In this environment, how can we demonstrate the value of data analysis?

How to reflect the value of data analysis in a scenario? The author of this article points out that data analysis should be conducted based on business needs rather than simply applying fixed analysis methods, and that the value of data analysis can be enhanced by accumulating experience and serving the business.

Students often complain at work that "the analysis they do has no business value, and there is no response after the report is sent out." How can we make data analysis reflect its value? Today, I will explain it in detail based on a specific scenario.

1. Problem scenario

A classmate joined a company membership center and ambitiously wanted to establish a "data-driven business" mechanism. To this end, he planned a lot of:

• Establish a full-link data indicator system to reflect operational conditions
• Create 3,000 member tags to enrich member portraits
• Establish RFM model and stratify members
• Establish a prediction model to predict member purchase behavior
• Recommendation model to recommend items for members to purchase

After submitting the plan, the leader criticized him: "What business value does this have?!" The student was very confused. Whenever user analysis was mentioned, it was always about user portraits, RFM, stratification, and recommendations. Is there some mysterious power that I don't know? Why were these things criticized as "valueless"? ? ? ?

2. Let’s start with “What is value?”

This question starts with “what counts as the value of data analysis”. If we are a B-party company, a data product or consulting company, we can make a bunch of tools in advance, such as BI, data models, CDP, etc., and then sell them to the A-party. This is a direct reflection of the value of data analysis.

But let's change the scenario. We are now the client company, and the data directly serves the business. At this time, the value of data analysis is defined by "how much it helps the business". If:

• The business can already see the data we provide
• The forecasts we make are not needed by the business at all
• The classification we made is not understandable to the business at all
• The business can’t find a place to use the portraits we made

That means it has no value. This has nothing to do with "everyone does this kind of analysis". As long as our company's business does not use it, it will not be recognized.

Therefore, if you want to enhance the value of data analysis, you cannot "find nails with a hammer". You should first use a bunch of "user portraits, RFM, related recommendations, behavior predictions" hammers and then see where you can hammer. Instead, you should look at: "What does our business really need?"

Note! Don’t just ask what the business needs, because it is very likely that they need you to tell their fortune.

3. Dismantling business issues

If you ask the business directly, “What do you want?” you will most likely hear:

• I want to predict the sales volume of products next month with 100% accuracy
• I want to know the reason for user churn with 100% accuracy
• I want to predict the number of participants in an event with 100% accuracy
• I want to be 100% accurate in determining what users want

In the end, they may even say to you politely: "If I can't do it 100%, 98% is fine. Let's not worry about it."

If we really use this demand to do it, it will definitely not work. Why? Because product sales, user churn, and user demand are all comprehensive results after being influenced by business means. For example, the user did not want to buy, but the business sent a coupon, and the user wanted to buy. At this time, you must first know: whether the business will send a coupon, and how much coupon will be sent, in order to predict the result.

This means that even if you can model and predict, you must first know: "What is the business means?" This is obviously inconsistent with business requirements. These people are still waiting for us to make 100% accurate predictions and then take action based on the prediction results. So don't swallow business requirements directly, but disassemble business requirements and find the focus that can be made.

For example, “100% accurate knowledge of the reasons for user churn”, ask one more question: What can you do after knowing it?

• Some users contribute very little. If they leave, is it really worth winning them back?
• Some users no longer have demand for our products. Will we really improve our products?
• Some user demands conflict with our company’s existing users. Do we really want to please these people?
• Some users are here just to get the free stuff, do we really have to satisfy them without restraint?
• …

Combined with the business scenario, we can break down the requirements. We will find that in most cases, the business can only do very limited things. In particular, to increase certain indicators in the short term, the only possible means is to send coupons or send messages, and that’s it. Long-term improvements are closely related to the overall business planning. Therefore, focusing on what the business can do can ensure that the data can be used by the business.

A common requirement here is to break down the needs, not according to what the business can do. Instead, they simply go around in circles at the data level. For example, to analyze user churn, they pull out the list of all churned users, and then throw in a lot of descriptive statistical results such as gender, age, past consumption, etc. This may seem like a lot of data, but it is still out of touch with business actions. After reading it, the business is still confused and will still ask, "So what? So what can I do?"

4. Find business blind spots

After breaking down the problem, you can further look for business blind spots. Finding business blind spots is very useful for improving the value of data analysis. Because the business is not completely ignorant of data, especially data related to their KPIs, which are generally closely monitored. If the data you throw out is something they already know, then you will definitely get a complaint like "I already knew it, what's the point of your analysis?"

so:

• Business doesn't know what
• What the business is uncertain about
• What does the business want to test?

This is where the value of data analysis can be best reflected

For example, in the case of user churn, it is very likely that the business side is regularly issuing coupons every month, and has already mastered a lot of basic data, such as:

• How many users are churned each month
• How many coupons are issued each month to recall users?
• How many channels are used to send push notifications each month?
• Conversion rate and recall rate of each channel

At this point, if we talk about these data again, even if we use the banner of "establishing a complete indicator system", people will still criticize us for "I knew that a long time ago". So based on the existing data, we can cut down to the following: what are the additional things that people want to know.

Here, whether to modify the existing rules is a key issue. If not, then make permutations and combinations based on the existing push channels, push copy, and push coupon types. Find the method with the highest conversion rate. If you want to modify the rules, then see what can be done in the business? For example, increase the denomination, change the time point of push information, change the push method (for example, based on user fission), and change the push content (for example, push a hot item instead of a coupon).

Whether to modify is closely related to the business. Because in many cases, the business does not have the authority to change the strategy. The strategy is already set by the boss and can only be modified. Sometimes, the business wants to modify the strategy. At this time, a large amount of data support is needed. It is necessary to first understand: which direction the business wants to change.

For example, the business side wants to increase the coupon strength. At this time, the idea of ​​data analysis is:

• Find out which groups are most likely to be recalled by coupons
• Identify the valuable segments of these groups
• Calculate past consumption of people and make recommendations on coupon strength

In principle, talents with consumption power are worth recalling vigorously. These segmented data are what business bosses need to emphasize when they ask for resources in order to dispel their concerns. Providing data at this level can greatly support business actions and reflect the value of data.

Note that the analysis here is still:

• Distinguish users’ past consumption and calculate user value
• Differentiate users’ sensitivity to coupon collection/use, and label them as “coupon preference”
• Differentiate users’ response rates to push information and find high-response groups

But after being clear: "This is data that the business particularly needs", this data will be recognized by everyone when it is given out. Instead of doing an RFM first (the segmentation rules have not been confirmed with the business) and then throwing it out, the business will be confused: So what? So what can I do? ? ?

5. Accumulate experience

Another very important thing that data analysis can do proactively is to accumulate experience around the goal. Business departments are often concerned with their own projects and tend to ignore the cross-effects between different projects. For example, in the case of user churn, it is very likely that the churned users are fans of certain products and have seasonal shopping needs. They will not respond to the coupons blindly issued by the membership center, but will respond to the activities pushed by the product department.

At this time, the data department can actively collect activities from various departments and display the full picture of how various activities affect users, based on users. This kind of panoramic data is rarely seen by business personnel in their own departments, and it is easy to trigger business thinking. Therefore, it is especially recommended that the data department actively collect various business actions and integrate them around the same business goal. This is also a way to reflect value.

VI. Summary

From the above, we can see that data analysis should be based on business and serve the business. This is not just empty talk, but can only be achieved by combining specific business scenarios and business needs, specifically discussing the feasible scope of the business, breaking down business problems, and answering them one by one.

Author: Down-to-earth Teacher Chen

Source: Down-to-earth Teacher Chen (ID: 773891)