Building a Generative AI Venture from Portugal: João Aroso, CEO of Leadzai (Part 3)

Sramana Mitra: So then how long did you do this business? And what kind of revenue levels or what kind of scale were you able to achieve?

João Aroso: We did that business for two years and one of my co-founders from my first startup is still there today. I left, but the company is still very healthy.

Sramana Mitra: And what did you leave for?

João Aroso: I left because I wanted to create a startup from scratch.

Sramana Mitra: OK. And what was the premise of that?

João Aroso: I don’t think it’s my thing to do restructuring. It’s an interesting business, but it’s definitely not the same thing as being an entrepreneur and being growth-focused. So, that definitely wasn’t for me. My co-founder was doing an excellent job there, but it wasn’t the right fit for me.

Sramana Mitra: So, this new startup that you left for, what was the premise of that? What did you want to build there?

João Aroso: It’s what I’m doing today. That’s Leadzai. We built a customer acquisition platform that uses AI to create campaigns automatically. It makes sense how all my experience came together to this because I’m very interested in supply and demand for SMBs. How do SMBs acquire customers? This is literally what I’m doing today. They’re customer acquisition campaigns. We do lead generation powered by AI.

Sramana Mitra: Go back to when you were leaving the previous company and tell me the story of how you built it. This is 2024. This story starts in what, 2016?

João Aroso: Yes

Sramana Mitra: Go back to 2016. What’s happening in the market? How do you start this new company? What are the details of the story?

João Aroso: In 2016, I realized that restructuring the business is not for me. I leave to create a startup from scratch, which is what I like to do.

I struggled to identify a solution that was easy for SMB advertising. When I started looking at what I wanted to do for my first startup, I understood that I was very fascinated by digitalization of SMBs and how SMBs could be present online and show themselves to customers.

In 2016, website was a commodity. Everyone had a website. Back in 2009, that was probably a significant part of the challenge and a high percentage of SMBs didn’t have a website.

Fast forward roughly ten years, everyone has a website. But the point is, having a website is no longer a distinctive factor, nor relevant. Traffic is relevant. So, how do you create efficient campaigns online to acquire the right traffic for the website?

So, that’s what I wanted to focus on in 2017. That was the main challenge for me. So, we started looking into machine learning, AI, and how those growing technologies could help us generate the best possible traffic for SMBs.

For probably two years till around 2019, we were trying to figure out our space within the advertising space. So, we raised some money from some early investors, 1.5 million in Lisbon, now this time from business angels and early-stage VC. We started building a platform to create campaigns. So, in 2018, we even developed our own LLM to be able to create ads on behalf of the SMBs under the assumption that everyone has a website if they’re running campaigns online so we can scrape the information from the website, feed that to our LLM and automatically with that, create a campaign, an advertising campaign that would be optimized towards acquiring the best possible traffic

Sramana Mitra: When you call it LLM, did you call it LLM then or are you using today’s parlance to describe something you did then that was different?

João Aroso: I don’t think we called it LLM. I know that we discussed NLP as a functional area of the company.

Sramana Mitra: In terms of specs then, you were trying to take natural language input, process that and generate campaigns based on that. Is that what the workflow that you were trying to achieve was?

João Aroso: Yes, I remember us using Wikipedia data and free newspapers data to train models, to understand keywords, to generate ad copy, and to do sentiment analysis. So, we did train a model to achieve those results, but I don’t recall honestly if we used the acronym LLM.

Sramana Mitra: That doesn’t matter. I am trying to understand what’s working in this platform.

João Aroso: Yes, but it was an NLP model.

Sramana Mitra: Did it work?

João Aroso: It worked decently. It was not amazing, but decent enough for us to launch an MVP, but not enough for us not to give up on it when GPT-3 came out. So, when we saw GPT-3, we thought we’d use GPT-3 because it’s a waste of our time to keep on developing our own model.

Sramana Mitra:  So, bridge me to the point. You started in 2017. You were playing around with the industry to figure out where you wanted to position and what you wanted to do. Around 2019 is when you launched this MVP. In that period, you had a $1.5M investor funding with which you built all of this.

Then GPT-3 comes out in 2020, and that’s when you decided to switch the core engine and plug GPT-3 in.

This segment is part 3 in the series : Building a Generative AI Venture from Portugal: João Aroso, CEO of Leadzai
1 2 3 4 5 6 7