Data-driven dating: how we launched a corporate dating service at Sber

If you are interested in AI, my cases and experience, subscribe to my Telegram channel, where I share insights, practical advice, and the latest news from the world of artificial intelligence.

Let's help data engineers find happiness

In 2019, I led a team of data engineers at Sber. As I got to know the team, I noticed that the guys looked... tired. Of course, they were all professionals, their tenure in the company was longer than mine, but there was a certain lost look in their eyes. When I talked to them a little longer, it became clear that they were just lonely. Some didn't have enough time for their personal lives, while others had already forgotten what it was like to meet people outside of work chats and meetings. People around were getting paid, completing tasks, but they were not truly happy. I had one colleague in my team, let's call him Lesha. A smart guy, he taught at the university, loved the theater, but he had a problem: his personal life was not going well. He tried to meet people on Tinder, and according to him, he spent more than 30 thousand rubles in a week on dates with girls. However, he did not have much success. Moreover, he was very busy: work, preparing lectures - he had almost no extra time. Once he expressed the idea that he would be helped by a service that could find not just random people, but those who share his interests and values.

It so happened that my team and I were engaged in data analysis tasks in the strategic department, using HR data. Therefore, we had detailed information about all the bank's employees: general information, education, psychological test results, financial and socio-demographic data, as well as the digital footprint of employees from various corporate systems. We used this information for management dashboards, conducting ad-hoc analytics, issuing recommendations for strategic decision-making by the company's management, and developing ML models to assess team effectiveness and predict staff turnover.

Bored? Maybe it seemed so from the outside. But I was increasingly convinced: this data is a real treasure, allowing to personalize the experience of each employee. Not only to make HR processes more efficient, but also to help people in something more personal.

After numerous conversations with colleagues on the topic of loneliness, incomprehensibility, and other things, I came up with the idea of a pet project: I decided to try to apply machine learning to organize acquaintances within the company. It is no secret that in large corporations people often remain in their "bubble": they communicate with a certain circle of people, rarely make new acquaintances, although there may be dozens or even hundreds of people close to them in spirit nearby. In fact, I decided to build dating on ML.

Idea of a corporate dating service

In "Notes of a "nerd", or why I don't have a girlfriend" the lecturer of the Department of Business Analytics at HSE Armen Beklaryan used an analogy with the Drake equation to estimate the probability of meeting a "soulmate". Analyzing demographic indicators, education level, and subjective attractiveness parameters, he estimated the possibility of finding an ideal partner in Moscow to be "only" 6 times more likely than meeting an alien in the Milky Way. Such calculations illustrate the limitations of mathematical models in matters of human relationships. It is clear that the task is difficult, but is it possible to somehow increase this probability?

My colleague, a full stack developer, and I applied to the SberUp corporate accelerator with this idea. We were selected among 1500 other teams. We presented a project called VanLav, in which we planned to use machine learning algorithms to find the "perfect partner". We decided to additionally collect compatibility data: we started forming questionnaires, conducting personality tests, and finding out the values and interests of the participants. The key message was to find a person with whom you "speak the same language", with whom you want to chat in person, go for coffee, or maybe even go straight to the registry office.

Within the accelerator, we went through several eliminations, each of which became a challenge for us. We were taught to analyze the market, correctly form a value proposition, calculate unit economics, and confidently present the product to investors and the bank's top management.

Initially, VanLav was aimed at the external market, but we had an idea: why not test our algorithms right inside Sber? We decided to test the MVP on bank employees before entering the open market. This is how the idea of SberDating was born - a corporate dating app that became a separate product, the first of its kind in Russia.

At first glance, it may seem that creating a dating app within a corporation is a strange idea. But if you think deeper, everything falls into place. Work is a place where we spend the lion's share of our time. There is already a basic cultural and value filter here, as employees are selected not only for their professional skills but also for their fit with the corporate culture. This means that the likelihood of meeting a like-minded person within the company may be higher than somewhere "in the city", where there are practically no filters.

Find and delete

We worked on the project after our main job. Therefore, we often stayed up late at night: I was responsible for the database and algorithms, and my colleague developed the backend and frontend (python + react). SberDating differed from mass dating services in that we applied a whole ensemble of ten AI models. Why so many? To increase the diversity of matches and the likelihood of successful matches. We selected people based on the similarity of interests, psychotype, and other factors.

We tried to abstract from formal parameters: position, geography, or bank department - all this faded into the background. The main thing was character, interests, values, and life approach. This shift in focus to the "inner world" of a person allowed us to find those combinations that might never have surfaced in the usual work environment.

An especially important element was personality typology. We relied on the "Big Five" model, known in HR practices around the world. As a result, an intelligent matching system appeared, which not only brought together similar people but also took into account the nuances of their personalities. For example, an introvert and an extrovert can be interesting to each other under certain conditions, and two people with a close value profile have an increased chance of developing conscious acquaintances.

Unlike mass dating applications, where the user endlessly swipes, SberDating was built around the idea of "conscious dating." We introduced a limit of 10 likes per day. Psychologists suggest that a person can adequately perceive and evaluate about a dozen potential options over a certain period. Thanks to this limitation, the user does not "fall" into an endless stream of profiles but thoughtfully studies each one, evaluates interests, reads the description, and looks at photos.

The goal is simple: "Find and delete." We want people to find "their" person and leave the app for real life, where they can go for coffee, discuss common hobbies, or work on a project together. Against this backdrop, SberDating has become not just a dating service, but a kind of social experiment in conscious dating within the company.

Successful launch: 6000 registrations in one day

Despite the fatigue, we gave it our all, met the deadlines, and managed to launch our product. The launch was timed to coincide with Valentine's Day - February 14. This marketing move turned out to be very symbolic and effective: about 6000 people registered in the app on the first day. We even had to urgently purchase an extended package of emails for the mailing service, as we expected a maximum of a thousand people.

Among the success stories were the most unexpected scenarios. One employee found his old kindergarten friend in SberDating, whom he hadn't seen for many years. Another installed the app, forgot about it, and after some time received a message from a colleague - this is how communication began, which turned into a real friendship. And our Alexey found a companion for a theater trip on the first day.

On demo day, I came out with a pitch in which I told Alexey's story. I showed his journey: from meaningless spending on Tinder to using our app, where he was able to meet a girl he talked to for a long time and with pleasure. This emotional story made everyone understand that the problem is real and we can really change something. Alexey's story "hooked" and demonstrated that our product is not an abstract idea, but a concrete solution to real employee pains. We won the accelerator and were offered an investment of 2 million rubles (which I ultimately did not take).

What's next?

It is important to note that initially our solution could only work on employees. It was the presence of internal data about personnel, digital footprint, profiles, and test results that allowed us to create such an accurate and interesting recommendation system. However, I really wanted to go beyond the corporation and try something in the outside world.

I launched VanLav – a telegram bot with anonymous voice dating, which solved the same problem: connecting people based on their interests, characters, voice messages. The feedback from users was positive, but I quickly faced the question of monetization. It seemed to me that this format should either be paid from the start or integrated into the functionality of an already existing large social network with a large user base. A small bot by itself did not provide enough opportunities for earning, and attracting users was difficult. In the end, I did not understand how to make a large-scale business out of it.

After these experiments and the completion of the AI transformation project at Sber, I decided to leave the company and become the technical director of a venture studio that creates startups based on artificial intelligence. This is a new stage of my career, which I plan to talk about in the next article.

The story of SberDating is an example of how data and machine learning can be not only a tool for optimizing business processes but also a means to help people be happier. We started with a team of data engineers who were immersed in routine, and after a few years, we created a product that really impacted people's lives. Among the stories I heard later were thanks from those who found a life partner (or companion) through our application. This cannot but inspire.