Collection in the consumer finance industry is to seek a dynamic balance between costs and benefits.

Ignoring the cost premise and talking about collection, there is no comparison. Intelligent collection is a partial cost reduction and efficiency increase, and the entire collection methodology has not changed.

The optimization of collections must be combined with changes in the bad debt rate. Cost optimization while keeping bad debts unchanged is the real way to reduce costs and increase efficiency. Otherwise, it will be a waste of money.

For financial companies, the biggest risk is people.

There is a saying in the financial industry that "one-third of the loan is divided into seven parts for management."

Whether it is a bank or a current mutual financial institution, the importance of collection is reflected in two aspects: first, it can minimize the loss of bad debts; second, it can seize higher-risk businesses and seize higher-risk businesses through strong collection capabilities. Gain benefits.

The default risk of the customer group of mutual financial institutions is higher than that of the credit card customer group. Subtle changes in the collection effect may cause profits or losses in the order of millions.

With the rise and fall of cash loans, the collection industry seems to be ushering in a spring.

On the one hand, the number of customers is exploding rapidly, on the other hand, customers are becoming increasingly difficult to collect, and industry-wide bad debts are soaring.

Intelligent collection has emerged as a new method.

AI is hot in 2018. Is smart collection really the future of the collection industry? not necessarily.

What is smart collection?

According to public reports, intelligent collection mainly uses artificial intelligence technology to optimize the entire collection process.

Let’s first take the entire collection process as an example. The following is the definition of overdue indicators for common collections. Usually, an overdue amount of 9 (M3+) is defined as non-performing, which is what the industry often refers to as the non-performing rate, and an overdue amount of 18 is defined as bad debt.

Of course, the definition of bad debts and the accrual and write-off mechanisms are different for each company, so we will not mention them here for now.

Definition of Overdue Collection Indicators

In this process, reminders will generally start by phone a few days before each repayment date (m1, m2, m3, etc.) The user's repayment date is coming, please pay attention to timely repayment. This method is used by Alipay, Baitiao, and the credit card centers of various banks.

Starting from m1, once the payment is not repaid, it will become overdue. Generally, a few days after the due date, a reminder phone call will be made. This is not a reminder to repay, but a reminder to repay. Therefore, as shown in the figure above, the entire collection intensity will gradually increase until it is overdue and will be dealt with through very strong means such as judicial, outsourcing and door-to-door. Of course, the cost will also be very high.

Throughout the entire collection process, overdue customers must be continuously analyzed through data reports, targeted collection strategies must be formulated, and adjustments must be made based on the collection indicators for each month.

This includes different aspects of work such as phone calls, outsourcing, and quality inspection.

Simplified diagram of the collection process

What aspects has been changed by smart collection?

The main innovations mentioned in smart collection technology are the following links:

1 Smart phone calls (inbound and outbound calls)

2 Smart phone calls Division of cases (mainly involving collection strategies)

3 Intelligent reports

4 Intelligent quality inspection (quality inspection)

Intelligent phone calls: greatly reduce labor costs ? Not necessarily

There are two types of smart phone calls. One is typical recording and broadcasting, that is, the speech information of the voice robot is manually recorded and imported in advance, which is very similar to Taobao's robot customer service.

The other is real-time artificial intelligence voice reply, which can be understood as iPhone's Siri or Amazon's Alexa for the time being.

The core of intelligent customer service is to increase reminders to users, increase frequency and reduce costs.

Traditional collection agencies use manual phone calls. Many financial companies will outsource this business, and the outsourcing will allocate the financial company's collection orders according to the number of agents. One seat equals one person. The average daily per capita production capacity in the industry is around 200-300 passes.

As for the profit sharing of collection companies, commissions are basically given in different proportions according to the amount to be collected.

Here is a reminder, don’t be attracted by the samples on the official websites of many artificial intelligence companies. Generally speaking, the official website is a polished version. If you want to know how the robot customer service performs in the collection business, you must do a test.

There is a current view in the industry that artificial intelligence has significantly reduced labor costs.

From the perspective of the cost of each order, it does have a greater advantage than manual labor. However, if this cost reduction is combined with the later bad debt collection indicators, the results may not be so optimistic.

Currently, there are different billing standards for mainstream telephone calling platforms in the collection industry

For pure manual dialing, the industry basically follows the dialing rate per call, which is 2 yuan/ Pass. The current mainstream charging method for artificial intelligence to replace manual dialing is based on the length of use, which is basically about one yuan or 0.5 yuan per minute.

Since the average duration of each manual call is basically about one minute, we can compare them on an approximately uniform basis.

Currently, iFlytek is doing relatively well in the field of artificial intelligence, and its fees are relatively higher than other suppliers in the industry. Let’s take a middle range of 1.5 yuan. Other suppliers also follow the median value, which is 0.5 yuan.

Assume that we also need to make 200 phone calls every day. The different costs are as follows:

Note: The bad debt rate here refers to the later indicators caused by different collection methods. change. The source of artificial intelligence data is business test numbers conducted by several mainstream consumer finance companies in China.

Pure manual labor can at least reduce the bad debt rate by one point compared with artificial intelligence. Let me repeat, do business testing and don’t trust any description on the company’s official website. If it is assumed that the amount to be called is 30,000 yuan per day, it will be around 900,000 yuan a month.

If you choose artificial intelligence:

If the bad debt is reduced by one point, the loss will be: 900,000 * 1% = 9,000 yuan.

Cost reduction, saving: (400-100) yuan * 30 = 9,000 yuan.

Basically a tie.

If we lower the unit price of artificial intelligence further to 0.25 yuan. Then the cost can be saved by 350 yuan * 30 = 10,500 yuan or 9,000 yuan.

If the gap of bad debts is not one point, it becomes 2 points. 900,000*2%=18,000 yuan 10,500 yuan.

The above numerical reasoning basically uses the business average in the industry. There will be an upper and lower deviation, but the amplitude will not be too large.

From the above simple deduction, we can draw at least three conclusions:

1. In terms of unit price, artificial intelligence has significant advantages over pure manual labor and can greatly reduce the average unit price. Cost, no question.

2. If the later collection effects are combined and the bad debt rates caused by different methods are also taken into consideration, it remains to be seen whether artificial intelligence is more cost-effective.

3. From a single point of view, the losses caused by the increase in bad debt rate are far more terrible than the cost reduction.

Finance is a business with a strong lag in risk control, and all changes must take into account the changes in bad debts that may result.

If you ignore the bad debt rate and carry out various optimizations unilaterally, it is tantamount to discarding the good and chasing the last.

Intelligent case division (collection strategy): If the human variable cannot be controlled, the value of artificial intelligence is not great

This is the core work of the entire post-loan collection. Different customers are contacted in different collection methods and collection frequency at different times. Developing an optimal collection plan between cost and recovery rate based on different data feedback and industry experience is the value of the strategy.

Collection strategy basically involves the following factors: people (collectors), at what time (collection timing) and in what way (collection intensity/frequency) what different cases are handled (collection case characteristics).

Whether it is automated WeChat, text messages, phone reminders or manual collection, the formulation of strategies requires human design. So far, artificial intelligence cannot replace human functions.

Intelligent reporting: Automated reporting has long been popular in the industry

Reporting runs through the entire collection management. For now, there is not much difference between the automatic reports provided by artificial intelligence and the reports built into the collection system.

Basically, for companies with MIS systems, the collection systems can provide automatic report generation in different dimensions. Many collection systems sold by third-party companies like Huateng in the industry include built-in reports. This technology has been mature for many years.

At this point, artificial intelligence has no significant advantage. Even whether the reporting level of artificial intelligence can be equal to that of MIS reporting requires further observation.

Intelligent quality inspection: Artificial intelligence is better than manual labor

In the field of quality inspection, artificial intelligence currently has a leading advantage.

Whether you build your own collection team or outsource it to a third-party collection company, you need to regularly access recordings to control the quality of collection calls. This is the so-called quality inspection. For example, no insults, threats, clear business descriptions, etc.

The traditional quality inspection method basically uses a certain amount of samples randomly selected from the collection recording library, recruits a group of quality inspection specialists to listen to them one by one, records the problematic cases, and calculates the proportion. .

In the case of artificial intelligence, the sampled speech samples can be converted into text through speech-to-text method, and then automated keyword information retrieval can be performed. At this point, artificial intelligence can theoretically reduce costs significantly.

But there is a problem here, but in reality, the problem of collection quality is that the collector has a problem. When the business scales up and the team that relies on manual quality inspection in the early stage is basically stable, a normal collector will not suddenly experience quality deviations.

This quality inspection is not completely random sampling. It can be intervened through human management, and there is also the theorem of large numbers. Therefore, as the scale increases and becomes stable, the effect may not be that great.

A truly effective system in the field of collection: automatic dialing system

Automatic dialing system: after getting the list of expected accounts, it sorts the accounts that need to be called by phone according to the set rules. Use algorithms to optimize the order of calls and maximize the productivity of each agent.

An automatic dialing system is a must for large collection agencies.

The following is a comparison of some business numbers before and after using this system:

Note: The numbers in the picture come from the book "Consumer Finance Scripture"

But The application of this system is highly dependent on the professional skills of post-lending strategists. The rules of the automatic dialing system are determined by managers to ensure that high-quality collectors answer as many calls as possible while maintaining a high call completion rate.

For example, how to ensure that an excellent collection agent will immediately send him the latest case the moment he hangs up the phone, seamlessly. In addition, the automatic dialing system is not a new product brought about by AI in the past two years. This is a very mature industry product and has been used by major credit card centers at home and abroad for many years.

However, this system also has obvious shortcomings. Basically, it is necessary and possible to use it on a scale of more than a thousand people. For a collection company that is too small, the cost of purchasing such a system is too high. Secondly, the automatic dialing system needs to rely on different variables in a huge database to allow the algorithm to make the most informed decision. If the sample size is small, it won't mean much.

What is the core of collection?

Collection is a labor-intensive industry. The core of collection ability is whether it can collect debts as low-cost and efficiently as possible.

The core of collection lies in the formulation and actual implementation of collection strategies. Under certain cost control, how to minimize the potential overdue risk or the harm of actual overdue behavior. This greatly considers the business capabilities of a professional modeling and strategist.

As shown in the figure above, in the entire post-loan collection process, collection intensity and collection effect actually have completely different strategies.

For consumer finance companies (except for cash loans, the business model is completely different), when overdue has just begun, there is no need to adopt the strongest collection intensity, but the collection effect at this time should theoretically be The best.

As time goes by, the difficulty of collection gradually increases, and the possibility of bad debts further increases. At this time, collection power should be increased.

There is another important income from collection, which is fines. Even under certain strategies, financial institutions intentionally make certain customers overdue or overdue for a long time.

Financial institutions are not stupid, and they have their reasons for doing so. Because good collection must be the best balance of the three factors: bad debt losses, penalty income, and collection costs.

The above figure simply lists the comparison of three different collection strategies.

The optimal collection is to maximize the positive value and reduce the negative value. These three factors are interrelated, and it is not easy to influence the whole body.

This extends to two core links: identification of high- and low-risk customers and creation of collection scores. At least as far as the current status of mainstream debt collection agencies is concerned, artificial intelligence cannot replace people.

Traditional financial institutions basically assign high-risk customers to the most experienced debt collectors. Because for high-risk customers, institutions cannot wait aimlessly, or increase the intensity of collection after the overdue time is extended. Instead, they must adopt high-intensity collection strategies as early as possible to get customers to collect their money.

How to identify high-risk customer groups among customers requires the use of a collection scoring system. The collection scoring system can be based on past customer performance and statistical methods, plus industry collection experience (including changes in user repayment probabilities that may be caused by different collection strategies, etc.).

Artificial intelligence has indeed improved efficiency at some single points, but this improvement needs to be considered in the overall picture of collection costs. Otherwise it is meaningless.

In addition, the current artificial intelligence outbound calls are still far behind real people. When more and more customers recognize that this is a robot speaking, will it have the opposite effect?