From Developer to 2-Time Successful AI Entrepreneur with Exits: Behamics CEO Valon Xhafa (Part 4)

Sramana Mitra: How did you decide who was going to be the CEO?

Valon Xhafa: It was a common agreement. We didn’t have a concrete discussion about who’s going to be the CEO. For us, the vision mattered more. This was a really important thing. I learnt from my friends that they have a lot of issues with their co-founders about this specific decision instead of focusing on building the MVP.

Having a co-founder with whom you have a common agreement and having open communication is really important. We have different perspectives on different things. My co-founder is more like going step by step and checking stuff. I’m a bit more into going faster and spending more money. We average out in the middle.

Sramana Mitra: You’re very lucky in finding somebody with whom you can have that fluid conversation with such little experience of knowing each other. It’s a very uncommon scenario. It’s actually very risky to start a company with someone you just met.

Valon Xhafa: That’s true. Both of us were aware of that.

Sramana Mitra: When investors gave you money, they had to know who’s going to be the CEO. You took the CEO role?

Valon Xhafa: Yes, these were just formalities. The most important thing for us was to build the product. We knew there was a huge gap in the market. We found some investors who also shared the vision with us. They were flexible on how we run stuff. From friends who are founders, they didn’t have a lot of flexibility. Pretty much, investors are founders instead of just being investors.

Sramana Mitra: Where did the customers come from?

Valon Xhafa: Through our network. That’s how we did it. Of course, the first POC is hard to get. You have nothing. We managed to work through it through our shared network. We were lucky that we got some big brands.

Sramana Mitra: They were willing to give you access to their data to develop your data?

Valon Xhafa: Yes.

Sramana Mitra: That’s another important point – where do you get the data?

Valon Xhafa: Exactly. It’s like a chicken-and-egg problem. We managed to pull it off by using synthetic data to approximate models. When we got the real data, we found out that it’s actually not that far from the synthetic data. That was another big thing. Synthetic data was not widely used.

Sramana Mitra: Not only that, synthetic data doesn’t work as well with all problems. In this case, synthetic data could proxy things fairly well. What was good about your previous company is that the data is public. This is a different kind of data.

Valon Xhafa: That’s true. Even now, we don’t need customer-related data. It’s always hard to build an AI product where you don’t have data, but you need data to build a product to do the pilot testing. You have to be able to navigate through that.

Sramana Mitra: What nuggets did you come up with as you were modeling this data? What were the levers that you could control to achieve what you were trying to achieve?

Valon Xhafa: The initial approach was that we need to come up with use cases. For example, the size wasn’t right. Maybe the product looked better in the pictures. We started working around the use cases and brainstormed how to tackle it.

One thing that we did is, you can guide the customer to get the correct sizing. Later in the process, we found out that if we wanted to develop stuff, we had to change the approach. We had to find new use cases and reasons why customers return products. This is where AI comes in. AI is pretty good at not only predicting behavior but also explaining the behavior.

If we predict that you’re going to return a specific product, you can use the same AI to explain why. We moved from a brainstorming approach to an AI-driven approach to finding use cases. That was a gamechanger. Now we can see use cases that wouldn’t even be addressed. We used the same approach that is used to discover drugs to discover why customers do specific behaviors.

If the customers abandon the cart because they added too much products, which is very common, you leave. You get confused and distracted. What happens is that as soon as the customer has three or four products, you can remind the customer to check the cart. We even compare with coupons and these even outperform coupons. We finally found an approach that can provide us with important results instead of us trying to spend a lot of resources on having a large data science team. Now we have AI that tells us where to look.

This segment is part 4 in the series : From Developer to 2-Time Successful AI Entrepreneur with Exits: Behamics CEO Valon Xhafa
1 2 3 4 5