A review by Sebastian Kielmann, CEO Picalike GmbH
Almost exactly 20 years ago, I switched from research on text indexing and search to image analysis and feature extraction. I was attracted by the fact that images are language-independent and rarely have duplicate meanings. When I first started working on image analysis, I mentioned this to my boss at the time, who had a PhD in physics. His reaction to this was decisive for the 20 years that followed: he said that what I was trying to do was not feasible. There was no Arxiv.org at that time, AI was not very popular, GPUs were not really used for matrix multiplication yet and corresponding papers from research were hard to get if you had an extremely slow line in deepest Walldorf / Wiesloch like me.
But where there’s a will, there’s a way… It was then that I realized how open and willing to share the research community was. There was no Twitter, Facebook, Instagram or anything like that. For me, that time doesn’t seem so long ago, but my kids can hardly imagine such a world. Conversations and exchanges were through conventions or mailing groups. For those who have also been analyzing images for a while: Open-CV was just getting started at that point.
I was researching image analysis ⎼ as a hobby on the side ⎼ and studying management and marketing, working in e-commerce or consulting. Then I met Dr. Tilo Höpker during a flight and we became friends. Over a meal together, I presented my idea to him: a recommendation technology whose basis would come from data from product images, rather than from texts. The idea had come to me because I was working on a product search engine at the time and saw how much erroneous data there was in the product feeds and how much information about the product was missing, even though everyone could see it, based on the product image.
Tilo thought the idea was great, would also jump right in to start a company that offered that. But he wanted to get a second opinion. At the next meeting, Daniel Raschke, who has since become a good friend of mine, was also present. We didn’t know each other at that point, but we were immediately on the same wavelength. The three of us started Picalike GmbH in October 2010. 10 years ago now.
My first lesson was that primarily a market has to be ready for a technology, not the technology has to be ready for the market. Anyone who has ever had to bring a technological product to market knows what I mean. I’m not a born sales guy, like to avoid people and crowds, don’t want to be in the spotlight, and talk a lot about the technology and too little about the benefit. But anyone who founds a start-up also has to take care of sales. The first customers were found quickly. Fashionhype.com and weare.de quickly went from customers to development partners and then friends and part of the family.
In addition to the pure data from the images, customer behavior and correlation among products, as well as filtering by all data about the product, were soon included to provide even better recommendations. Similarity Recommendation (similarity detection, like “similar items”) was then joined by Complete-the-Look Recommendation (outfit recommendation, like “this goes with that”). For us, as for any startup, the time quickly began when you have to do a balancing act between customer support and further development. Not every customer needs the same thing, and we were a fine but still small team. During the day I helped with customer support, in the afternoon I pitched Picalike wherever I could, and in the evenings there was further development. There were days when I thought we were going to take the next giant step (technologically or revenue-wise) and there were days when I wondered if it was really a good idea to leave a secure job. I don’t regret taking the leap into self-employment, but there are many things I would do differently today.
One day my phone rang (it was a Hansenet line back then, who remembers it?).
It was a young guy from the OTTO Group who had read an article in the Chamber of Commerce magazine. We talked and he did some intros on OTTO.de, Bonprix, Shopping24 and others. This gave me the opportunity to introduce our technology to the big players. This phone call was crucial for the next 9 years. I presented not only to the aforementioned companies, but also to the OTTO Group itself, which decided to invest in us. Before signing the shareholder agreement, we had resolved that when it was signed, we would all go mega partying. We had wild ideas about parties at the Doll House on the Reeperbahn or about organizing a big private party. And then this: after the signing, everyone went home, went to bed, slept for a few hours and the next morning the work started all over again.
In no time, we assembled a highly motivated team, moved to a larger office and invested in marketing, product development and research. We quickly grew from a handful of customers to dozens of customers, opened an office in the US (first SF and then moved to NY), started partnerships with prestigious companies, and were pleased with the growing interest. Everything was looking good. We grew, new clients came in, became more and more international and had very exciting projects and cases in mind with us.
As time went on, we noticed that the momentum of our growth was slowing down. Projects were getting more complex, budgets were getting smaller, competition was getting more diverse. Our approach of “customer request -> research -> project -> product” was showing its weaknesses. We had to admit to ourselves that research is fun, but not so easy to scale financially. And so we analyzed the weaknesses of our business model, the strengths of our team and system, and went for a start-up within the start-up. In other words, a soft opening.
Welcome to the world of Portfolio Analysis 🙂 Through several analyses, we realized that we were too dependent on the availability of technical resources at our clients, that we needed to create a quick and easy entry point into our Picalike world, and that we needed to deliver demonstrable, repeatable value to our clients. Thus, we started with multiple A/B tests, case studies, surveys, interviews, and new positions within the team. It was also important that research remained a high priority in our day-to-day. With the help of our customers, partners and the team, we started the development of OnSight analytics. The planned development time was half a year. However, reality is not a good friend of planning: two years later, we launched the open beta. Some challenges were bigger than expected, the day-to-day business could never be neglected, and we had to learn a lot from our pilot partners. But from the beginning we had committed ourselves not to compromise on quality, even if it would take longer, and we still stand by that. Even at a time when Corona has changed all our lives.
The last few months have been like the first few months: a lot of ups and downs. Moods of optimism followed by setbacks. But what remains is that in the last 10 years no day has been like the other, many stories have been written, we have learned a lot, and now for the next 10 years we want to continue to learn and put into practice what we have already learned.
Thank you to the team for the incredible support and loyalty, to our clients for the great and always exciting collaboration, and to everyone else we have met along the way over the last 10 years for enriching our lives and shaping our path.
Stay healthy and take part in the next stages of our journey.