Data analysis implementation process

Teacher Chen uses the case of a large after-sales chain service provider to reveal the entire process of data analysis implementation. The article not only provides a systematic analysis framework, but also emphasizes the importance of management methods in the data implementation process. If you are a data analyst, business decision maker, or a reader interested in data analysis, this article can provide you with valuable insights.

"The data analysis you did is spot on!"

"In addition to writing numbers, can you give us some practical suggestions?"

"I've seen the numbers, so what? What are we going to do?"

This kind of complaint is often heard in the office, which makes the data students very depressed. What is the implementation? Today, we will explain it systematically through an example.

Problem scenario: A large after-sales service chain provider accepts service demands from manufacturers, enterprises, and individuals. After receiving the demands, the customer service generates a work order, which is then assigned to the self-operated service point or outsourced service provider to complete the service on site.

Now it has been determined that the North Star Indicator is: the actual number of work orders completed. Question: How to conduct further implementation analysis.

1. Common mistakes in data landing

Many students said as soon as they saw the question: "Teacher, I know this question!" Work order = number of requirements * conversion rate. Since we want to increase the number of completed work orders, what we need to do is to combine the two indicators of number of requirements and conversion rate:

Make it high!

So the way data is landed is:

First, we need to sign more manufacturer customers

Second, we need to sign more corporate clients

Third, we need to increase personal traffic

Fourth, improve customer service efficiency

Fifth, we must strengthen on-site management

Sixth, improve the skills of the master

Look, this suggestion is so specific and practical...

Forehead……

First of all, it is true that these indicators should be set high, and these six points are indeed suggestions. The problem is: these are all correct nonsense. Even if you are not a data analyst, everyone "already knows it", so of course they should be set high. How can they be lowered?

The core of inferring business implementation actions from data is: priorities. Through data analysis, we can find out which are the key points and which are auxiliary. This is the value of data calculation. If we don’t calculate, everyone will know that this needs to be strengthened and that needs to be strengthened.

2. Step 1: Sort out the business process

The first step to implementation is to stop talking about numbers at the data level and just talk about numbers. You need to first figure out which business process the data comes from and which process it is affected by. The business layer is generally sorted out from coarse to fine, like peeling an onion. For example, in this case, although there are many business roles involved, they can be divided into: demand/supply based on the clues received by the customer service (as shown in the figure below).

The goal is to increase the number of completed work orders. The first level to be prioritized is whether demand/supply match.

Looking only at a single month/overall supply and demand situation, there may be three states:

Demand ≥ Supply

Demand = Supply

Demand ≤ Supply

The first level of judgment determines the subsequent landing direction:

Demand ≥ supply, improve supply capacity

Demand = supply, continue to observe/reduce supply costs

Demand ≤ Supply, develop customers and expand demand

This is the implementation suggestion of V1.0. Please note that the real suggestion will not be given in such a colloquial way, but is calculated:

This step seems simple, but it actually contains a mystery: how to determine which end is bigger?

Step 2: Establish judgment criteria

The judgment criteria are by no means as simple as letting the boss make a decision. Demand ≤ supply is relatively easy to observe, for example, the average number of work orders per after-sales technician is small, the average salary is low, and there is a lot of staff turnover.

But if demand ≥ supply, there is a high probability that there is no data record. For example, an individual customer calls in, but can’t get an appointment with a repairman; a corporate customer calls in, but has to wait in line for a long time before repairing, but because they signed an annual contract, they will not get angry for a while.

These situations make it difficult to assess supply capacity and the data is not true. When the company loses customers, it is too late to react.

Therefore, it is best to analyze the standards separately. For example, for manufacturers/corporate customers, it is necessary to match the service terms when signing the contract (for example, the order must be processed within 24 hours of receipt); for individual customers, it is necessary to look at the completion rate after the customer initiates the demand, and from the completion rate, eliminate the customer's reasons (such as being too expensive after asking for the price, not being able to find anyone at the door, asking casually, etc.) to calculate a relatively accurate number.

This is the work of finding the judgment standard. After establishing the standard, it is necessary to reach a consensus with all departments to achieve unanimous approval.

This step is very important. The reason why many students have difficulty in implementation is that they only have numbers but no judgment. Or the judgment conditions are not rigorous, which leads to many vague business definitions and arguments when they go deeper. This naturally makes it difficult to implement (as shown in the following figure)

Step 3: From short-term to long-term

Note: Over time, there may be seasonal changes. For example, a particular device is more likely to fail during the summer/winter when it is used more frequently. Therefore, after establishing a standard based on one month, you can look at the situation for the entire year and further identify the problem.

For example, the overall situation is: demand ≥ supply, but:

Occasionally (1 or 2 months)

Frequent (occurring for more than 3 months)

Sustainability (new/near end of life)

Seasonal (occurs in a specific season)

The corresponding priorities are also different, and the implementation suggestions that can be guided are also different (as shown below)

5. Step 4: Find the key points and focus on them

After the overall situation is determined, we can look at local problems. For example, in the first stage, we identify the problem as coming from the supply side, that is, insufficient supply. So how can we further analyze it?

First of all, the business has three lines, and we need to distinguish which one is the focus of the three lines. Because toB customers such as manufacturers/enterprises and toC individual users are two fundamental development ideas. Not only do they have different proportions in the total number of work orders, but they also have different importance to future development. It is very likely that toB is the lifeline of the company.

Under the influence of different importance, even if the current data is the same, the judgment on future development may be different. You must make a judgment first and then look down (as shown below)

6. Step 5: From the whole to the details

Secondly, after-sales service is delivered by region, so which region is particularly serious and which region is exceptional. This is relatively easy to understand, as the customer needs and store/master configurations in each region are different. It is very likely that remote areas are outsourced for the second time. Therefore, locking in the problem points also helps: focus on the big and let go of the small, and solve the areas with prominent problems first.

There are strategic differences here: if you really see that demand in a certain area is particularly strong and it is all outsourced, you will most likely choose to "replace the outsourcing and set up a service point yourself" rather than "the outsourcing needs to maintain the performance development trend."

In practice, it is never a matter of maintaining a good indicator or improving a poor one. There may very well be a third option.

7. Step 6: From local to details

Finally, after-sales service is carried out in two steps. It is necessary to distinguish whether the customer service is slow in dispatching orders or the after-sales execution is poor.

This breakdown is the most complicated. Because the customer service cannot dispatch the order, it is likely because the work of the regional/large customer service team is already saturated, or because of normal reasons such as holidays, or because of objective reasons such as waiting for parts to be transferred. Therefore, without detailed data such as the service team/supporting conditions/holiday conditions after receiving the order, it is difficult to say whether it is a problem with the dispatcher or a problem with the service.

When considering implementation plans, the more detailed the problem, the more likely it is that you will find that there is no data at all after getting to the very fine details. Use whatever data you have at hand, which is also the basic principle of analysis.

8. Step 7: From data to management

In the face of missing detailed data, management methods can be used to coordinate data construction. For example, customer service is required to complete the allocation within 30 minutes of receiving the order if the order is received for the first time and there are no abnormalities after sales. If there are any abnormalities, manual feedback and annotation will be given.

You can also check the inventory of parts in each region in advance and mark out-of-stock labels in advance, so that when analyzing, you can distinguish which parts are delayed due to waiting for parts. You can also require the service master to check in and out of the system before or after the service is completed, so as to count the master's on-the-job status and infer whether the master is saturated.

Note that these management methods themselves are also helpful for performance. They can detect supporting problems in advance, monitor front-line behavior to reward more work and more pay, and timely identify areas with growth potential.

Therefore, using these business benefits as bait can promote the implementation of management measures and achieve the purpose of data collection, thus killing two birds with one stone. If there are no good management measures, it is very likely that data cannot be collected and naturally cannot be implemented. If there are no business benefits, even if the top management intervenes strongly and pushes the software down, the business will not cooperate and fill in the data randomly, and the data will still be a mess.

IX. Summary

If you want to implement the data, you have to do it step by step from coarse to fine, eliminate various anomalies, hit the key points, and finally rely on and implement it in combination with management methods.

Instead of simply:

1. If any indicator is low, raise it

2. If any indicator is high, keep it high

It is not something that can be solved by simply coming up with a "God of Power and Invincibility General Model". For example, some students saw: "Oh! Customer service is sending an order!" and immediately responded like a conditioned reflex: "Let's follow Didi/Meituan and build an artificial intelligence order dispatching model."

Well, this is after-sales service. The damage rate of the machine will not continue every day like taxi/takeout, and the demand is limited. And door-to-door service also involves the issue of accessories, so it is not possible to mess around. So if you want to do it in detail, you have to go deep into the business process and patiently peel the onion.