Practical RFM analysis to increase repeat sales

Almost half a year ago I wrote an introductory article what RFM is. The first thing I did before writing the previous article - I had prepared an RFM analysis for our platform, to understand the picture with our clients. I liked this method as the easiest way to visualize the behavior of the clients. Yes, it was interesting, but very soon I realized that there is no practical benefit in this picture.  

Numbers mean nothing if they do not change our actions tomorrow.

This time I will share with you the experience and tips how to benefit from RFM even without a three-year sales history. We provided the RFM analysis to dozens of different shops and faced a number of obstacles on the way to achieving real results. Before to describe them all, let us once again recall what RFM gives us. It is based on three indicators:

 
I think the Monetary indicator is not necessarily tied to money. It can be any filter superimposed on the basic parameters of this method: Recency and Frequency. For example, the information portal can be considered by views of the page or the depth of viewing pages on the website. RF matrix can be considered separately for different types of clients, filtering by the client's source, by region, by category of purchased goods, by age, and more.
If you divide all your clients by recency of last purchase into several groups and the number of purchases made, it is possible to construct a matrix which shows how to divide your clients into groups based on their activity:

There are much more groups, but we will get back to them later. I usually build scales by frequency and recency, according to the principle: from bad to good. So by the recency: on the left, those who bought a long time ago, and on the right - who recently bought. By frequency: those who made a single purchase are lower, and who made a lot of purchases - on the top. Then the visual separation of groups will look something like this:

We have mentioned it before, but let's look in details how to split the base into groups and mark scale, how many groups to allocate and what to do with them next.

The most important indicator is — «Recency». The more time has passed since the last sale, the less likely the next purchase. And this probability drops very rapidly. Let's review the "Recency" from the beginning. For example, we have three clients (square, circle, triangle) and on the graph, we noted the time when each client had made a purchase.

Please note, for some reason I drew a graph of limitations in the opposite direction from good to bad. I am sorry and promise to correct it later :(

The task is to determine how many clients fall into each of the periods by recency. It is important to realize that the recency is the indicator which takes into account only the last purchase, previous purchases will be marked by the frequency. So the correct answer to the puzzle is:

If you calculate how many clients fall within each segment, it is possible to build charts, which may look something like this:


Once again there is a question: which of the three options is the best? 

I often hear that «the red (3)» — because it is stable in fact, it shows that we are constantly attracting clients who make a purchase and do not come back again. It is good that figure is slightly growing, but still «purple (1)» is better. As it can be seen that the majority of our audience recently bought something and is strongly involved in the buying process. Of course, the worst is the — «green (2)» график. In this case, we had a surge of activity (may be New Year or strong investments in context), and then we lost everything.

Most books on the analysis, including the popular edition of «Strategic Database Marketing»by Arthur M. Hughes, proposed a very simple mechanism of separation of "Recency" by segments: sort all your contacts by recency and divide into 5 equal groups. The same is recommended to do with the frequency and the monetary:

This method works and helps to break all the clients on: recently bought, rarely, long time ago, a very long time relatively to each other. But what it is «when»? If we can not say exactly «when», then it is too hard to evaluate this group. For example, in the neighboring group can get contacts that bought something in the same day. It means that we will treat differently the same clients... Why then do we conduct such an analysis? I will try to show the entire depth of the problem on two examples:

Example 1: Everything was well in your store, and suddenly no purchase within a month. Everything is bad, but if you split into equal parts - nothing will change, and the last buyers fall into the segment «recently bought».
Example 2: If you do not have sales data for several years yet, and you have just started your business, the borders will be very different every day, and there is no point to use the results of this analysis.

Seasons, promotions and holidays also strongly influence the overall picture.

Any contact list’s segment, which you can not say anything specific about is the poor segment

It seems to me logical to fix not quantity, but time. Then the optimization of the number in the group will be our priority. We will need to get more contacts in the group, «recently bought» and as little as possible in «a long time ago». 

 

It remains only to determine which time limits are better to expose. To do this, ask yourself a few questions:

For companies which do not have the data, these periods can be assumed. There are some charts with the difference between purchases (which I will discuss later) might help for companies «with history».


I gave an example of the schedule with the difference in months between the first and second purchases. Most stores have similar pictures as in Fig. Here we see that 50% of all repeat purchases were made during the first two months. During the six months, up to 75% percent make their second order and 90% of those who buy again — buy within 13 months. Which one of you can make a practical conclusion? I think, there is no need to give a discount to those who buy without our help. Discount prices should be 2 months after the first purchase. Before this, it is necessary to show the relevant offers, to be a household name and the main thing — to be useful to the client.

If you build a matrix difference between any adjacent purchases, you get approximately the following: 


It is interesting, that the difference between the first and second purchases is always greater than between the second and third. And between the second and third purchase is greater than between the third and fourth, but the more purchases, the less influence to the difference between the each next one. It turns out that a person needs some time to believe at first. Maybe next time he/she will try to buy elsewhere or just will wait for a little. But the more often he buys, the less time between purchases. Jim Novo in his book «Drilling Down« says that the diagram falls in the beginning and then stabilizes, and then begins to grow. This is very similar to the life cycle of the client, when he buying less and less, gradually ceases to be the client. But I've never seen this picture in practice.


Please note:: to obtain these figures better to use the median, then the average . I will try to explain the difference. The average is the sum of all values divided by their number, and the median is the value of this element, which is in the middle in the sorted sequence. That is guaranteed 50% of all values are less or equal to the median and the other 50% are greater than or equal to the median. The median is better than average because it is free from influence of «outliers». Those values, which are very rare, but much beyond the most. Typically, the median is much less than the average.

It would be good to take a look the seasonality that is usually 12 months. After looking at all the figures, we can conclude how to sort the groups by the period. For example:

Great email-marketing

Try Now

But even with this approach some issues remain open. And what if some clients have a short natural period of purchasing and others have a long one? In this case, I create a profile for each of the client and a gradation for each profile which ought to be normalized by recency axis to relative categories:

There is the same story for the «Frequency», but almost always I get the similar scale. Those who made:

Well, that is it for RF analysis. If you look at the RF matrix, knowing how to divide segments, then it is clear who are the newcomers: those who made a purchase recently. Disposable — made a purchase a long time ago, and probably have already forgotten about it. It is unlikely that they will come to us again. Jim Novo said that these clients are always 50-60% of the total client base, and this should be accepted. Usually, I see 70% of the base in this “sad” segment. In order to make sure that the segment is “sad”, you can try to return someone and for that Jim Novo offers a strategy — «to accept», but if you do not believe, start with:

  1. those who have a few items in bill
  2. those who have a bigger bill
  3. those who have a bigger bill- probably, they will come to you for a second purchase, than those who are «happy with everything».
If you can not get them back - you won't get back anybody.
The only way to get the clients back for the second purchase - is to contact them as soon as possible but not too early.
If you contact them too early, you will be intrusive and offer extra murderous for business discounts.
 
Let's look at the RF in the dynamics. It is worth to note that we always start with a good (recently bought) on a scale by recency and slipping into bad (have not bought for a long time). And as soon as the purchase is made, we always come back again to the most optimistic segment:
 
 
There is an opposite situation with the frequency. We always begin with the poorest segment (1 purchase) and it is getting better with each next purchase. You almost have no chance to get back (if you count the frequency of a certain field (eg, 2 years), the frequency may be reduced)
 
 
 
And now look at the whole picture in the RF matrix:
 
 
 
We always start in the segment «recently bought" and «1 purchase». We dream that all clients were in the VIP segment. And the quickest way to do this is a permanent purchase. If the client does not do anything from the beginning, he «slides» by recency into «rarely» in the segment of «one-time client». Our goal is not to let him go, that is why automatic reactivation emails have been created ;)
 
 
 
There is another sensitive issue: what if a pause «delayed» and the client did not buy anything for a long time (significantly exceeding the client life cycle). I think, in this case, he will come to us again, only if we «buy» him with good context or SEO or he will just stop liking the shop where he had been bought it all before. In any case, we will come to a completely different person with different interests and abilities. Therefore, we propose to make a «reset» like in a children's game.
 
 
 
The main feature of this approach to the RFM is that it can be applied on the first day of using the online store. You can immediately set up triggers based on the client inactivity and to observe the dynamics of the activity of your clients. Who are you going to lose? What should you do to get clients when it is still possible?

That’s it for now. I promise to write one more digest where I will try to give answers to the following questions:

I wish you to have good clients who go by the shortest path: from newcomers to the VIP :)

 

Get professional expertise

🔒 GDPR, CCPA, CASL Compliant. Your data is safe and secure with us.