Eight models of data analysis: detailed explanation of RFM model

Hello everyone , I am a little girl who loves learning xiong Xiong Mei .

What I share with you today is a frequently mentioned , But models that are seriously undervalued ：RFM Model .

One 、RFM The basic idea of

RFM The model consists of three basic indicators ：

• R： The time since the last consumption
• F： The frequency of repeated consumption in a certain period of time
• M： Accumulated consumption amount in a certain period of time

RFM In the model , The meaning of the three variables is very specific ：

• M： The more you spend , The higher the user value , The more we should focus on .
• R： The further away , The more likely users are to lose , The more you wake up the user .
• F： The lower the frequency , The more you need to use one-time means （ Like promotion 、 A gift ）, The more frequent , The more you can use continuous means （ integral ） To maintain the

therefore RFM Can directly derive action suggestions from data , It's a very easy way to use .

Two 、RFM The small example

Let's look at a specific example ： A taxi trip APP, Pressed RFM Format , Count user data （ Here's the picture , Sample data only 100 strip ）, The current leader requires ： Analyze users . How to analyze ？

First step ： First look at M. Distinguishing user value comes first , First identify who is the big customer , Who are the small customers , The idea of the later work is clear . We can use the tenths , Simply layer users , See which are key customers （ Here's the picture ）.

After grouping , You can open a PivotTable , Look at the proportion of consumption in each group .

wow ！ The first group of users contributed 40%+ Consumption , The first three combine , common 30% User contribution 74% Consumption , What a big customer , Therefore, it can be classified as follows ：

• The first group ：VIP3（ highest VIP）
• The second group 、 The third group ：VIP2（ The consumption of each group accounts for more than 10%）
• Fourth 、 Group five ：VIP1（ The consumption of each group accounts for more than 5%, Less than 10%）
• be left over 5 Group ：VIP0（ Single group consumption accounts for less than the whole 5%）

You can use one here IF sentence , To sort （ Here's the picture ）.

After classification, you can observe where the consumption threshold of each group is , For example, the threshold of the first group is 798 element / month . When making operational strategies , Probably for convenience , Find the nearest integer . So you can make a manual adjustment , hold VIP3 Change your store to ： Consumption within one month 800 element . Similarly , Other thresholds can be adjusted in the same way .

After adjustment , We have separated our major customers / Small customers , You can do the next classification . The next step is to R. How to determine the R What about the classification of ？ It can be determined directly according to the business characteristics . Like a taxi , Even if you need to take a bus again , It's impossible to go out every day , therefore R The value does not need to be set too short , Otherwise, they shout in people's ears every day ：“ Come on, take the car, take the car ”, It's too harassing users .

R Values can be classified in weeks . There are working days and rest days in a week , If the user really just needs , At the latest 1 It's time for Zhou to take a bus （ Here's the picture ）.

After sorting , You can make a crosstab , Observe the difference VIP Our customers are R Value distribution （ Here's the picture ）.

look ,VIP The higher the rank ,R The smaller the value. , and VIP0 Users of , Unexpectedly 80% already 2 I haven't been here for more than a week , Or there's really no need , Or it's lost . such , Yes VIP0 Analysis and suggestions , Also very clear ： Combined with the weather 、 The holiday season 、 Activities and other specific scenes , Give a small discount , Wake up users with a single taxi coupon .

For very high value ： Take out real gold and silver , Maintain a good relationship

For very low value ： Wake up regularly , The one that comes back is a

For not high, not low , We should distinguish between behavior .

For example, in this case VIP1 Type user , Two levels of activity Obviously , A wave of people are very active , A wave of people are silent , And their consumption power is almost the same . There are two basic strategies ：

For highly active , Launch a bundle XX Day discount package , Lock in subsequent consumption

Low active for , After a period of deep sleep , Launch large incentives , Stimulate secondary consumption

Under such thinking ,F It can be used as a reference , from VIP1 in , use F It is important to distinguish between high and low active people , Then formulate specific strategies .

So that's a simple one RFM analysis , And each customer group has targeted business suggestions .

If you just stop here , That's a pity ！ because RFM The value of the model goes far beyond that .

3、 ... and 、RFM A variant of

RFM The true value of , lie in ： It is a kind of Use time 、 The frequency of 、 Quantitative relation , Distinguish between light and heavy users Methods . In many business scenarios , Can use similar ideas to solve problems .

such as ： Examine the active behavior of users , Can also be divided into RFA

• R（Recency）： Last active time
• F（Frequency）： lately 1 Active frequency in the week
• A（amount）： lately 1 Accumulated active time in the week

Now ,RFA Combine , It can also clearly distinguish light and heavy users . also , according to RFA Combine , We can also find the next operation idea , For example, the following two users , It looks similar in general , But it can be based on behavioral characteristics , Set different content recommendation schemes , Activate user ：

Four 、RFM The shortcomings of

Be careful ,RFM The disadvantages of the are obvious ： It only takes into account the number of user actions , It does not consider that the user is What for? . For example, use RFM Examine user consumption , One key point is missing ： What do users buy . alike RFM The number , It may be quite different , such as ：

• R： - 30 The day is not spent
• F： lately 1 Months only 1 Secondary consumption
• M：1000 element

stay RFM Classification , Those who meet the above conditions are the same type of customers . But , If we find that ：

A user ： Take advantage of the big promotion , Hoarding 1000 Yuan shampoo 、 Shower gel. 、 hair conditioner 、 A paper towel

B user ： Take advantage of the big promotion , Bought a 1000 Yuan's air conditioner

Even if RFM The classification is consistent , We also know that ,A And B Users are two completely different kinds of people , Should adopt 2 Class to activate the consumption strategy . therefore ,RFM The model can be used , But it should be combined with the consumer categories of users , Think carefully .

author ： Sister Bear . New in the data world , Like data analysis 、 data mining .