How an AI-startup sold a product that hasn't worked yet: the story of TASS

14 October, 2022


TASS is an intelligent video surveillance system that increases offline shop sales by up to 15%. The startup provides two services: Visitor analytics – collecting data on customers (counting the number of visitors and determining gender and age), and CJM – customer movement within the shop (identifying visited showcases, shelves and not popular zines). The startup is in the Seed stage and has been operational since 2019. But the startup's journey was far from typical. 

We talked with project founder Shahzod Umirzakov to find out how it came to be that one company bought their AI product that was only one-third functional.

"A few years ago I was working as a bus conductor for my father," says Shakhzod. There was a shortage – the conductors wouldn't give the proceeds in full. So I suggested that my father install video cameras that would keep track of how many people enter the bus. So my friends and I created MVP – we soldered, tested, and installed the system in 80 buses ourselves, for which we charged $25 a month. But it turned out that our market was very small: there were not many private buses, the state fleet refused, and other countries already had their own solutions. 

It became obvious that it was unprofitable to work only in passenger transport in the region, and we started thinking about scaling up. At that point we began to experience difficulties – the employees had not received their salaries for several months. Finally, our first major client turned up. But he agreed to install our system in his shop chain, provided it would not only count the number of people, but also give him an analysis of customers – their gender, age and how much time they had spent in the shop. So we saw a new vector for the development of technology, and began to look for artificial intelligence developers. 

While we were fine-tuning the processes and looking for new programmers, the team was still working purely on enthusiasm. And the tasks were quite difficult: artificial intelligence learns from a lot of data, and many hours of CCTV footage had to be processed. And so, the day before our product launch we did the final testing. The result was poor – the system showed only 32% accuracy. We were on the verge of failure. 

Of course we could have postponed the testing in order to improve the technology. We decided not to miss out the first money, and decided to analyse the video manually. All night long we watched the videos, counted the number of people, determined their gender and age. And in the morning, we went to a meeting with the founders. 

Surprisingly, the presentation was successful. We received our first order, thanks to which we paid off our debts and paid our team's salaries. But the artificial intelligence still needed some fine-tuning, so for about 2.5 months we continued to manually count shoppers already in 30 branches, while our programmers were training the system. Finally, we achieved an acceptable accuracy of 80%.

Our system is now much more advanced, with 98% accuracy. We are always profitable: we have over 500 connected locations, and are already starting to expand into Kazakhstan, Azerbaijan and Turkey. As it turned out later, our customers had no idea about our scheme. Apparently it was worth going against the rules at the time, in order to finalise the product as quickly as possible and keep the team." 

Now, the mission of TASS is to transform the retail industry in Uzbekistan, so that in-store payments can be made automatically, without a cash register or clerk. Such technology already exists in the US. And in 2023, they plan to launch a pilot in their country.

Telegram   /   Instagram   /   VС.ru   /   LinkedIn