The Center for Data Innovation spoke with Hristo Hadjitchonev, chief executive officer and co-founder of A4Everyone (Analytics for Everyone), a company based in Bulgaria that offers analytical tools to various sectors. Hadjitchonev discussed how data science can improve business efficiency by predicting the demand and addressing the need for talent.
Eline Chivot: How did your background and experience lead you to create A4E, and what is the vision behind your organization?
Hristo Hadjitchonev: The founders of the company have more than 70 years of combined experience in delivering data-analytics projects, so we are not youngsters. I have a Master’s degree in analytical modeling of economic systems, in the mathematical science behind AI and machine learning. I met one of the other co-founders and now our chief scientist, Professor Alexander Efremov, when we both were students. While he was pursuing an academic career, I was always in between, I had a background as a software engineer, after that I had a university career in data science, and after that in management, always connecting all these dots and getting deeper into various fields—data science, machine learning, and AI.
The idea behind A4Everyone dates back to 1999.
At the time, which was during the dot-com boom, I was working extensively for American companies. I started to work with big data—the term had not been coined then—as well as on delivering datasets. I saw that all the things we were delivering to these big companies at that moment could be given to any enterprise, regardless of their size. The biggest problem at the time was the very high cost of the computation, and only big firms could afford dedicated servers or data centers. The second problem was the cost for any project related to that. In time, these triggers were resolved. Computation costs dropped significantly—almost by a thousand times today compared to then, due to the diversity of cloud-based options we now have.
Back in 2015 I was the operational director of one of the world’s biggest analytics companies. The firm was active in the field of delivering data science to large global banks, by developing solutions related to scoring, decision making, and fraud prevention. I was responsible for the development process of the entire portfolio. But, you know, I had something in my head, an idea. With a team of four, we decided to try to apply what we are doing for big enterprises to SMEs. That’s how A4Everyone was born: With a little bit of passion for what we were doing, and with a bit of a dream that this could and should be accessible to everyone.
We have evolved over our five years of existence. What we deliver is the automation of decisions and processes. So, replacing humans in particular decisions, and doing better, faster, and more precise decision making, thereby freeing up time for humans, so they get to spend it on the creative side of things rather than on the repetitive, day-to-day work.
Chivot: How do you use AI in your analytical software to build viable business cases for companies in demand prediction?
Hadjitchonev: Everything that we deliver is built on AI and machine learning technologies, and via a cloud platform, with our own solutions. They are fed on a daily basis with a feed of data from our clients; this data is then crunched and analyzed, and new models are generated afterwards. Based on these models, we build prediction and forecasting—as well as the decision-making, which applies some form of business rules and particular business needs. These are unique to each customer. To be able to get into their processes and automate decisions, we have a process with our clients: On a daily basis, we automate human decisions, identifying how many units are needed, where to supply them, or what to order from suppliers. It is not only about predicting the demand. Based on the prediction, we also help clients make a decision on what to do, but in a completely automated way.
To give you an example—a retailer has one function, dealing with buyers which have two important jobs. On a daily basis, they have to order in a way that the warehouse is kept full of the exact-needed amount of goods. One of our clients is a cake producer which has 42 patisserie locations in Bulgaria and Romania. They produce fresh cakes—and because they are not frozen, their shelf-life is just three days. If you get the demand estimates wrong, either you will be out of stock, or you will have a huge amount of waste. So, for these three days, you have to have a very tiny balance between the demand you are going to face and what you have to produce and deliver. This client has 42 locations, meaning that there are 42 individual models accommodating different factors, such as the weather, which itself includes 17 different parameters: The rain, the speed of wind, the temperature, the clouds, etc. The weather could have an impact of about 20 to 30 percent on your demand. So, for instance if tomorrow in Bulgaria, it is minus 15, and there is snow, you will sell 30 percent less cakes. You also have “special days” that matter for your delivery, for instance, Mother’s Day.
When you have all the computation power you need and can experiment with automated and machine-learning based systems, you can incorporate a lot of such factors and parameters, and you can explore which are correlated to sales, and to what extent. We were even able to experiment with weird things like moon phases, which are said to affect human behavior (though we did not find correlation, and that’s especially unlikely in the cake business).
Our internal know-how makes us able to automate a mathematician’s work to solve problems. We automate the data science work. The traditional way is to work with data scientists, who build a project and a static-model—that works for a while. But after some time, the model has to be tweaked or even rebuked, and if you have to build an individual model for hundreds or hundreds of spots and have to rebuild them often as well, it is too expensive.
Chivot: Can you give a few examples of the benefits companies can get from using data analytics?
Hadjitchonev: Our business can directly impact the profit and losses of the companies we work with. If our solution is not performing properly, say if we automate contract negotiations for a health insurance company, and the contract is not fit, that company is at risk of a huge loss. In other words, we are making companies more efficient, more profitable, and probably one of the most direct things that we deliver is a better control over the processes.
With our solutions, clients very often find hidden gems of knowledge in their data, insights they were not previously aware of. For example, one of our clients had launched a six-week advertisement campaign for a product, across various media. Because we were trying to predict their demand, we were able to evaluate the impact of promotions on their required stocks as part of the model. We figured out that their promotions were working for almost four weeks, but beyond, their marketing spending’s effects were reduced to zero. Using that insight (and the money they were able to save), they squeezed their advertisements down to four weeks instead of six and became more efficient with more efficient advertisement campaigns.
There are direct benefits to automation, which include reducing human cost by making things more efficient, less biased, and more stable. In retail, you need consistency every day of the year but that is the case for all sectors.
Further, we cater direct business solutions to a particular business need. We set very financial key performance indicators like waste, profitability, and revenue, and we directly impact these numbers. We can measure these indicators. A lot of other companies are trying to sell the data science itself, but to me this doesn’t make sense. We provide a business solution, to solve a need and if that solution fits well, no one on the business side cares if that is achieved with neural networks or regressions. We don’t sell reports, we don’t sell data crunches, or data visualization. We deliver based on our platform, so the entire support of the infrastructure is part of the service. The company doesn’t need to install anything. The only thing we need to do is integrate our service with their existing ERP (enterprise resource planning) system and can continue using their normal way to work.
In addition, human decisions are generally biased—and with the coronavirus crisis, that became even clearer. Clients told us they were pleased to be able to rely on our solution, because right now, they said, people are not able to decide what to do, whether to understock or overstock, because of a lack of visibility. For example, right now it is even more important for the financial sector to take the right, evidence-based decisions, especially in the credit industry. Consumer behavior is changing dramatically. Our scoring system constantly adopts the new data which it is being fed.
To give you another example, we work with a large fast-moving consumer goods store in Bulgaria, for which we completed the full automation of orders to suppliers. That company supports everything as a platform—they have over 16,000 different goods, more than 500 suppliers of various goods, every day orders are sent to their suppliers, and their warehouses must be ready to meet that demand. They have excellent logistics as they deliver within two hours of ordering. Within just one month, as the coronavirus crisis broke out, their business increased by almost 120 percent. Our system was able to adapt to that change automatically. At first, our system recommended a huge amount of items to be purchased. Our client said that this was not possible, and that the amount had to be reduced. But two days later, they called and said their warehouse was already empty. Our system turned out to have worked perfectly, and most importantly, to reduce their waste by three times. Brick-and-mortar stores are suffering at the moment, and online delivery is not prepared to compensate for all the shops that have shut down.
Chivot: Why are few companies are using data analytics? And how is A4Everyone helping by making data analytics more easily accessible to various businesses?
Hadjitchonev: I recently attended a presentation by the data science team director of a large Dutch insurance and financial firm. He explained that even a company with large revenues can have problems with its internal data science team. Firstly, internal data science teams are very expensive to recruit and maintain. He has a team of 125 data scientists, his budget is €40 million annually, but data scientists are, right now, mainly excited to work on different things. They start something, and after one or two years, they leave, and try something else. It’s a struggle to hire, rehire, train, and keep up with the demand for talent. Secondly, resorting to external companies delivering data science as a service may be less expensive, as these costs are distributed among different companies and you don’t have to bear them all with an internal data science team. Thirdly, external services can provide higher quality, because they meet the demand for many more than just one client. If you can use an existing data science service, you better use it rather than build it yourself, and instead have your data scientists focusing on internal things that are very unique for your business, which they can’t find anywhere else, or which distinguishes your company’s performance from that of others. The problem is that too many companies are internally applying data science, for now, because the servicing in AI is still only starting. We are really one of a few companies trying to apply this business model.
Let’s consider two interesting numbers. The World Bank estimates there are 125 million companies worldwide. Now, in comparison, the number of data scientists, globally (including mathematicians), is between 1,000,000 and 1,500,000. In the future, there will not be a lot more than that. But not all will be good mathematicians. And for now, the ratio is only 1:125. Part of our business model aims to try to solve that gap, for as many companies as possible.
Chivot: A4Everyone provides tools to various industries, such as insurance, manufacturing, wholesale, customer finance businesses, and retail. Which one has the upper hand in the way in which it has been leveraging AI?
Hadjitchonev: The most advanced, with the highest penetration of usage, is the banking and the financial sector. Further behind is manufacturing—by far. The penetration in retail and wholesale is not high, but it is the most promising because retailers engage in high competition. They have to adapt as fast as possible. Using technologies like AI is becoming a differentiator between successful and unsuccessful companies. There are other factors, naturally, but generally, good management is always looking for every option to make business as efficient as possible, and that includes adopting AI. This could be many other things, such as better processes, logistics, and client support. It is not only limited to using AI or not. But it does reflect a form of mentality, and some form of brave pioneering to experiment to try and be better.
In Bulgaria, businesses are quite open to use and adopt such technologies. Throughout our first three years, we needed a lot of investment to build our platform and be able to deliver. Focusing primarily on Bulgaria and the region was enough to break even, which is quite good for a startup operating in a very limited space. At the end of the last and fourth year, we successfully closed a next round of financing. Our current target for expansion is the DACH market—Austria, Germany, and Switzerland. We already have leads there, some projects that we started, and we are very optimistic with what we are seeing. To some extent, our automation has really helped businesses navigate through this crisis. Imagine if 20 percent of your people are in quarantine, while your daily operations must continue. We help with something that can be automated and help people focus on other things that can’t be automated. We are seeing a potential to expand further and target all of Europe and later, the world.
