GoodsForecast: how mathematical models and forecasting algorithms replace marketers

Andrey Lisitsa - co-founder and CEO of GoodsForecast. Since 2005 - developer, manager

projects and one of the departments in the company"Forexis". He graduated from Moscow State University in 2005, the department of system analysis. In 2009, he entered the postgraduate program at the Computing Center of the Russian Academy of Sciences, and in 2016 he received an MBA degree after completing the training program of the RANEPA and Kingston University in London. At GoodsForecast, he is responsible for finances, building processes and developing new business lines.

Sergey Kotik - co-founder and development directorGoodsForecast. He also worked at Forexys in the positions of a developer, a project manager and a department head. He graduated from the University of Moscow State University in 2004, the department of mathematical methods of forecasting. In the company GoodsForecast supports transactions and develops affiliate programs, as well as attracting cash investments in business.

Forecasting consumer demand andcontrol over the processes of ordering products allow manufacturers, distributors and retailers to work most efficiently. And most importantly - help them to avoid two negative factors: the reboot of the warehouse and the inadequacy of the product range.

In the beginning were the algorithms

GoodsForecast was established in 2013 on the basis ofForexis, founded by a group of mathematicians and cybernetics from the computer center of the Russian Academy of Sciences (RAS) and Moscow State University in 2000. The main mission of Forexis was the use for commercial purposes of scientific research by Academician Yury Zhuravlev - scientific works on the theory of local optimization algorithms and the algebraic system of algorithms.

Yuri Ivanovich Zhuravlev

"In the 50s before Yuri Ivanovich Zhuravlev,then young scientists had the task of analyzing information about gold deposits, - says one of the founders of GoodsForecast, Andrei Lisitsa. - The search for gold deposits was a very difficult and expensive business. Zhuravlev was the first to build a mathematical model that made it possible, with great accuracy, based on the available data, without conducting test drillings, to say whether it was necessary to conduct geological prospecting at the expected location or the probability of finding gold would be too small. Thus, his well-developed career has received a new twist. What he did in the Soviet era, his scientific school continues to do today, in fact, is a development of the machine learning field, which is so popular today. ”

Zhuravlev Yuri Ivanovich - Soviet and Russian mathematician. His fields of activity include applied mathematics and computer science. The main areas of research are discrete mathematics, the theory of recognition and prediction, as well as Boolean functions. The new directions created by Zhuravlev include the theory of local optimization algorithms and the algebraic theory of algorithms, which consists in an algebraic approach to the problem of synthesizing correct algorithms.

The basis of the algebraic approach of Zhuravlev isidea of ​​using parametric families of algorithms. He believes that the process of solving poorly formalized tasks, which are not possible to cope with, makes it possible to find solutions to specific problems of this class. Based on this approach, results were obtained on solving the so-called canonically difficult problems.

Based on the test algorithm used by scientistsa completely new direction in recognition was created, which is based on the use of discrete analysis. The unique model for calculating estimations created by Zhuravlev is considered as classical today.

By 2000, Zhuravlev already had enoughextensive scientific school. One of his leading students, today an academician of the Russian Academy of Sciences, Konstantin Vladimirovich Rudakov is a famous mathematician. Zhuravlev and Rudakov, together with one of their graduates, decided to create a commercial structure based on the computing center of the Academy of Sciences, which will be based on the achievements of the scientific school and use mathematics to benefit business and government agencies, that is, to analyze data and carry out custom projects in this area.

“And“ Forexis ”functions as a kind ofan incubator, - adds Sergey Kotik, co-founder of GoodsForecast. - When a company gropes for a task that can be scaled, when it realizes that its solution is needed not only by the client, but by the market as a whole, and this can be used to make a serial history, for a while this trend grows inside Forexis, and then allocated to a separate company. So it was with Antiplagiat, Antirutin company, with us - with GoodsForecast. In the near future, Forexis will have other subsidiaries. ”

Stocks under control

In the Russian market of analytical andconsulting services there was a high demand for the preparation of professional forecasts for medium and large businesses. To work effectively, companies needed a clear sales and purchasing plan. To compile it, specialists systematized data on the activities of the enterprise. This process took up to several months. But when the information was finally prepared, the information gathered was out of date. Help was needed with the assistance of outside specialists who own an innovative method of systematization, analysis and forecasting. So in Russia began active digitization and IT implementation in business.

Sergey Kotik. Photo: Eugene Feldman / "Hightech"

These are the opportunities offered inGoodsForecast: based on mathematical models, the first forecasting systems were created. However, Forexis was not only engaged in forecasting. Analytics and trading monitoring systems were successfully used as a working tool on the Moscow Stock Exchange. Experts of Forexis modeled the schedule of flights at Domodedovo Airport, and also made a forecast of the degree of demand for the carriage of goods by rail to Russian Railways. For students and researchers, Forexis has created the Antiplagiat service, which allows to determine the percentage of text uniqueness.

"We use different mathematical models, -explains Sergey Kotik. - Those of them that are used for forecasting are different from those used for solving optimization problems. We make a start from a specific case, from a set of customer data, which our analysts investigate and analyze. Our company is engaged in the following: software development, data analysis, business consulting. It is this set of our capabilities that allows us to implement current projects. After all, in order to correctly set a task and associate it with mathematics, it is necessary to understand clients' business process well. And in order to solve the problem, you must be able to build models and customize them. Well, in order for this solution to work on large amounts of data, meeting the requirements of reliability and fault tolerance in a multi-user structure, you need to be able to clothe the developed models in the form of industrial software. ”

In 2013, the company GoodsForecast announced itselfas an independent legal entity, becoming a subsidiary of Forexis, and in the same year became one of the residents of the Skolkovo IT cluster. Five years later, its annual turnover reached 100 million rubles. The company's staff reaches about 50 people. In addition to the administration, sales and marketing, there are a project office, an analytics (mathematics) department and four product lines that develop, refine and implement software.

"The process of direct sales to large customerspretty standard. Reputation and word of mouth work. Someone went to an acquaintance, someone addressed himself, someone went to “cold”, got acquainted with someone at conferences, - says Sergey Kotik. - The cost of our projects is different, from hundreds of thousands to several tens of millions of rubles. And there were a lot of interesting projects. For example, we have been cooperating with Baltika since 2008. Almost solve all the problems associated with forecasting in their company. There is a very interesting project in terms of complexity with TechnoNIKOL - this is the optimization of production lines. Innovative projects in production are all usually complex and very individual. An interesting project now with the company Knauf. It consists of two fairly large parts: sales planning and optimization of production planning, that is, the distribution of this plan across their numerous production sites in Russia and the CIS countries. If we call regional projects, then we work with the Chelyabinsk company “Unichel”. They have one of the largest networks - more than 600 stores. Now there we finish the project on planning of inventory management. There, too, there are very interesting moments connected with the specifics of the shoe market.

Demand tops

Obvious effectiveness of the implemented programsstill not the guarantor of one hundred percent adoption of an innovative product. One of the negative factors affecting the speed of advancement of forecasting and planning programs on the market is the incorrect introduction of data by customer companies. In order to proceed with the forecasting, GoodsForescast specialists have to help them with the initial data at the preliminary stage. Such a need has become the reason for supplementing the company's activities with the provision of consulting services in a chosen direction.

Andrey Lisitsa. Photo: Eugene Feldman / "Hightech"

“Before we enter into a contract,quite a long period of communication with the client. In order to offer him the concept of the project, to describe the time, cost and agree on these parameters, it takes from one month to six months, and sometimes more, "- says Sergey Kotik.

All work on the project is divided into three stages:

  • Development of technical specifications for the implementation project. Everything is prescribed in it: system functionality, scenario of its use, algorithmic apparatus, acceptance criteria.
  • The process of solution implementation and refinement, if theyare needed. This includes integration with data sources, customization of algorithms, user training, acceptance tests. According to the results of this stage, the system is launched into trial operation.
  • Trial operation. When the system is already used, but perhaps not at full capacity, not at all, for example, the company's products or not at all warehouses. Any errors that occur are corrected, algorithms are configured. At the end of this stage, the entire system is launched into commercial operation.

What can the company's solutions based on mathematical solutions:

  • Replenishment makes the process of managing reserve goodsautomatic, while there remains the function of adjusting the quantity of goods ordered, which is important for distributors, producing organizations and retail.
  • Planning includes a set of opportunities for strategic and tactical sales planning.
  • Distribution optimally distributes the plan to the production sites and estimates the duration of the fulfillment of the tasks set in relation to the number of client orders.
  • Scheduling solves the problem of optimal scheduling of production lines with the aim of maximum satisfaction of customer orders with minimum costs.
  • Promo predicts the effectiveness of promotions, conducting themanalysis. The system determines how the sales volume will change according to the results of the promo, using the ratios of its various parameters and the dynamics of demand. If the action is conducted for the first time and there is not enough own data for evaluation, then similar activities are taken in the same region by the same parameters, but with a different discount depth.

Challenges and successes

"Difficulties in managing projects, of course,there are different - political, technical and sometimes even economic. If we consider the technical aspects, here the key point is the quality and structure of the initial data of the customer, - explains Sergey Kotik. - Within the framework of integration, difficulties can always arise, and they, as a rule, are very individual. Sometimes there are very specific problems. For example, a customer wants to make a project, and his IT specialists are very busy, their tasks are scheduled a year in advance, and they say they will not participate in the project. This is especially true for large companies. For example, we are currently working on a project where employees implement it without the help of their IT specialists, the business customers themselves provided us with all the necessary data. That is, in fact, did a titanic work. Even in spite of the fact that the IT department of the global head company has given up on everything, the project is on, other departments are coping with the task ”.

Partly, the GoodsForecast company is focused on large customers, because it is they who are missing the data volumes that it makes sense to analyze in an algorithmic way.

"We work with big business, above all,because the optimization that we do still gives the customer a significant benefit in large volumes. Imagine a stall that sells for 100 thousand rubles a month. If for him to build a forecast of demand and create complex models of inventory management, he will start earning 45 thousand instead of 40. However, the project itself will cost several million rubles. It is simply unprofitable, - says Sergey Kotik. - Our product should be introduced only when the company has a significant turnover. It is unlikely that the project will be implemented cheaply because each company is very individual and each of its products has its own characteristics. And this already requires certain labor costs. You can not take a certain model, one for all, adjust it and get forecasts, recommendations on orders - or to optimize production.

Sergey Kotik. Photo: Eugene Feldman / "Hightech"

However, we are working on the idea of ​​creating someuniversal solution that would allow us to scale and help small businesses in their work. But for small companies today are relevant other tasks. They are more important automation, in particular, the introduction of accounting systems and high-quality data management. Now there are a lot of convenient accounting systems for small businesses that are associated with trade and production. But if we are talking about such a small business that conducts sales in a paper magazine, recording all data with a pen, then no optimization is needed now or in the foreseeable future. ”

The main difficulty, according to Andrei Lisitsy,became unavailability of Russian enterprises to work with specifics. Enterprises have high hopes for analyzing data volumes, but at the same time they do not care about the fact that they are engaged in their own business, which will help to effectively use the result of the analysis.

"It is important to understand that artificial intelligence is notwill solve all the problems by clicking the fingers, adds Andrei Lisitsa. - It is not enough to simply acquire a powerful north and software platform, load data into the system. Competent specialists are needed who will organize data collection and, using them, set up an algorithmic apparatus in the context of the task. The experts who can decipher the results and use them in commerce are important. Even an experienced manager cannot understand directly the logic of the work of complex models and the influence of indicators on the result. So, he will not be able to manage the system, benefiting the enterprise. ”

What the coming year is preparing

Today, the Russian market analysis and forecasting demonstrates strong growth. According to GoodsForecast, in 2019 it will grow by at least 30%.

“We expect the same growth of our own turnoverat the end of the year, - adds Sergey Kotik. - The greatest demand will be in use of inventory management systems, the forecast reserves of goods and sales planning. We will develop solutions related to forecasting in the field of promo. It enjoys steady interest in the market and is indeed very large - at least 60% of the goods are sold through promo. Such sales are very labile and difficult to predict work. Because, firstly, the “producer-retailer” chain is involved, and secondly, many factors influence it. ”

Andrei Lisitsa, Sergey Kotik and Daniil Kanevsky (Analytics Director). Photo: Eugene Feldman / "Hightech"

In 2019, GoodsForecast will present a number of newproducts, one of which is based on the analysis of data of fiscal data operators (CRF). By check from the outlets, you can analyze the consumer basket, identify which products are usually bought at the same time in one hand, and evaluate the performance of cashiers. On the basis of such data, it is possible to optimally build the display of goods at the point of sale, predict the workload of cash desks, make up a schedule of staff shifts.

“We are developing already existing products: In particular, we add functionality that will allow us to create new ones on the basis of existing products, ”says Andrey Lisitsa. “Plus, we are now solving more problems related to optimizing production: how to plan production correctly, in order to meet the forecasted or current demand as much as possible and at the same time level the costs.”