Recommendation system is a program which gives recommendations based on the data about the user and the item. In this work, the user is a sales person, and the item is a company.

Such system includes the whole process -- from receiving information to presenting it to the user.

Every stage is important. The information you collect will determine which algorithms you can apply. Good algorithms give good, useful recommendations. Criteria for the evaluation of the results allow you to choose the most suitable algorithms. However, none of this will work unless you properly present and explain the advice to the user.

This work has a certain degree of uniqueness and there are several reasons for that. First, the Russian factoring market is very young, and not all Russian companies are aware of the very fact of existence of such financial service.

Chart 1 Chart 2

Let us consider, for example, the segment of small and medium-sized enterprises (SMEs) on chart 1. According to the website bankir.ru of 03/15/2017, more than 2 million small and medium-sized enterprises, including sole proprietors, are registered in Russia. However, only 6 thousand of them use factoring services, which is less than one percent.

Typically, the extent of a financial service is measured by such indicator as the market share of that financial product in GDP. The share of factoring in GDP has been growing steadily since 2009 and reached 2.3% in 2012. On this basis, the Russian market can be called developing, since in developed European countries this figure can reach 10%, and in Eastern Europe - 5-7%. Thus, there is a great potential for the development of factoring in our country. Small and medium-sized enterprises account for a quarter of the total volume [8].

The composition of the Russian factoring market has not changed significantly over the past year. The most active are still about 35-40 factoring companies and banks. Most market participants are banks (seven out of ten market leaders). A sharp increase in the number of market participants is unlikely in the near future.

Let us consider the main factors that made the high growth rate of factoring in our country possible. Statistics show that the market is developing quantitatively due to the growth of factoring services without recourse (with the protection of the client from credit risk, i.e. the risk of non-payment of the buyer) and the increase in the number of transactions with large buyers.

This is primarily due to the expansion of the market leader - VTB Factoring, which actively offers factoring services to major corporate clients of VTB Group. In addition, other major banks, such as Promsvyazbank, Alfa-Bank and Petrocommerz, are the leaders of the factoring market. These players are focused on working with large customers and are ready to offer them a product such as factoring without recourse.

It is worth noting a number of events that indicate the qualitative growth of the market. This is primarily the development of electronic factoring technology.

In addition, in 2012, we saw the first example of real cooperation of factors in customer service. Promsvyazbank and the National Factoring Company (NFC) implemented a joint transaction when the first factor provided financing to the client, and the second provided protection against the risk of non-payment of the buyer.

In addition, the same NFC actively offers banks and other factors joint products, such as syndicated factoring (when the bank attracts a client and can provide funding for the transaction, and NFC provides a full range of factoring services: evaluation and analysis of buyers, management of receivables, debt collection, surety for the buyer).

Another driver of market development is the improvement of market infrastructure. Last year, electronic factoring began to develop actively. Amendments to the Law on Electronic Digital Signature were adopted, which allowed launching electronic factoring with a legally significant electronic digital signature. This technology allows the use of fully paperless electronic document circulation.

The supplier and the buyer can sign a contract with the certification authority, which is the guarantor of the authenticity of the electronic digital signature. After that, the whole process of document exchange within the framework of factoring transaction takes place remotely and is confirmed by electronic digital signature.

The supplier certifies the invoice with information about the delivered goods, the buyer confirms the acceptance of the goods. The supplier transfers to the factor a signed registry of monetary claims for funding. The factor checks it and confirms the electronic digital signature.

Due to the use of electronic factoring, suppliers and buyers receive a number of advantages: a significant reduction of the document circulation period, cost savings and information protection. The client has the opportunity to receive funding within a few hours after the physical delivery of the goods.

In addition, it saves a significant amount of money on the delivery of documents to the factor, stamp producing, drawing up of paper documents and its storage (and the set of documents for each delivery may consist of several items). Another advantage of electronic factoring is the legal protection of signatures. The certifying center confirms its authenticity, which can further serve as proof of the rightness of the factor in court.

Electronic factoring as a technology can be successfully applied to any product: factoring with recourse, non-recourse factoring, reverse factoring, guarantee factoring.

As for the product line of Russian factors, the main product on the market is traditionally factoring with recourse (almost 67% of the market in 2012). The share of non-recourse factoring accounts for 32%, international factoring accounts for about 1% of operations.

The volume of international factoring operations is constantly increasing, which is facilitated by the intensification of Russia's foreign trade turnover. In particular, the creation of a Single Economic Space with Belarus and Kazakhstan will allow to actively develop factoring operations for export/import from/to these countries.

Currently, only eight players offer this type of factoring in Russia. For the successful development of export and import factoring it is necessary to cooperate closely with international factoring associations (IFG and FCI), which provide the infrastructure for cooperation of factors from different countries.

Despite certain difficulties in the development of the market, factoring has great potential, which allows to predict the continuation of growth rates at the level of 15-20% in 2018. A number of factors are lobbying for changes in public procurement legislation so that factoring and changes regarding deliveries to state structures could be applied on clear legal grounds.

If these gaps in the legislation are eliminated, it will give the market an additional 30 or even more percent growth. In the usual situation, new products, the penetration of factoring into new industries and regions will remain the main drivers of growths.

Moreover, historically, factoring transactions in the segment of state-guaranteed order were practically inaccessible due to the lack of regulated rules in the regulatory enactment, according to which the company-contractor of state-guaranteed order could assign receivables of the state body. According to the "Law on Public Procurement", the change of supplier is not allowed under state contracts, which actually means a ban on factoring in the field of public procurement. Only in 2018, the factoring community together with entrepreneurs initiated appropriate amendments to the legal acts, which were approved by a number of ministries [4].

Now, the use of factoring in procurement is also extremely rare. The reason for the unpopularity is, first of all, the lack of clear regulation of this procedure and contradictory law enforcement practice. Although there is no direct prohibition in the legislation on the contract system for the assignment of rights of claim under government contracts, in practice, the customer may have problems at the stage of transfer of funds to the account of the financial agent. To make such payment, it is necessary to make changes to the contract in terms of the name and bank details of the recipient of funds, but because of the rules of paragraph 5 of article 95 and paragraph 2 of article 34 of the “Law on the Contract System” the cases of introducing changes to the contract concluded at the end of the tender are strictly limited. The current edition of paragraph 7 of article 448 of the Civil Code of the Russian Federation is an additional obstacle. According to it, there is a ban on the transfer of obligations from the contractor to third parties under contracts concluded on the basis of tender. The new edition clarifies that this restriction does not apply to monetary obligations. This amendment appears to be the first step to simplify the use of factoring in procurement.

The opinions of public authorities on the use of factoring in public procurement are different. Thus, the Ministry of Finance of Russia is of the opinion that, in accordance with the budget legislation, factoring under the state or municipal contract is not allowed (Letter № 02-02-04/13740 of March 11, 2016) [4].

The Federal Antimonopoly Service of Russia and the Ministry of Economic Development of the Russian Federation adhere to the opposite opinion. In accordance with the clarifications of the FAS (Letter № AD/38430/13 of October 2, 2013), the law does not prohibit the assignment of a monetary claim in public procurement [4].

Judicial arbitration practice has recently tended to the fact that factoring on government contracts does not contradict the current legislation.

The following risks can be identified for the customer:

Ш non-receipt from the bodies of the Federal Treasury of authorization of payment on details of the factor;

Ш payment to the improper person if the monetary claim is assigned several times;

Ш non-return of paid funds upon detection of defects of goods, works, services. In accordance with the new edition of article 833 of the Civil Code of the Russian Federation, in case of a supplier's failure to fulfill its obligations under the contract the customer will be able to make a claim to return the money only to the supplier, but not to the financial agent;

All these events once again confirm the fact that the Russian factoring market is at the stage of formation.

Secondly, the construction of a high-quality recommendation system, especially for such specific product, is simply impossible without access to private banking data. In other words, such system can be built only by a person working within the system and having access rights to customer databases.

Unfortunately, despite the serious efforts of banks to update and develop their IT platforms, the main forces are still focused on the development of platforms for servicing individuals and classic credit products. As a result, such areas as leasing or factoring are left out of modern IT approaches for processing, storing and using data.

The building of a recommendation system for internal use by employees of a factoring company would also not be possible without modern IT architectures for Big Data processing. Technical limitations in computing and resource power were a stop factor that prevented implementation of such ideas and algorithms 5-10 years ago [1]. To confirm this conclusion, it is enough to look at the chart MIPS (million instructions per second):

Chart 3 The computing power of modern computers has more than doubled in the last 8 years [1].

2.Theory background and market research

Recommendation systems have been one of the most well-known, easily monetized and replicated applications of artificial intelligence systems for several decades.

Many have heard the story of the recommendation system in the supermarket, which, based on the shopping list, understood that a young girl was pregnant, and began to send the advertisements of diapers before her parents found out about the pregnancy...

Ten years ago, a famous competition organized by Netflix for movie recommendations led to new models and major breakthroughs in recommendation systems. The market of such systems has since increased more than tenfold, and since then research, new ideas and models have been increasing, and the market continues to grow. But first, let's look at what is a recommendation system.

In the supermarket the recommendation system draws conclusions concerning what else can be offered to you in a way that will encourage you to buy it. The recommendation system draws conclusions on the basis of your behavior in the store, how often you go there, what is your route through the store, which counters with goods you pass by, what are you interested in, what you look at, everything you touch and buy. The aim of the system is to sell you more; ideally, it works absolutely magically: you've just thought about something, and there is a discount on the necessary product right in front of you just in time. In general, the aim of the recommendation system is to help the business to sell more by means of timely recommendations to the client in the right place, at the right time and through the right communication channel.

However, there are two harmful stereotypes that still prevent the wider use of recommendation systems and other similar models and methods. Firstly, in the minds of both the general public and businessmen, recommendation systems are still associated more with "entertainment" applications. For business, "recommendation systems" is something about Netflix, IMDB, last. fm, Pandora... O'f course, you can also recommend products in supermarkets and online stores -- Amazon is also at the forefront, but that's all. Secondly, it seems to many that even if theoretically a recommendation system could be useful, in reality it is too difficult, abstruse and requires a complete restructuring of the entire process of data collection and processing, as well as changes in business processes, logistics and inventory management principles, and not many people are ready to go for it. Businessmen who have gone further, can not understand what the ROI is from investing in such stories (considering the transformation of the business model), and therefore also refrain from investing so far, reading in the media and social networks stories about how well it turned out for someone. We will try to dispel these stereotypes, because in fact recommendation systems can be useful for almost any business, and to start recommending, the data that is collected along the way is usually enough, and the ROI is sufficient to start the project right away [3].

What is a recommendation system, how does it usually work? The classic task of the collaborative filtering is to predict what products (in the broadest sense of the word) users will like based on what products they liked in the past. Mathematically, this task can be thought of as the task of adding a table: imagine a huge table where the rows are users, the columns are products, and in the cells there are estimates of how much this user liked this product: for example, the rating of a movie in the form of stars or the rating of a product bought in the online store. We don't know anything about the vast majority of the elements of this table -- usually one client can estimate only a very small part of all products -- but in some cells there are known estimates, which are the input data for this task. The task of the system we want to build is to predict how much the customer will like those products that he has not yet estimated, and then recommend the best products personally for him. You can recommend either just those products that are likely to please him, or taking into account the profits that they will bring to the business. This problem is usually solved according to the "like like like" principle: alike users will like the same products, and the likeness between users can be estimated through what products they liked in the past.

At this point, the head of the content provider or even the store usually sighs sadly and says: "But I do not have any ratings... Yes, I have the customer base, I know what products they bought/read/used, but how do I know that customers did not like?.."It does not matter: there are many methods for one-class collaborative filtering, in which it is enough to know only the fact of consumption, without any negative estimates. For example, Facebook knows what users “like”, but it hardly has the technical ability to record what the user could have “liked”, but did not do it. These methods are a little more complicated mathematically (many of them are based, for example, on non-negative matrix decomposition), but for business the essence does not change: by using the database consisting of customer actions, you can evaluate the likeness between customers or products and extrapolate this likeness to new recommendations.

The main difficulty in all collaborative filtering systems, with or without ratings, is the cold start: how to recommend products that almost no one has yet seen, and what to recommend to users who have just come to the system; usually you have to resort to simpler, non-personalized schemes of recommendations (the most popular products, etc.), and then, as statistics are collected, move to personalized schemes.

At this point, content providers and stores are already in business, but the head of the bank sighs no less sadly: "It's convenient for you, you have one user read a dozen articles on the website, and you have the statistics, you can recommend something. And all our customers are new -- how many deposits or mortgages one person needs in one bank? One long cold start; your recommendations don't work for us..." But even here the answer will be found! The answer is usually to use the additional information that the business has about its customers.

For example, what might a recommendation system look like for a bank? When a person becomes a bank customer, and often even when he is just thinking about it, the bank usually already knows a lot about the person. Passport data already provide information about the gender, age and marital status, as well as geo-referencing (where the client lives); questionnaires that are usually filled out by new clients, provide information about the level of income, the structure of consumption of the client and so on. This allows us to move from the consideration of individual customers, which are determined only by their wishes and their experience of interaction with products, to generalized models that highlight the general preferences of customers, similar to each other in something, for example, "25-30-year-old girls from the Tverskoy District of Moscow." In such models, collaborative filtering is usually supplemented by clustering methods, splitting a large set of objects into subsets of objects similar to each other. It should be noted that not only natural demographic segmentation, which many banks do manually (or even automatically), is possible here, but also the use of much more detailed and interesting information, for example, the structure of the client's expenses, which can be estimated by the flow of funds on his bank card.

All of the above, as a rule, does not require any serious movements and large additional costs from business: you just need to prepare for analysts special data sets that already exist in the CRM-system that you use. The main substantive part of the project will be training the artificial intelligence systems and testing models on these (already historical) data, analysis and improvement of the resulting recommendations, taking into account the wishes of the business. And only at the stage of final application, the integration of ready recommendations into business processes some meaningful interaction between the team of analysts, the business function and the IT Department of business may be required.

Thus, recommendation systems are a large class of models that can help almost every business, whatever its specifics and subject area.

Now we are very close to the main part of our study, which in turn will consist of two global tasks and several subtasks:

1. Theoretical: identification of key economic criteria in determining the company's need for factoring.

Ш Identification of the data source for the analysis of the economic potential of a company when using factoring

Ш Identification of key economic indicators

Ш Analysis of the obtained indicators

Ш Risk analysis of the factoring market and the algorithm for their minimization

2. Practical: building a recommendation system

Ш Clustering of the customer database

Ш Building a recommendation system

Ш Cold start problem and its solution

Ш Calibration of recommendation system

Ш Economic effect and analysis of the results

4.Theoretical part