Shakhzod Umirzakov and Jamshid Khakimjonov from Tashkent launched their startup, TASS Vision, in 2021. Their team developed AI-powered cameras for stores that use computer vision to track foot traffic, identify shoppers’ gender and age, and help store owners boost their profits.
Today, TASS Vision’s solutions are installed in over 2,000 stores across 10 countries, including Kazakhstan, Uzbekistan, Peru, and Turkey. So far, the startup has already attracted $1.8 million in investment.
For the joint project 100 Startup Stories of Central Asia by Digital Business and Astana Hub, TASS Vision co-founder and CEO Shakhzod Umirzakov shared how he once manually counted visitors to train the neural network, why AI cameras are a game-changer for retail, and what’s next for the startup when it hits a $100 million valuation.
«My team and I were installing equipment on buses at night»
— How did the idea for TASS Vision come about?
— Back in 2018, when I was 19, I was helping my dad with his business. He ran a small bus depot and worked in passenger transportation. One of the biggest problems in that industry is that bus attendants often pocket part of the revenue, and it’s hard to prove. To fix that, I suggested a solution that could count passengers.
It was an infrared counter — when a passenger walked through the door, they crossed an invisible beam, and the system recorded their entry. I spent about three months building the device and installing it on the buses. Once the attendants realized their cheating could be detected, they started working honestly. Within two months, revenue nearly doubled, and the project fully paid for itself.
Then I put together a team and we started onboarding other bus depots. I remember how we’d spend our nights installing equipment on the buses and crash in the mornings. During the day, we wrote code, 3D-printed the device casings, soldered microcontrollers, and then headed back to the depots in the evening. That project eventually sparked the idea to analyze shopper behavior in retail.
— Why didn’t you stick with the transportation industry?
— In 2019, while I was in Almaty, I noticed they were already using transport cards — the Onay system for cashless payments on public transport. That’s when I realized something similar would soon come to Uzbekistan. Once cash payments disappeared, our device wouldn’t make much sense anymore. So I figured it was time to move on and switch industries.
I started looking for a new idea. I made hundreds of cold calls to retail stores, construction companies, cafés, and restaurants to figure out where there was real demand for analytics. I’d just ask one simple question: “How many people visit your store? And how do you count them?”
The reactions were mixed. Some people would say, “Why do we even need that?” But others were genuinely interested. That’s when I realized that in offline retail, hardly anyone really understood who was coming into their stores or how that traffic affected sales. I saw an opportunity to make in-store analytics as accurate and accessible as it is in e-commerce. Not long after, Jamshid Khakimjonov joined me. He had a strong engineering background and experience with computer systems. Together, we started building the first version of what would later become TASS Vision.
— Who was your first client?
— Our first client was Texnomart, a major electronics retail chain in Uzbekistan. Their head of marketing told us they were already using a system to count how many people entered the store — but what they really needed were deeper insights, like age, gender, and customer behavior. So we promised that within three months, we’d develop a new device using smart cameras and computer vision to analyze video streams and turn them into real-time analytics. Not just how many visitors came in, but who they were and how they behaved.
We planned to finish the project in three months, but in reality, training the neural networks and developing the algorithms took more time than expected.
To meet the contract terms, we had to get creative. We were supposed to deliver daily stats, so we told the client that AI was identifying visitors. But in reality, I was doing it manually. I watched 12 hours of store footage at high speed on two screens, and in about an hour and a half, I processed a full day’s data. I counted visitors, determined their gender, logged everything in Excel, built charts, and sent out reports. I did that every single day for 99 days straight. As they say — fake it till you make it. It was the only way to keep the client.
But that experience actually helped us understand how AI should work. By comparing my manual results with the model’s output, I saw that the AI was only about 50–60% accurate. We found the errors and kept training the algorithms. A few months later, the system finally started working on its own
«For two years, we worked on designing our own smart camera — from the circuit board to the optics and the casing»
— Sounds like the early days weren’t easy. You didn’t have the team or resources yet, but the client was already expecting results. How did you manage it all and build up your AI expertise along the way?
— When it came to expertise, at first we had no real understanding of how AI worked — neither me nor the team. This was back in 2019–2020, before ChatGPT and the big AI hype. There were hardly any specialists in Uzbekistan or even in nearby countries. So we basically started from scratch, learning through YouTube, articles, forums, and Google. We just kept trying, failing, and trying again.
Progress was slow because we didn’t have mentors or the resources we needed. Training neural networks requires serious computing power, which is expensive. So we had to find workarounds. For example, our Head of AI, Sirojiddin Nuriev, was studying in South Korea at the time and helped us access the computing resources of his university’s lab, where students were allowed to run experiments. That’s where we trained the first versions of our model.
Step by step, we built up our in-house expertise, brought the model to stable accuracy, and were finally able to move from experiments to a full-fledged product.
Today, we have around 40 employees, including 12 engineers at our main office and 2 researchers with master’s and PhD degrees in AI and computer vision from universities in South Korea.
— What are the advantages of the TASS Vision system?
— Most stores today still use basic sensors that react to any kind of movement. Because of that, the system might count someone as a customer even if they just walked past the entrance or waved a hand nearby.
The error rate of those basic sensors can reach 40–50%, but they’re still widely used because they’re cheap — and retail is a low-margin business. AI-based analytics, on the other hand, is much more accurate, but it comes at a higher cost, usually between $500 and $1,000 per store per year.
We came up with a solution that combines the best of both worlds: the high accuracy of AI systems and the low cost of simple sensors. This became possible because all the data processing happens directly inside the camera.
— How does the system actually work?
— Usually, AI systems connect to existing in-store cameras and analyze video through the cloud, meaning the footage is sent to a server for processing. That setup is expensive because you have to pay for cloud storage and computing. We decided to take a different approach and moved all the processing directly onto the devices themselves. As a result, the cost dropped by a factor of ten, and the data is now processed in real time.
For two years, we worked on designing our own smart camera — from the circuit board to the optics and the casing. In a way, it’s similar to how Apple operates: we develop the design and specifications in-house, then outsource production to partners in Shenzhen. After that, we install our own software and firmware on the devices.
With a lifespan of five years, our camera ends up costing the store just $2.50 per month, compared to the $250 typically paid for cloud-based AI solutions. That’s nearly a 100-fold difference.
«A 1% increase in conversion at just one store is enough to cover the cost of our system»
— How many clients are you working with right now?
— Today, our systems are installed in over 2,000 stores across 10 countries. We brought in clients directly, without using intermediaries. We knew exactly who we were targeting — non-food retail chains like electronics stores, pharmacies, and clothing retailers. There aren’t that many of them in each country, maybe up to 100, so we went after them one by one. We built a list, made calls, and reached out directly to their offices. Step by step, we grew our sales and reputation through personal meetings and word of mouth, not through marketing.
— What exactly does the system measure, and how does it help retailers?
— We don’t just count visitors — we deliver insights that help boost sales. For example, the system shows how effective marketing campaigns are by tracking how many people came in after seeing an ad, and how many of them actually made a purchase.
Marketers can see which campaigns actually worked and which ones just burned through the budget. The system also calculates customer acquisition cost (CAC), something that was nearly impossible to track in offline retail before.
— How does the data impact the work of sales assistants?
— We help stores measure conversion. For example, if 100 people walk into a store and 5 make a purchase, that’s a 5% conversion rate. Now, managers can actually see those numbers and factor them into employees’ KPIs.
When sales assistants realize the camera is “watching,” they start paying more attention to each customer — and that directly impacts revenue. Poor service is one of the main reasons for low conversion. Often, someone walks into a store and no one approaches them, asks questions, or offers help, so they leave without buying anything.
— Do you have any examples where this had a major impact?
— Yes — one major electronics retail chain completely changed its motivation system after using our data. They removed fixed salaries and based bonuses entirely on conversion rates. It was a risky move, but it paid off: sales jumped by 63% in just three months.
In essence, the data helped them build a transparent and performance-driven motivation system. And for large chains, even a 1% increase in conversion is enough to cover the cost of TASS Vision. On average, our cameras pay for themselves within six months.
— What’s the startup’s business model — how do you earn revenue?
— We offer four subscription packages — Standard, Plus, Pro, and Enterprise — with monthly fees of $29, $49, $69, and $119, depending on the features included.
The Standard plan is our basic package. It includes visitor counting, age and gender detection, and group recognition (for example, a family or a group of friends is counted as a single visit). It also features employee identification using badges to keep the data clean, plus access to a mobile app, email notifications, and basic statistics.
The Plus package includes more advanced features and reports, such as automated AI-generated insights, benchmarking, and POS integrations. The Pro package takes it a step further with face recognition, which makes it possible to distinguish between new and returning customers and gather deeper analytics for each store.
The Enterprise plan is designed for large retail chains and includes personalized modules, custom integrations, and priority support.
Currently, around 70% of our clients are planning to upgrade to the Pro plan, which includes five additional modules.
«ChatGPT actually recommended us to a partner in Peru»
— Let’s talk about investment. How much funding have you raised so far?
— In the beginning, we didn’t have any capital, so we invested our time and knowledge. We relied on client prepayments to keep things moving. Our first $100,000 came from three angel investors, one of whom is the founder of the Texnomart retail chain. Later, we received support from UzVC, and Kazakhstan’s Activat VC invested $150,000. Our participation in 500 Global brought in another $100,000. Altogether, we raised around $350,000 in the early stages, enough to launch the first version of the product and start scaling.
In 2025, we closed a $1.5 million funding round. The lead investor was the European fund Purple Ventures, with a $500,000 check. The round also included Pragmatech Ventures, IT Park Ventures, Big Sky Capital, SABAH.fund, and several others — eight investors in total.
— How difficult is it for a startup from Central Asia to attract European investment?
— It definitely wasn’t easy, even though everything looked great. We had an 86% margin, our ARR was close to $1 million, and we had a clear business model. Still, I personally reached out to over 250 venture funds, and almost all of them turned us down. Most said something like: “Sounds interesting, but you need traction in Europe or the USA.” In Central Asia, it’s relatively easy to raise a pre-seed round, a bit harder to raise seed, but entering Series A is almost impossible. There are very few large funds in the region willing or able to do full due diligence. The turning point came in December 2024, when we met with the European venture fund Purple Ventures. After six months of negotiations, they agreed to lead the round. That changed everything. Once they joined, other investors followed. Purple Ventures became the first independent European investor to make an exception for a startup from Uzbekistan. They believed that global tech companies could come from Central Asia.
We used the funding to purchase equipment, build out our R&D team, and expand into new markets.
— When did you start expanding into international markets, and how do you manage operations across different countries?
— After Uzbekistan, our first international market was Kazakhstan in 2023. Today, our systems are installed in over 10 countries, including Kyrgyzstan, Azerbaijan, Mongolia, Tajikistan, Turkey, Saudi Arabia, the UAE, and Peru.
Retail faces the same core challenges everywhere — understanding who’s coming into the store, how staff are performing, and what influences conversion. The only real differences are language and time zones: in Peru we speak Spanish, in Turkey it’s Turkish, and in Central Asia we use Russian.
We use two models. In countries like Kazakhstan, we establish a direct presence. We open a local company, hire a team, and handle sales and support ourselves. In other markets, we build a network of distributors. We provide them with the product and full training. They earn a 20–30% margin and help us scale more quickly. That’s how we entered Kyrgyzstan, Azerbaijan, Mongolia, and Tajikistan. By the end of the year, we plan to expand into Georgia and Armenia as well.
Sometimes, partners find us on their own. For example, in Peru, we were contacted by an electronics retail chain. They said ChatGPT had recommended us. We sent them cameras and ran a pilot in 50 locations.
Now, they’re preparing to scale the system across 1,100 points of sale.
— What are your plans for the near future?
— By the end of the year, we plan to finalize the Pro package, which includes face identification and tracking the time customers spend in-store. We also aim to onboard new partners in Armenia and Georgia.
In the next 3 to 5 years, our goal is to onboard 15,000 points of sale worldwide, grow our annual revenue by at least 150%, and reach a $100 million valuation. After that, we plan for an M&A, either merging with or selling the company to a strategic partner, most likely a major player from the USA. For me, that’s a natural next step. TASS Vision is my first startup, and after this, I plan to build even bigger companies.
— What areas are you interested in exploring next?
— I’m interested in agricultural technologies, water supply solutions, and sustainable development — areas that will shape people’s lives over the next 50 to 100 years. I want to build products that stay relevant for the long term and can be used by millions of people around the world.