Building a business prediction model is a great method

In data analysis and business decision-making, building an effective business prediction model is the key. This article will introduce in detail how to combine business with algorithms to build a reliable business prediction model, help business departments better understand the prediction results, and guide specific business operations.

It is difficult to make a prediction model;

After making the forecast, it is extremely difficult to let the business know what to do next.

Today, I will show you how to combine business with algorithms to make reliable predictions.

1. Why do we need to build a business prediction model?

The above problems are caused by poor selection of prediction methods. Algorithm model predictions, regardless of the simplicity or complexity of the algorithm, have a common problem: they cannot reflect the business process. As a result, when the business side wants to adjust business behavior based on predictions, they don’t know where to start.

In this case, it is necessary to build a business prediction model. Today, we will explain it systematically.

First, let’s look at a specific problem scenario:

A toB raw material supplier, downstream demanders include:

• Large customers, with framework contracts and circular purchases
• Large customers, no framework contract, monthly purchases
• Small and medium-sized customers, no fixed contracts, purchase when there is demand
• Develop new customers every month (mostly small and medium-sized customers)
• Actively developed customers/passively visited customers every month

Now the business side needs to predict the customer purchase volume for next month, and hopes to provide guidance on the specific work of departments such as major customer sales/small and medium customer sales/new customer advertising placement. Question: How to make the prediction?

2. Business prediction model, how to do it?

Business prediction model is a method that uses business assumptions as input variables to predict business trends. This is different from algorithmic models. The input features of algorithmic models often have no business meaning and therefore cannot guide specific business operations. Business prediction is designed to make up for this shortcoming.

For example, in this scenario, the factors that have the greatest impact on customer demand are the customer's own production plan and the relationship between our company and the customer.

However, it is difficult to get accurate data for these two dimensions. If it is a small or medium-sized customer, it is very likely that there is no production and procurement plan at all. They are just floating around and will do whatever orders they get. If it is a large customer who has not signed a framework contract, each purchase must go through the bidding process again, and it is very likely that other suppliers will intercept it halfway. Therefore, it is difficult to make predictions directly from these two aspects.

The work to be done at this time is divided into three parts:

First: Sort out business processes and find data indicators that can be monitored

Second: Sort out business characteristics and distinguish stable/unstable factors

Third: Sort out business assumptions and output forecast results

Fourth: Track forecast results and correct process problems

Step 1: Sort out business scenarios

In this case scenario, the business process is relatively simple and clear, that is, the customer pays the money and our company delivers the goods. However, the order amounts of different types of customers are different, and the delivery difficulty is different, so they can be considered separately (as shown in the figure below)

Step 2: Sort out business characteristics

This step is critical. By combing through the characteristics of each business line, we can find the stable/unstable factors in each time period. The stable part is the basis for prediction, and the unstable part is the means to control the prediction results.

In this case scenario, in terms of procurement requirements:

• Large customers who signed the framework ≥ large customers who did not sign the framework ≥ small and medium customers
• Industries with good development trends ≥ industries with average development ≥ industries with poor development
• Old customers ≥ new customers referred by old customers ≥ new customers actively developed ≥ new customers passively visited

Therefore, by first labeling the customers accordingly and then looking at the data in groups according to different label types, you can calculate the values ​​of the following key indicators and observe whether their trends are stable through historical trends.

• Renewal rate of old customers
• Renewal amount for old customers
• Number of new customer leads generated
• New customer lead conversion rate
• Amount of first order from new customers

Note that some factors cannot be directly quantified and need to be converted. For example, "the industry development trend is good", there are at least two ways to confirm:

1. Data method: label all enterprises by industry, check the statistical data of the industry, and then look at the development data of enterprises contracted by our company.

2. Manual method: For all sales, regular revisits to new customers/visits to old customers must last more than 5 minutes, and data must be collected.

How to choose a method? Answer: Since it is a business forecast, the method that can be influenced by the business is preferred, that is, the manual method. Because the manual method can not only collect customer information, but also collect two key information: business actions and business judgment ability.

Just imagine: if the salesperson is perfunctory and careless even in the return visit/visit, can there be any orders? Definitely not. Therefore, measuring business action is also an important part of business forecasting.

If in this process, it is found that some business departments are:

• High staff turnover rate
• Poor staff execution
• Few effective visits
• Visit and chat with customers
• Feedback is wrong nine times out of ten

Then the problem is obvious: poor business capabilities lead to poor business.

This is very important. Students must remember that since you are making predictions based on business behavior, you must consider the business behavior thoroughly. Don't try to mix half with business considerations and half with data for your own calculations. This will muddy the waters and make it difficult to evaluate whether it is good or bad.

Step 3: Output prediction results

With a clear classification, we can output the prediction results. The output method is very simple:

• If there are stable parameters, apply them directly
• If there are no stable parameters, the business can fill in the estimated parameters by themselves.

The results are summarized and calculated like this (as shown below):

Note that the business does not fill in the estimated parameters randomly, but needs to have a basis. As shown in the figure, the estimated results that obviously violate the law of development are invalid. And this behavior itself can also become the input of the model: the business side lacks the ability to evaluate its own capabilities and required resources.

In this way, while giving the business forecast results, we also give the assumptions that need to be guaranteed, such as:

• Assumption 1: Customer demand in the XX industry is not affected by the export exchange rate
• Assumption 2: The new lead conversion rate is no less than 5%
• Assumption 3: Business execution effectiveness is above 90%

These assumptions can be used directly as evaluation indicators in the tracking phase, and response plans can also be prepared in advance, so that even if some minor problems occur, they can be corrected directly, and major problems can be perceived in advance, saving the workload of tracking and reviewing.

Step 4: Tracking prediction results

As they actually occur, the results can be tracked based on the forecast assumptions.

• When business trends are not good, problems can be warned in advance.
• When the problem actually occurs, you can check the hypothesis to find the problem point.
• For problems that have response plans, you can directly activate the plans to solve the problems.

This can guide business actions very well (as shown below):

Note that in the above 6 situations, only the unexpected problems that the customer foresaw belong to forecast failure. Why was such important information as the price suppression by the major customer not foreseen in advance? Both the business department and the data department should reflect on this. If a black swan problem really occurs, it is likely that there is a change in the internal personnel of the customer or the opponent has played a dirty trick. At this time, the forecast will indeed fail, but it has nothing to do with the forecast itself. These factors cannot be predicted, and at this time we can only think of a solution when reviewing the situation.

3. Business Forecast Model, Advantages and Disadvantages

The biggest advantage of the business forecasting model is that it can completely put an end to the chicken-and-egg problem of "Is it the inaccurate forecast that leads to poor performance, or is it the poor performance that leads to inaccurate forecasts?" It clearly tells everyone: it is because the business is not done well that the forecast is inaccurate!

And we can tell you in detail that the poor performance is due to the following business reasons, thus guiding business development.

• Inadequate follow-up of new customer leads
• Inadequate visits to old customers
• Old customers did not apply for preferential prices
• Poor development capabilities in key industries

The biggest disadvantage of business forecasting models is that forecasts rely on human judgment. Therefore, forecast results are particularly affected by team morale.

Generally, when team morale is high, the predicted values ​​are too high, and the error correction capability is also too high; when team morale is low, the predicted values ​​are too low, and the error correction capability does not exist at all. Overly biased judgments will affect the implementation of the model and thus fail to achieve the desired effect.

Therefore, both business forecasting and algorithm forecasting should be equally important. Algorithmic models can directly provide overall data based on past development trends, so they are used to assist in judging whether the current business side has overestimated/underestimated the situation, so that the leadership has a basis for using management methods and can encourage the business department to make the right judgment.

Business prediction models are suitable for use when the business side can actively exert influence and change the results. However, in some scenarios, the business side is passive, such as customer service, after-sales, production lines, etc. Customer calls are affected by many factors, such as promotional activities, new product launches, advertising, etc., but none of these influences can be controlled by customer service. In this case, it is not appropriate to use business prediction models, but algorithm models to directly estimate the total volume of calls next month to evaluate human resource arrangements.