Business prediction model, how to build?

The business process is suitable for impact on the service manager. When we change the results; many times, when we do predict models in work, it will often encounter a lot of problems. For example, there is no way to respond business, causing follow-up to adjust; this article shared We have learned about the construction of business forecasting models.

Do the predictive model is difficult, do an explanatory forecast model, it is more difficult!

When many classmates do predict, they will use the smooth and self-return of the time series, and they can be asked by the leaders:

Predict the standard, so? Do you have a standard? What is predicted? Who is the business? Where is the resources?

The students often asked a mist, I don’t know what to answer.

This problem is that the prediction method is not good. Algorithm model prediction Leading the business parties don’t know where to start when the business parties want to adjust the business behavior.

In this case, you need to build a business prediction model. Today, we explain the system.

First see a specific problem scene –

A TOB raw material supply enterprise, downstream requirements include:

Large customer, and signing a framework contract, cyclic procurement large customers, no framework contracts, monthly batch purchase small and medium-sized customers, no fixed contracts, there is a need for new development customers every month (in small and medium customers) each month Customers / passive home customers with active development

Now the business side needs to predict the customer’s purchase quantity next month, and hope to guide the specific work of the large customer sales / small and medium customer sales / new customer advertising. Q: How do you predict?

First, how do you do business prediction model?

Business prediction model, that is, the way business assumes as input variables, predicting business trends. This is different from the algorithm model, and the input characteristics of the algorithm model are often there without business, so it is not possible to guide the specific business operations, and business prediction is designed to compensate for this shortcomings.

For example, in this scenario, the biggest influencing factor in customer needs is the two dimensions of the customer’s own production plan / our company and customer relationship.

But these two dimensions are difficult to get accurate data. If it is a small and medium customer, it is very likely that there is no production and procurement plan at all. They are the wind and rain, and there is a single child. If it is a large customer who has no signature contract, each period of purchasing has to go back to the bidding process, which is likely to be halfway by other suppliers. So if you want to predict directly from two aspects, it is difficult to get started.

The work to do at this time is divided into four parts:

Card business process, find monitorable data indicators combing business features, distinguish stable / unstable factors carding business assumption, output prediction results track prediction results, correct process issues

1. Step 1: Sir, business scene

In this case scenario, the business process is relatively simple, that is, the customer pays for one hand, and our company is delivered; but the amount of orders for different types of customers is different, and the delivery difficulties can be considered separately (as shown below):

2. Step 2: Sir of business characteristics.

This step is the key. By combing the characteristics of each line, you can find a stable factor / unstable factor in each time period, the stable portion is the basis for predicting, the unstable portion is a means of controlling the prediction results.

In this case scenario, in terms of procurement needs:

Big customer of the framework ¡Ý Unmotted framework ¡Ý Small and medium-sized customers to develop trends ¡Ý Developer industry ¡Ý new customers who have become introduced by old customers ¡Ý new customers who actively developed ¡Ý passive home new customer

Therefore, first put the customer on the corresponding label, then packet data by different tag types, you can calculate the value of the following key indicators, and observe whether the trend is stable through historical trends.

The renewal rate of old customers, the renewal amount of the old customers, the new customer clues, the number of new customers, clues, new customers, new customers

Note that some factors here cannot be quantified directly, and transformation needs to be transformed. For example, “The industry has trend”, at least two methods to confirm:

Data Method: Tags all enterprises, then check the industry statistics, and then see our company’s signing company development data. Artificial law: All sales, regularly return to new customers / visit old customers, visit the time of 5 minutes, and reclaim data.

How to choose a method? A: Since it is a business prediction, it is a method of applying an impact on operations, ie artificial law. Because of the collection of data in person, it is possible to collect customer information, but also collect two key information of business actions and business judgment capabilities.

Imagine: If the sales visit / visit is perfunctory, it is uncomfortable. Can you have an order? It is definitely no; therefore measures the business action, it is also an important part of itself.

If in this process, find some business units is:

Personnel removal rate high personnel implementation effective to visit less than a number of visits and customers judges 10 times

That problem is very obvious: can’t business, leading to business.

This is very important, the students remember, since the business behavior is predicted, it is necessary to consider the business behavior. Do not try to doped half of business considerations, mix half the data to calculate it, so that the water is mixed, it is difficult to evaluate Hello.

3. Step 3: Output prediction results

With a clear classification, you can output the forecast results. The method of output is simple: there is a stable parameter, direct set; no stable parameters, business yourself fill the expected parameters;

This calculates the results (as shown below):

Note that business yourself fill in the predictive parameters is not a mess, you need to have a basis. As shown, it is ineffective to obviously violate the development rules. And this behavior itself can also become the input of the model: the business side capacity is insufficient, and it does not assess its ability and the required resources.

Also gives the business forecasting results, it also gives the hypothesis that need to be guaranteed, such as:

Suppose 1: XX industry customer demand is not affected by the export exchange rate; assuming 2: New clue conversion rate is not less than 5%; assuming 3: Business execution validity is above 90%;

These assumptions can be directly prepared for the inspection indicators of the tracking phase, which can be prepared in advance, so that even if some small problems can be directly corrected, there is a big problem that can be perceived in advance, saving tracking and replicating workload.

4. Step 4: Track the forecast results.

When actually occurs, the tracking results can be tracked based on prediction assumptions.

When the business trend is not good, you can early warning questions in advance. When the problem actually occurs, you can check the assumption to find the problem point. For problems with the counterpot, you can directly enable the plan to solve the problem.

This will make a good guide business action (as shown below)

Note that the above 6 cases, only the customer is expected to be out of the problem, belongs to the forecast failure. Why do you have such an important information in a prices, but there is no early prediction, business units and data departments must reflect.

I really encountered the problem of black swan, it is likely to be a change in the customer’s internal person or the opponent’s black trick. At this time, it will predict the failure, but it is not related to the forecast itself; these factors can not be predicted, at this time, when you can only Think of ways to.

Second, business prediction model, advantage and insufficient

The biggest advantage of business prediction model is to completely end: “In the end, it is not predicted, leading to a good performance; or does not have good performance, leading to predicting” this chicken egg, egg growth chicken.

It clearly tells everyone: it is because the business is not doing well, so it is not allowed to predict! And you can tell you very well because of the following business reasons, leading to a good performance, thereby guiding business development.

New customer clues are not in place, old customers, visitors, no, old customers, do not apply for the preferential price focus industry development capacity

The biggest disadvantage of business prediction model is to predict the judgment of dependencies; therefore, forecast results are particularly affected by team morale.

Generally, when the team morale is high, the predicted value given is large, the error correction capability assessment is also large; when the team morale is low, the predicted value given is small, and the error correction ability is not existed. The judgment of too biased will affect the landing of the model, which is not effective.

Therefore, business predictions and algorithms are predicted that both should not be biased. The algorithm model can directly give a total of overall data based on the past trend, so it is used to assist judgments: the current business party is an overestimation / underestimation of the situation, so that the leadership is based on the use of the management means, beat the business department. Correct judgment.

The business predictive model is suitable for use in the operational party to actively display, and when the results are changed. But some scenes, the business side is passive acceptance, such as customer service, after-sales, production line, etc. Customer incoming, promotional activities, new products listing, advertising, etc. Total amount to assess human arrangement.