Recommendation System Development for Potential Factoring Service Customer Identification
Development of an algorithm for identifying potential clients of factoring services based on information from databases of Russian banks. Its implementation in the VBA programming language. Recommendations for improving the efficiency of sales staff.
| Рубрика | Программирование, компьютеры и кибернетика |
| Вид | дипломная работа |
| Язык | английский |
| Дата добавления | 01.12.2019 |
| Размер файла | 1,6 M |
Отправить свою хорошую работу в базу знаний просто. Используйте форму, расположенную ниже
Студенты, аспиранты, молодые ученые, использующие базу знаний в своей учебе и работе, будут вам очень благодарны.
Accounts receivable analysis
Accounts receivable are divided into two types:
Ш standard (or ordinary) Standard type of receivables includes the issuance of a loan (execution of a contract with a certain amount) for a specific period. Such loan is strictly planned and must be settled before a specific date. Once the term expires, the debt goes into the category of overdue debt. Violating the terms of the contract, the debtor company receives fines and penalties.
Ш overdue (or unjustified).
Let's consider the following three cases of the ratio of accounts receivable and payable:
1. Accounts receivable is more than accounts payable. This means that the company delivers the goods with deferred payment.
2. Accounts payable is greater than accounts receivable. This means that the company operates on a prepayment basis. In other words, the buyer pays first and then receives the goods.
3. Accounts payable is equal to accounts receivable - this means that the company works without delay of payment and payment for the goods is carried out on the day of delivery.
The second and third cases to us, as representatives of the interests of the company providing factoring services, are potentially not very interesting because, based on the definition of factoring, it can exist only in the presence of a contractual delay.
But do all companies need factoring, because the existence of a contractual delay does not mean that the company will experience an owner's equity deficit?
For a more accurate analysis of the company's need for factoring services, it is necessary to analyze the above criteria over time. That is why the accounting for the year in quarterly breakdown was in the premises.
To achieve the greatest efficiency, let's look at a bunch of indicators: the change in revenue (sales volume) and accounts receivable over time.
In business, this indicator is often called day sales outstanding (DSO). This is an analytical indicator that is used to assess financial sustainability in the short term. It indicates the average period of time that passes from the moment the goods are sold to the buyer until the receivables are paid.
Here we can also observe three cases:
1. The volume of sales is growing, but at the same time accounts receivable grows proportionally. This means that the company increases sales while not changing the conditions for the provision of contract deferral.
2. Sales volume is not growing, and accounts receivable is increasing, this indicates that customers are paying worse and worse, and soon the company may have a shortage of owner's equity.
3. The worst case for the company, but perhaps the most promising for the factoring offer, is the situation when the sales volume decreases and the accounts receivable increases. This means that there is a clear problem in the company with the management of accounts receivable or there is a trade dispute between the supplier company and the buyer company. This situation may lead to overdue receivables, which will be discussed in more detail in the next section.
It should also be noted that it is not superfluous to look at such an indicator as “Bad debt reserve”. The very existence of such an indicator already shows the existence of a problem, and if you look at the ratio of this indicator to receivables, you can already make more unambiguous conclusions about the financial condition of the company. The large amount of reserves for doubtful debts indicates that the company has a rather poor pool of buyers who violate and do not comply with the contractual terms. Factoring as a financial instrument successfully used for receivables management, and as a consequence, well disciplining buyers, can be extremely useful for companies in a similar situation.
Let's consider a company where a "Bad debt reserve" indicator was equal to 0, but at some point it began to grow. This may mean that one of the buyers with whom our client is working is experiencing financial difficulties and there is a high probability that the client will want to transfer this buyer to factoring service.
Another important indicator is the debt/EBITDA ratio. This is a rather popular ratio among analysts, cleared of the influence of non-monetary items (depreciation). In the normal financial position of the organization, the value of this ratio should not exceed 3. If the ratio exceeds 4-5, it indicates too much debt burden on the company and the likely problems with the repayment of their debts. For enterprises with such a high ratio it is problematic to attract additional loans.
Like other similar ratios, the debt to EBITDA ratio depends on industry characteristics, so it is more often compared with the values of other enterprises within the industry. In addition, it should be noted that such items of expenditure as the purchase of new equipment that affect the outflow of funds will not be taken into account in the calculation of this indicator, since the purchase itself does not change the financial result of the organization, and depreciation allocations are not involved in EBITDA. Another important point - when assessing the indicator, it is assumed that the accounts receivable of the organization is regularly paid by buyers. If the company increases the unpaid debt of buyers, it certainly worsens its solvency, but does not affect the ratio of accounts payable to EBITDA.
Summing up, the following list of key economic criteria that affect the customer's need for factoring can be made.
1) Ratio of capital investment/significant investment in equity capital to the balance sheet total of the company (ratio 1);
2) Ratio of accounts payable to accounts receivable (ratio 2);
3) DSO indicator, or rather its ratio to the year (ratio 3).
4) Ratio of Bad debt reserve to accounts receivable (ratio 4);
5) Debt/EBITDA indicator (ratio 5);
In the next section, these criteria will form the basis for building a recommendation system to identify potential customers of factoring services.
Of course, we could dive into the analysis of the economic activity of the enterprise in more detail, but with a deeper analysis, the specifics of the market, the industry and the customer segment begin to play a very important role. Moreover, the main task of this work is to develop a recommendation system for fast initial processing of the client's database and each new criterion that requires calculation will increase the time of data processing.
It should be noted at the outset that this work will not consider external and internal operational risks associated with the provision of factoring services, for example, the criteria responsible for fraud and illegal financial transactions. This study is conducted using the premise that all customers who have entered the database for processing by the sales department have already passed the internal security check of the bank by default.
Risks
The growing financial instability in Russia makes it necessary to solve the structural problems inherent in the banking industry, to identify effective tools to ensure financial stability and improve risk management in the banking sector. There is no doubt that the Central Bank is taking active steps in this area, including carrying out planned work on the implementation of the Basel II and Basel III agreements. However, in conditions of constant changes in the situation on the banking market and in the Russian economy as a whole, continuous monitoring and analysis of their condition, adaptation of regulatory and supervisory measures, introduction of regulatory innovations are required, so the topic of this study is extremely relevant.
When the bank decides to set a factoring limit of a particular type, it is necessary to remember the following:
1. The risk of fraud in factoring transactions is one of the highest in the field of credits to legal entities and individual entrepreneurs. To reduce it, it is important to monitor the financial condition of the client and the payment discipline of his customers, as well as the documentary base.
2. The client must have all the necessary certificates, licenses to trade, which is financed by the factor, otherwise the shipment transaction may be invalidated.
3. It is important to diversify the factoring portfolio in order to reduce the risk of a high share of overdue factoring debt.
4. It is necessary to conduct an expert assessment of the liquidity of the goods traded by the client (including the presence of encumbrances on this product (collateral, etc.), the absence of negative dynamics of market demand for it, the price policy of the client. After the start of financing, it is useful to regularly monitor the markets in which customers work to control the decline in demand for the goods sold.
5. The pledge of the owners of the client according to the factoring agreement increases the chances of factor in the recovery of debt in distressed debt.
6. Factoring without recourse increases the risks of the factor. The debtor in this case is the only debtor of the factor.
7. Subtle factoring increases the risk of fraudulent transactions on the part of the client. The debtor is not notified of the assignment of claims and continues to make payments under the delivery contract to the client. Financing in some cases is carried out without verification of shipping documents. The responsibility for the timely repayment of the debt lies with the client, who is obliged to monitor the receipt of funds from the debtor and independently transfer funds to the factor.
8. The use of electronic document management in the financing of invoices increases the risks of fraud on the part of the client. Verification at the time of financing in this case in some cases is not made by the factor. In the future, the goods may not be accepted by the debtor and therefore not paid.
9. It is necessary to check the delivary contracts and the history of the customer's payment relationship with the debtor for exceptions, subject to the terms of payment of deliveries, including mutual offset. For such debtors (delivery contracts), it is proposed to use a reduced financing ratio (proportional to the share of offsets and delays in payments in the total amount of payment under the contract for the analyzed period).
10. When a factoring limit is opened for a customer for a debtor, for which the limit is set for another factor, the risk of fraud increases. An unscrupulous client may provide one invoice for funding both factors. Alternatively, such customer may be asked to enter into a new delivery contract with the debtor for a new limit and begin financing after the mandatory submission of a notification signed by the debtor on the assignment of claims under the new contract.
11. Prior to financing, it is useful to inform the client that it is necessary for him to transfer payments erroneously transferred by debtors in payment of assigned claims to the client on the day of receipt of such claims to the factor account. If the requirement is not met at the financing stage, it is recommended to suspend the financing of deliveries to the debtor. After the start of financing of the client, it is necessary to continue monitoring the financial condition of the client and debtors of their payment discipline.
The list of risks faced by banks and companies providing factoring services can be continued, but most of the operational and credit risks are already described in the professional literature.
Since the purpose of this work is to develop a recommendation system, the impact of the key economic criteria obtained in the previous section on the financial condition of potential companies should be assessed first.
At this stage, we are faced with the following problem: despite the fact that there are only five criteria, it is impossible to clearly determine the optimal combination of criteria values for the company. This is due to a very strong influence that the company's industry and its business segment have. In other words, the economic cycle of a company operating in the food industry and supplying goods to a large federal commercial network will be radically different from the economic cycle of a company operating in the field of metallurgy. Accordingly, the terms of the contracts, in particular the deferral of payments, and the concept of "healthy" receivables for these two companies will be different.
This problem can be solved by using the already developed database of factoring company. For each segment and each industry, you can create a portrait of the average economic indicators discussed in the previous section. This approach will be a conservative risk, since it is likely to eliminate the appearance in the recommendation of the company, the economic performance of which is very much deviate from the average for the industry and segment.
Practical part
Data clustering
In any economic community, with the exception of very small ones, the market should be divided into sectors that can be controlled -- in this case, you can correctly place the emphasis in the formation of the marketing strategy and adequately assess the risks of concentration by sector.
Based on the current banking practice in the field of factoring, distinguish 7 main business segments can be distinguished:
1) Micro
2) Small
3) Average
4) Large
5) Largest
6) Regional public sector
7) Military-industrial companies
At the same time, within each segment, the company can be attributed to one of 20 industries. For example, oil and gas industry, food industry, telecommunications, etc.
This means that within the framework of the study 140 risk profiles of clients will be compiled, consisting of a specific industry and segment. In other words, the analyzed factoring portfolio will be divided into 140 clusters.
Chart 9
At the next stage, the key economic indicators ratio 1...ratio 5 will be calculated for each company based on the latest relevant accounting. After that, within each cluster, averaging of these indicators by the number of companies included in the cluster with a weight in the amount of the established risk limit will be performed.
Calculation example:
If,
C (p, q) - cluster (segment, industry);
n - number of companies in the cluster;
- risk limit for n-th company;
i - serial number of the basic economic criterion;
- value of the i-th basic economic criterion for n-th company;
value of the i-th economic criterion for c-th cluster;
Then,
As a result, a vector consisting of 5 elements will be obtained for each cluster, each of which corresponds to the average value of the criterion within this cluster. A set of these five criteria will be the risk profile of the average client in this cluster.
Chart 10
To assess the probability of reliability of a newly arrived client, it is enough to calculate the mean absolute deviation (MAE) of its basic economic indicators from the basic economic indicators of the portrait for the cluster to which the new client will be assigned.
MAE can have values from 0 to +?, where 0 is a company that coincided with the portrait of the "average" company in the cluster in terms of its basic economic indicators, and +? is a company that is infinitely far from the average cluster company in terms of its indicators.
Cold-start problem
One of the issues raised at the recent ACM RecSys conferences (the main international forums on recommendation systems) was the so-called problem of constant cold -start of recommendation systems.
The classical cold-start Problem (CSP) is widely known. In a recommendation system that offers "users" certain "items" based on the existing history of preferences, it occurs when new elements appear in this well-established scheme. This is either a new user whose preference history is empty or a new item that has not yet been selected by any user. Many people know that collaborative filtering algorithms (as well as many other methods) in their original form are not applicable to such situations. However, in many real-world situations, CSP often becomes a cyclical problem for already known users or items -- for example, in cases where some users rarely appear in the system, change their interests (the so-called user volatility). This problem is called the Continuous Cold-start Problem (CoCoS).
The solution to the CSP problem is determined by the type of recommended objects -- for example, there are widely known practices of using basic sets of recommendations and using hybrid systems that combine a combination of collaborative filtering and content recommendations (the appearance of which, in general, is obvious in cases where descriptors can be assigned to item). When accessing third-party data, it is common to extract supporting information, for example, from social networks. However, all of these methods are not suitable in the case of the CoCoS problem, since it is assumed that after the user has become "known", he remains so for an unlimited amount of time.
As in case of CSP, the CoCoS problem can take the form of one of the following types -- CoCoS by user (Continuous Cold-Start) and CoCoS by item (Item Continuous Cold-Start). The first case that we have already touched on occurs when new or infrequent users are found in the system. The second -- if there are new items or those whose properties may change over time. In the case of native recommendations, for example, these are articles whose content can be changed by the authors (contrary to expectations, this is quite a common situation).
Unfortunately, this problem does not need to be searched for a long time, because we can face it as soon as in our list of potential companies there are those that do not belong to any of the already known clusters. This means that for these companies there will be no average portrait for the industry and segment, and thus to calculate the value of MAE for such companies will be impossible.
A visual demonstration of this problem is shown in the chart *:
Chart 11
Cluster C area (medium, coal industry) is highlighted in red. It was not possible to calculate the indicators for this area, because the system did not have data on companies from the coal industry of the middle business segment.
There is a natural question: how to assess the potential of the company for which there is no statistics on the industry and segment?
In this case, one of the ways to solve the problem is to create a risk profile of the client separately by segment and separately by industry. When portraits of the industry and the segment will be obtained, the target portrait can be taken for the average value of the key economic indicators of these profiles.
The formal presentation of this calculation will be as follows:
If,
C (p, q) - cluster (segment, industry);
n - number of companies in the cluster;
i - serial number of the basic economic criterion;
- value of the i-th basic economic criterion for n-th company;
value of the i-th economic criterion for p-th segment;
- the average value of the i-th economic criterion for q-th industry;
Then,
As a result, a vector consisting of 5 elements will be obtained for the cluster, each of which corresponds to the average value of the criterion within this cluster. The resulting vector will be our risk profile of the average client within the new cluster.
Chart 12
It should be noted that this approach is less accurate than the one that was applied at the first stage of building a recommendation system and this is due to business logic. Clustering allowed to take into account the specifics of the business not only within the industry, but also within its segment, since the economic cycles of large companies differ significantly from the economic cycles of small ones.
In the previous section, as part of the solution to the problem of risk assessment for new companies, it was proposed to compare the main economic indicators of the new company with the average indicators for the cluster.
The result of this comparison will be the mean absolute deviation:
Based on the results, the risk assessment system will be based on the following logic:
The lower the MAE value for the company in question, the better. This will mean that the new companies within each cluster will be ranked by increasing the average absolute deviation, where the first place will take the companies most similar to the average for the industry and segment, and the last place in the list of recommended will take on the contrary those companies that are most different from the average for the industry and segment.
This approach has a number of advantages. First, it is very easy to implement and can be written in almost any modern software environment. Secondly, it is a very conservative risk, because the company, which is closest to the average profile within the cluster, has a certain economic stability.
It's necessary to consider another important aspect. Any business is always aimed at maximizing profits and the banking sector is no exception. That is why we will have to deviate a little from the paradigm of minimizing risks and think about such an indicator as profitability.
First of all, it is worth noting the specificity of factoring services and how this specificity is related to the basic economic criteria described above.
The main part of the factoring company's income is the commission for factoring services and client financing. Drawing a parallel with classical lending, the main role in this case is played by the period of lending or in our case the period of factoring service. The principal difference between a loan and factoring is that in classical lending, the loan is paid once and for a certain period, and in factoring service, the client is financed on a permanent basis and the maturity of the client's debt to the factor is determined by the term of the contract deferral. This means that the longer the term of the contract delay, the longer the client will use the money, and accordingly the more factoring company will earn.
Let's consider the value of MAE obtained earlier.
= |
shows the deviation in the DSO value of the new company from the average in the cluster.
Let's consider the following situation:
Suppose, there are two companies belonging to the same cluster.
Suppose, the DSO for the first company is 54, for the second 56, and the average for the industry is 55.
Provided that all other indicators except DSO will be equal for the two companies we are considering, then the value of R for both companies will be the same, because |54 - 55| = |56 - 55| = 1.
Going back to the section of economic criteria, we recall that the ratio 3 - DSO indicator reflects the average term of the contract delay between the client and the buyer. Accordingly, if within one cluster there are two companies with all economic indicators being equal, except DSO, the preference should be given to the one that has more DSO, because the period of factoring financing for a company with a large DSO will be longer, which will lead to an increase in the commission for factoring services. The obtained dependence of the factoring company's commission on the DSO indicator shows that the previously proposed logic for building a recommendation is not optimal in terms of profitability.
Another important indicator for assessing profitability is the amount of the established limit on the company. The limit reflects the maximum amount of monetary claims assigned and financed by the factor within the framework of servicing a particular client. The factoring limit, even within the same cluster, may vary for different companies because it reflects the maximum amount of shipments during the contract deferral period. Accordingly, the greater the factoring limit, the greater the amount of monetary claims assigned to the factor and the greater the amount of financing, which means that the factor will earn more commissions when financing a company with a large factoring limit.
Each factoring company has its own methods of calculating the limit per client, which differ in detail depending on the company's risk policy, but they are all based on the analysis of similar indicators of financial and economic activity of the client and factors that positively or negatively affect the size of the limit.
A common methodology of calculation of the factoring limit is the analysis of the card 62 of the account, namely, the shipments and payments during the accounting period, usually it's a year. It should be noted that during the analysis of the card 62 of the account offsets and refunds are excluded, because the shipment data cannot be assigned and financed by the factor, and therefore should not be taken into account in the calculation of the factoring limit.
This means that the amount of the potential factoring limit should also be taken into account when forming the recommendation, and if two companies have the same basic economic indicators, then preference should be given to the one with greater the calculated value of the factoring limit.
Now we understand that the commission for factoring services depends largely on the amount of the factoring limit set for the company and the DSO indicator, which in turn reflects the potential funding period.
In order to consider the effect of the increased DSO and the factoring limit for the formation of the recommendations, let's consider the vector of mean absolute deviations obtained as the result of work of the recommendation system. For all companies that have the MAE value equal, we will calculate the value of the potential factoring limit and, depending on the value obtained, give priority to the one whose value will be greater.
Calculation example:
There are 5 companies.
Companies 3, 4, and 5 have the same MAE value.
Let's calculate the factoring limit for companies 3, 4, 5.
The company 5 has the largest factoring limit, and the company 3 has the smallest one.
Then our recommendation vector r is transformed as follows:
r = (1, 2, 3, 4, 5).
As can be seen from the example, companies 1, 2 have a lower value of MAE, thereby they will remain in the recommendation in their places. Companies 3, 4, 5 have the same value of MAE, and in order for the recommendation to be more accurate for companies 3, 4, 5 the factoring limit is calculated, after which their order in the recommendation is determined already on the basis of the obtained values of factoring limits. Preference is given to the company with a large factoring limit.
Building of the recommendation system
The task of the recommendation system is to inform the user about the product that he may be most interested in at the moment. The client receives information, and the service earns on the provision of quality services. Services are not necessarily direct sales of the goods offered. The service can also earn commissions or simply increase user loyalty, which then translates into advertising and other revenues.
Depending on the business model recommendations can be its basis, as, for example, TripAdvisor, or can be just a convenient additional service (as, for example, in some online clothing store), designed to improve Customer Experience and make navigation through the catalog more convenient.
In this work, the main task of the recommendation system is to help the employee of the sales department of the factoring company to identify the most promising and profitable customers for more efficient development of the database of potential customers.
At the previous stage of our work, we learned how to assess credit risk for our potential customers. However, risk assessment is not a sufficient condition for making a recommendation. As noted earlier, in the section of economic criteria, the most effective approach is to assess the economic performance of the company over time. In this case, the approach of assigning one of the three possible credit strategies is proposed:
1) growth
2) stagnation
3) decline
Each strategy describes the approach to be followed by a sales person.
If the company is assigned the strategy "growth" - this means that the basic economic indicators show that the financial position of the company is stable and the company is developing successfully. Such company is likely to be interested in factoring as a way to expand its commercial network and increase production. Typically, dynamically developing companies face a shortage of production capacity at some point. Factoring in this case can help to release part of the working capital for expansion or optimization of production. A striking example, well illustrating such a situation, can be seen in the field of food production. The company produced a certain product and supplied it to the regional trading network. The sales volume increased and this in turn led to an increase in the burden in terms of accounts receivable management. At some point in time, the regional buyer is absorbed by the federal trade network and offers the manufacturer to deliver the goods not only within one region, but also to neighboring regions. In this case, the production will face an obvious problem of expansion of production with all the ensuing costs and problems.
If the company is assigned a credit strategy "stagnation" - this means that the company demonstrates an acceptable debt burden, a stable financial position. As part of this strategy, a potential client may also be interested in factoring in terms of increasing production volumes, as well as the modernization of production cycles. Moreover, the stagnation of the company in the conditions of a dynamically developing economy and high competition can dramatically turn into a fall and financial problems. Factoring in this case can serve as insurance for small and medium-sized businesses.
The strategy of "decline" is a negative strategy and the companies to which this strategy is assigned are experiencing obvious economic difficulties or will experience them in the near future. The increase in the debt burden, the lack of working capital can lead the company to bankruptcy in the near future. Factoring service in this case may be in demand as a tool to minimize costs. It can also be successfully applied to improve the payment discipline of buyers, which will ultimately have a positive impact on the quality of the receivables of the problem customer.
For each of the economic criteria, it is proposed to assign an indicator reflecting the impact of the criterion on the credit strategy.
1 - criterion shows a positive trend in the results of the year
0 - criterion shows neutral trend in the results of the year
-1 - criterion shows negative trend within the year
Let's make a matrix of strategies of criteria value:
Table 13
|
Ratio 1 |
Ratio 2 |
Ratio 3 |
Ratio 4 |
Ratio 5 |
||
|
Growth |
1 |
1 |
1 |
1 |
1 |
|
|
Stagnation |
0 |
0 |
0 |
0 |
0 |
|
|
Decline |
-1 |
-1 |
-1 |
-1 |
-1 |
Since the indicator can be different according to different criteria, and the economic criteria have the same weight, it is proposed to assign a value equal to 0.2 to each of them.
Let's look at the process of forming a recommendation in more detail. In the previous sections, risk rating and credit strategy were identified for each company. The risk rating is simple - the smaller it is, the better - but the strategy is not so simple.
In general, within each of the proposed strategies factoring as a financial instrument may be useful, and for the strategy "decline" it is even necessary. The question of prioritization in this case depends on the risk of the company's appetite and the business indicators that the company must achieve. The answer to this question can only be given by the company's management and there can not be a universal solution.
Since the recommendation system is developed on the basis of the data of PJSC "Sberbank", the settings related to the identification of appetite risk and business indicators will be set in accordance with the policy of PJSC "Sberbank". The company adheres to the risk of a balanced approach in achieving its goals, and focuses on a product that would satisfy the needs of as many customers as possible. In essence, this means that the sales department employees should give preference to companies with a credit strategy of "stagnation". This conclusion is confirmed by the distribution of the companies on strategies chart.
Chart 14
The chart shows that 55% of all companies included in the pool for analysis belong to the strategy of "stagnation". 39% of companies show economic decline, which means that they are highly marginal for factors, but high marginality is accompanied by a high level of risk, which means that factoring companies experiencing problems with profitability and low margin should give priority to these companies. The remaining 6% of the companies is an ideal client for a factoring company, as it demonstrates the growth of economic indicators and will potentially be a stable partner. Unfortunately, such companies often refuse factoring services, which is due to the business logic and the cost of factoring services in the Russian market.
It is worth noting that the approach of determining the credit strategy is poorly formalized and, as noted earlier, is primarily determined by the credit policy of the company. In this work, we propose an approach for the analysis of vintage data factoring portfolio. For each existing company, a credit strategy should be identified at the time the company starts using factoring services. Having received the value of the industry strategy for each company, it is possible to determine the dominant credit strategy in the cluster, and accordingly to lay the following logic in the recommendation system: the priority within the recommendation is given to the company whose strategy within the cluster is dominant.
Let's consider the example:
There are three companies belonging to the same cluster. Their risk rating is the same, and credit strategies are different. "Growth" is the dominant credit strategy within the cluster. Then the company that will have also have the “growth” credit strategy will be given priority in the framework of the recommendation.
Summing up the results of this section, the following milestones can be highlighted:
1) For each company, a credit strategy will be determined based on the dynamics of the basic economic indicators of the company within the year.
2) Each company will be characterized by two indicators: risk rating and credit strategy.
3) The recommendation will be formed within clusters depending on the dominant credit strategy.
Calibration of the recommendation system in determining risk rating
The calibration process is a very important stage for recommendation system to function correctly. The main task to be solved at this stage is the task of taking into account the result of the recommendation system.
After the system works for the first time, a recommendation will be formed - all companies will be assigned a priority, and depending on this priority, the companies will be distributed to sales staff for further study. The result of the employee's work, from the point of view of the recommendation system, can be two scenarios:
1) the client was interested in factoring (a conventional unit in a binary system)
2) the client was not interested in factoring (a conventional 0 in binary system)
It should be noted that not every interested client will immediately use the factoring services of our company. Firstly, because it can already be served by another factor, and secondly, it can be a management decision of the company's owners, due to many factors. For example, termination or restructuring of a business. Perhaps the client will want to use the letter of credit instead of factoring. In any case, for us, as for the developers of the recommendation system, one thing is important - whether the recommended client was interested in factoring or not.
Based on the above in this section, customers who have expressed their interest in factoring or used factoring as a financial service will be included in the recommendation system and, accordingly, will affect the value of the risk profile of the average company within a certain cluster. This will be a kind of natural calibration process of our recommendation system in terms of determining the risk profile.
The process of calibration of the recommendation system in terms of determining the dominant strategy is similar to the process of determining and calibrating the risk profile. Clients are included in the recommendation system and accordingly change the ratio of strategies in the cluster.
It should be noted that refusals of recommended customers will also be recorded in the refusal database. Unfortunately, the statistics on refusals to date is not very representative, due to the fact that automated accounting for these cases has not been configured. However, as the statistical data of refusals are accumulated, the algorithm of the recommendation system can be developed and calibrated taking into account refusals. In the future, such an approach would help to improve the accuracy of the recommendations.
Practical implementation description
Stage 1.
Clustering and determination of the risk profile of the client.
The data of the factoring company's clients is used.
a) All clients of the factoring company are divided into clusters. Each cluster is defined by the customer segment and industry.
b) The following economic indicators are uploaded from the database for each company:
1) capital investments;
2) balance sheet;
3) accounts payable;
4) accounts receivable;
5) revenue;
6) provision for doubtful debts;
7) total liabilities (include both long-term and short-term debt, information about which can be obtained from the liability of the balance sheet of the organization)
8) profit (loss) before tax
9) interest payable
10) depreciation fixed and intangible assets
c) Based on the uploaded data, five main economic criteria are calculated for each company
1) Ratio 1 - ratio of capital investments to the balance sheet;
2) Ratio 2 - ratio of accounts payable to accounts receivable;
3) Ratio 3 - DSO/365, DSO - ratio of accounts receivable to revenue for the year;
4) Ratio 4 - ratio of the provision for doubtful debts to overdue accounts receivable
5) Ratio 5 - ratio of total liabilities to EBITDA.
EBITDA = profit (loss) before tax + (Interest payable + depreciation of fixed and intangible assets)
d) For each cluster, a weighted average of S1,..., S5 is calculated, where weights are the risk limits set for customers.
Stage 2.
Processing of the database of potential customers and the formation of recommendations.
The bank's database is used.
a) For each company from the bank's database five basic economic indicators ratio 1, ..., ratio 5, are calculated, similar to step (c) in stage 1, for each accounting date
b) Each company belongs to a specific cluster obtained in stage 1. The cluster is defined by industry and company segment.
c) For each company, the average absolute deviation of its basic economic indicators from the weighted average for the cluster (risk rating) is calculated.
d) If it is not possible to define a cluster for a company, a new cluster corresponding to the industry and segment of the company is formed. Similar companies will belong to the new cluster.
e) For a new cluster there is a risk profile separately by industry and separately by segment, it is not weighted by the risk limit, and is calculated as the arithmetic mean of indicators for the segment and industry. The risk limit is not used at this stage because the companies of the "Largest" segment have much larger limits, and their use is led to a shift of indicators towards this segment.
f) For a company classified as a new cluster, the mean absolute deviation of its basic economic indicators from the average for the new cluster is also calculated and the risk rating is determined.
Stage 3.
Identification of the credit strategy.
a) For each of the five main economic indicators its dynamics is determined: positive, negative or neutral.
b) For each company, the credit strategy (growth/stagnation/decline) is determined depending on the dynamics of economic criteria at step (a).
c) Once all companies are assigned a risk rating and credit strategy is defined, a recommendation is formed in accordance with the logic described in the section "Building of a recommendation system".
Stage 4.
System calibration.
a) The company included in the recommendation is processed by a sales department employee.
b) According to the results of work, the employee sets the criterion of the company's interest in factoring. (Yes - interested, No - not interested)
c) Companies with a sign of interest remain in the cluster and their economic indicators are included in the calculation of the risk profile of the client.
d) Companies without a sign of interest are excluded from the cluster and stored in a separate database.
Flow chart
Economic effect and analysis of the results
In the previous section, we discussed in detail the process of building a recommendation system for identifying potential customers of factoring services and implemented the proposed algorithm in the excel software environment.
In order to give an unambiguous answer to the question whether our project of building a recommendation system is effective and useful for a factoring company, first of all it is necessary to evaluate its commercial efficiency and analyze the results.
To assess the commercial efficiency, we will resort to simulation modeling.
Suppose we have a standard staff of sales department consisting of 10 people. Since the processing the database of potential customers is not regulated, assume that each employee chooses a company for further work at random, and the employee does not have any additional knowledge about this company.
Each employee can process no more than 5 companies per day. Then no more than 50 companies will be processed per day. It should be emphasized once again that employees choose companies for further work at random.
Now we use the Monte Carlo method and generate 100,000 scenarios of which companies could be selected by employees. Let's calculate the mean absolute deviation for each of the selected companies and average the result for the entire sample. Let's call the obtained value MAE_MK.
At the next stage, we will take the top 50 companies that have been recommended by our system and calculate the average MAE for them. Let's call the resulting value MAE_RS.
The results are as follows:
MAE_MK = 35%
MAE_RS = 7%
This means that the deviation of the companies selected by employees from the cluster profile risk is 35%, while the average deviation of the companies selected by the system from the cluster profile risk is 7%, which is 5 times better.
Thus, we can draw the following conclusion: when processing N clients recommended by the system, factoring companies will take the risk in the amount of q, and when processing N clients chosen by employees of the company at random, the factoring company will take the risk in the amount of 5*q. This suggests that the recommendation system reduces the risk by 5 times.
Of course, it should be noted that, like any system, the proposed algorithm is not optimal, and in the presence of non-formalized external data, insider information may be less effective.
In the section of building of recommendation system the approach of assignment of this or that credit strategy depending on dynamics of its economic indicators was described. As a result of work of the company the portfolio of clients which quality can be estimated in several ways is formed. However, at this stage there is no sense to invent, all the metrics for assessing the success and effectiveness of the company have been invented long ago. In general, there are three main metrics:
1) profit growth
2) sales growth
3) reduction of the share of a non-performing loan (NPL)
At its core, these indicators are directly correlated with credit strategies defined earlier. Thus, if a company forms its portfolio only from clients with a credit strategy "decline" in the short term it is possible to guarantee profit growth, because the cost of factoring for such companies is much higher, but as it was noted earlier, high marginality is associated with high risks. At the same time, we can't concentrate on finding just the perfect customers, as it increases the risk of loss of market share that can make the factoring product uncompetitive.
To test the recommendation system, a group of sales department employees was randomly divided into two subgroups A and B of 5 people each. After that, each of the subgroups was instructed to form a portfolio consisting of 10 new customers, so that the diversification of their portfolios corresponded to the overall diversification of the factoring portfolio of the company: 6% of customers with the credit strategy "growth", 39% of customers with the credit strategy "decline", 55% of customers with the credit strategy "stagnation". Given that each subgroup had to form a portfolio of 10 new clients, their portfolios had to include at least one client with the credit strategy "growth", four clients with the credit strategy "decline" and 5 clients with the credit strategy "stagnation". The sub-group A worked using the traditional method of cold calls, and the group B used the recommendation system. It should be noted that the database of potential clients was common for both subgroups.
The test results are presented in table 15.
Table 15
|
Subgroup |
А |
В |
|
|
Time spent on portfolio formation |
11 working days |
8 working days |
|
|
Number of processed companies |
27 |
19 |
The results indicate that the recommendation system can improve staff efficiency by 20 per cent or more, which is a very significant increase when considering the combined usefulness over a year or several years.
It should be noted that in order to fully assess the effectiveness of the recommendation system, it is necessary to conduct testing in the formation of more significant factoring portfolios, which include hundreds of customers. Unfortunately, within the framework of this study more extensive and detailed testing could not be implemented due to time and resource constraints.
Summing up the results of the evaluation of the economic efficiency of the recommendation system, it is worth noting once again that the introduction of the system not only has a positive impact on the level of risk taken by the company, reducing it by almost 5 times, but it also increases the efficiency of employees enhancing their productivity by almost 20%. Moreover, the use of the system allows form and diversify the factoring portfolio more efficiently depending on the market and economy, and to quickly and flexibly determine the further course of development of the company.
Task completion proof
The main goal of this work was to create a recommendation system to identify potential customers of factoring services. To achieve this goal, the following main tasks were identified:
Ш Identification of key economic criteria for assessing potential customers of factoring services based on the information contained in the financial statements (Form 1 - balance sheet and Form 2 -financial results statement);
Ш Creation of a recommendation system that would significantly reduce the time for processing of the database of potential customers and reduce the risks taken by the company;
Ш Assessment of the effectiveness of the recommendation system created on real data of the factoring company;
All of the above tasks have been completed. The sections "Economic criteria" and "Accounts receivable analysis" describe in detail the approach taken as the basis for obtaining the key economic indicators of the company. As a result of this work, 5 criteria were identified that most fully reflect the potential and interest of the company in the use of factoring services.
The second chapter of this study, which includes such sections as "Building of a recommendation system", "The cold-start problem" and "Practical implementation", describes the step-by-step construction of a recommendation system based on the criteria identified in the first part of the study. In the practical part of the study, the credit strategy for each potential company is determined. The credit strategy characterizes not only the financial position, but also the potential of the company for further cooperation.
Also, the problem of cold-start is considered and an algorithm for solving this problem using information available in the industry and segment is proposed.
The results and evaluation of the efficiency of the system are discussed in detail in the section "Economic effect and analysis of the results". The simulation model is used as the main tool which is based on scenario analysis, Monte Carlo method, which led to the conclusion that in the absence of additional external information about the client, the use of the recommendation system will reduce the potential credit risks associated with the provision of factoring services to companies by 5 times.
The analysis of economic and resource efficiency was carried out with the help of A/B testing, when one group of employees worked without the use of a recommendation system, and the second used the system in their work. As a result, the productivity and cost-effectiveness of staff using the recommendation system was 20% higher. The task of forming a factoring portfolio was completed faster, while the number of companies that had to be processed decreased.
Summarizing the results, it should be noted that the positive business effect due to the introduction of the recommendation system in the work of the company has been achieved, and the work process is optimized, which indicates the achievement of the objectives set at the beginning of this study.
Conclusion
Summing up the results of the study, it should be noted once again that the technologies of storage and use of big data do not stand still. Every year at the disposal of the world's business giants will be more and more information related to the personal data of customers. This trend sets not only new directions in the development of information technology, but also contributes to the development of new approaches in the field of commerce.
Подобные документы
Consideration of a systematic approach to the identification of the organization's processes for improving management efficiency. Approaches to the identification of business processes. Architecture of an Integrated Information Systems methodology.
реферат [195,5 K], добавлен 12.02.2016Lists used by Algorithm No 2. Some examples of the performance of Algorithm No 2. Invention of the program of reading, development of efficient algorithm of the program. Application of the programs to any English texts. The actual users of the algorithm.
курсовая работа [19,3 K], добавлен 13.01.2010IS management standards development. The national peculiarities of the IS management standards. The most integrated existent IS management solution. General description of the ISS model. Application of semi-Markov processes in ISS state description.
дипломная работа [2,2 M], добавлен 28.10.2011Review of development of cloud computing. Service models of cloud computing. Deployment models of cloud computing. Technology of virtualization. Algorithm of "Cloudy". Safety and labor protection. Justification of the cost-effectiveness of the project.
дипломная работа [2,3 M], добавлен 13.05.2015Overview history of company and structure of organization. Characterization of complex tasks and necessity of automation. Database specifications and system security. The calculation of economic efficiency of the project. Safety measures during work.
дипломная работа [1009,6 K], добавлен 09.03.2015Practical acquaintance with the capabilities and configuration of firewalls, their basic principles and types. Block specific IP-address. Files and Folders Integrity Protection firewalls. Development of information security of corporate policy system.
лабораторная работа [3,2 M], добавлен 09.04.2016The material and technological basis of the information society are all sorts of systems based on computers and computer networks, information technology, telecommunication. The task of Ukraine in area of information and communication technologies.
реферат [29,5 K], добавлен 10.05.2011Basic assumptions and some facts. Algorithm for automatic recognition of verbal and nominal word groups. Lists of markers used by Algorithm No 1. Text sample processed by the algorithm. Examples of hand checking of the performance of the algorithm.
курсовая работа [22,8 K], добавлен 13.01.2010Архитектура операционной системы Android. Инструменты Android-разработчика. Установка Java Development Kit, Eclipse IDE, Android SDK. Настройка Android Development Tools. Разработка программы для работы с документами и для осуществления оперативной связи.
курсовая работа [2,0 M], добавлен 19.10.2014Основные алгоритмические структуры. Запись алгоритма в словесной форме, в виде блок-схемы. Система команд исполнителя. Язык высокого уровня. Создание программы и её отладка. Интегрированные среды разработки: Integrated Development Environment, IDE.
лекция [61,7 K], добавлен 09.10.2013
