on the path that we're on today.

SPEAKER_01 18:19

Yeah, I think that's a really interesting perspective that you have to go back to the founding kind of hungry state and find revenue where it is and rebuild the business around the revenue that is actually available. I think people in my business are facing the same kind of crisis, you know. Certainly in content creation, not too many people want to pay a premium to have their blogs written by our experienced writers. Some are, but it's it's definitely reduced the margin in that business and the strategic interest and importance of content creation. So it makes sense. Kind of further on that on that theme of you know building solutions for your customers. Just a couple of weeks ago, and I guess about a

Next wave enterprise agents

SPEAKER_01 19:06

few weeks before we're gonna publish your episode, OpenAI and Anthropic announced joint ventures with management consultants to help enterprises deploy AI, which sort of validates exactly where you've taken the business and have been doing for a couple of years now. Now that the vocabulary has caught up with reality, what's the next wave of enterprise AI adoption going to look like? And where does a company like DataSort fit in?

SPEAKER_00 19:36

Absolutely. So we have to keep in mind in our world, maybe a majority of our friends and network are all in on AI. We understand and see the value. But we have to keep in mind that only 15% of the world is using this, even on a weekly basis, right? In the broader scheme of things, this is based, by the way, on the number of weekly active users reported by both Google and OpenAI. And so there is this very famous framework called crossing the chasm in the business world. And we have not crossed that chasm with AI just yet. And we're still working on it. And with OpenAI and Anthropic starting those joint ventures, I think that's going to push adoption forward. What they essentially confess to the world is that enterprise Chat GPT and Cloud Code out of the box are not cutting it. They're not so good that enterprises adopt it themselves. They need the help of these armies of management consultants to go in and understand the culture, the processes, the where the data is stored at every single one of these clients and implement something truly custom for that customer. Now, let's put that aside for just a moment here. The other major observation is that chatbots as a form of interacting with AI are an opt-in technology. What do I mean by that? You have to go teach your uncle, your coworker, the right time to be using a chatbot. And they have to remember in that moment, oh, I should use that chatbot in order to ask this question, as opposed to reaching for whatever tool they were using previously. Agents, which are now all the rage in Silicon Valley, this is an opt-out technology. Agents are running 24-7, reading the documents, doing all this work behind the scenes, and reaching out to the employee when they need the humans' inputs. And so the fact that it's switching from opt-in to opt-out means that adoption of this technology will naturally increase. Depending on the market reports you're reading, in 2025, it was between 5% and 30% of enterprises are successfully using AI on a weekly basis. But by the end of this year, every enterprise that is adopting these agents, it's going to be 100% adoption overnight. Just because of the nature of them being on 24-7. So, with all of this, I think adoption is going to accelerate a lot faster than people anticipate. This has a lot of second and third order consequences. The GPU crunch is going to be even worse than we're feeling in the first half of this year. But I also think this is a great opportunity for a company like ours to position ourselves as an alternative to those joint ventures. But we're going to be model agnostic. We're not going to lock you in to open AI and enthropic. We're not going to be using your workflows and your data to train the next generation of their models. We can set you up with something that you own rather than AI that you continue renting from others.

SPEAKER_01 22:44

I think that's an interesting proposition. You know, when we do positioning work, we need something to bounce against or to vilify, right? So here the villain is losing control of your data, right? That's the fear that you can, I think, profitably sell against. And I think it's a real issue for so many companies. They don't even have to be regulated. I mean, imagine Coca-Cola probably doesn't want its formula seeping out into Claude or ChatGPT. So I think there's a lot of opportunity to position against sharing your data to train models. I think that's like a primal instinct almost. People want to protect their corporate data and processes. Feels like a good one to me. I have one more thought I wanted to ask you to talk about, which is you've sort of been through a full cycle here. You've been a market leader, you've been disrupted, you had a pivot, you did survival mode, and now you have a clear path forward.

Founder advice and measurement

SPEAKER_01 23:47

For a SaaS or AI founder listening right now, who's in the middle of some kind of chaos of their own creation, or that's been foisted on them, and they can't yet see that other side. What's the one thing you'd tell them that nobody told you?

SPEAKER_00 24:03

There's a lot, but I think at least pertinent to the themes of this conversation. One thing that I realized is I spent the last two years. So actually, uh Ken, I come from a product management and engineering background. This is my first time doing B2B. I spent the last couple of years learning all the marketing techniques that were successful for the last decade. But what I didn't see at the time was that go-to-market techniques themselves were being fundamentally disrupted. And so what was more important for me, and this is something that I realized during a long haul flight, is I need to stop relying on what's worked for others in the past and kind of extract what the fundamentals are of marketing. But what we do now is we track every dollar that we spend on marketing. And that dollar can be spent with Google, with LinkedIn, on a conference, on a roundtable dinner, on whatever it is. Track every dollar and track every conversion and just figure out what is working there. Right. So this is very helpful for me from my like engineering background and mentality. It's just if something's not working, even if it's considered like table stakes and hey, this is what every other company does, doesn't matter. Throw that out the door. What is the market reacting to, or in this case, what's leading to actual sales conversions, and spend your time and money there.

SPEAKER_01 25:28

Sort of an inside baseball question on that. Are you looking at the full life cycle, like all the conversion events that a prospect might go through?

SPEAKER_00 25:37

We understand that there are certain things that are still geared towards top of funnel, and there are certain things that lead to direct sales conversations. So we do need to calibrate for that. And we have equations to do that accordingly.

SPEAKER_01 25:47

Yeah, that's the hardest part sometimes. You know, people get stuck on last touch, think that that's their best channel when it may be a great closing channel, but it might not have started the conversation. And speaking of conversations, this was an awesome one. Ivan, I want to thank you for being on SAS backwards. If people want to learn more about your company or reach out to you, how can they do that?

SPEAKER_00 26:10

Best place would be to reach out either via email. That'll be ivan at datasore.ai. I'll put that out there. You

Wrap up and contact info

SPEAKER_00 26:17

can send me a direct message, or alternatively on my LinkedIn profile, it'll be profile tag as i lee. So feel free to reach out either way.

SPEAKER_01 26:26

Great. And also obviously go to the datasore.ai website. If people want to reach me, I'm on LinkedIn slash in slash KenLempit. My advertising and demand generation agency for SaaS and AI companies is Austin Lawrence Group. We're at AustinLawrence.com. And I hope that Ivan's episode has convinced recalcitrant listeners to subscribe to the podcast, which you can do wherever podcasts are distributed. Ivan Lee, thank you so much for joining us on SAS Backwards.

Intro/Outro 26:55

Thank you, Ken. Thanks for listening to the SaaS Backwards Podcast, brought to you by Austin Lawrence Group. We're a growth marketing agency that helps SaaS firms reduce churn, accelerate sales, and generate demand. Learn more about us at www.austinlawrence.com. You can email Ken Lempett at kl ataustinlawrence.com about any SaaS marketing or customer retention subject. We hope you'll subscribe and thanks again for listening.