Jim answers questions from fellow Drillers
Hi again folks, Jim Novo here.
Traditional RFM execution is focused on giving a snapshot view of customer likelihood to respond / campaign profitability across large and varied customer databases. But is that all it can be used for? Heck no! If you understand the basics of how and why RFM works, and you understand your customer database, there’s a ton of different very valuable customer scoring operations you can accomplish. Interested? Get out the Drillin’ tools …
Q: I recently purchase your book “Drilling Down.” Really enjoying reading it!
A: Well, thanks for the kind words!
Q: I had a question about the implementation of the RFM model against email campaigns. Say we have a client that has done this:
- Sent out 2 emails to entire database – in June and July
- Sent out 3 targeted emails to a specific segment of database – in June, July and Aug
From my CTR and Open Rates I know that the targeted segment performance is better. For my scoring I am using the following:
- Recency, last email responded to, and
- Frequency, number of emails where an action (a click-thru) was taken
So the question is when trying to apply an Recency / Frequency RF score to the entire database, do you / can you use all 5 email programs? Would Recency include the email to the specific segment in August? Would frequency include the segment that received the email in August?
A: The fact you are asking this question tells me you understand the methods better than you think you do. The correct answer is yes, and no, depending on the objective of the scoring. As long as you **understand** that there is the potential for the marketing to the target segment to skew the scoring of the overall group, then you are thinking about the problem correctly. Whether you decide to do the scoring as “everybody” or you score the targeted segment and then score “everybody else” separately really depends on what you are trying to accomplish / the objective of the effort.
Q: Wouldn’t the folks in the targetd segment potentially have a higher RF score?
A: Absolutely. But this is not bad, I mean, a bunch of them responded, so they “deserve” a higher RF score, yes? Isn’t response good, and so they should have higher scores?
If you look at the case for scoring the entire database, it generally tells you who is most likely to respond to **any** campaign. If you know a lot about your customers, you probably will not send them all the same campaign, but create different campaigns for different customer segments and hope to generate sales (or sign-ups, or downloads, or whatever). What you are really getting from scoring everybody together is identifying specific individuals or groups who:
1. are most likely to respond, and
2. appear to be defecting so you can be proactive and go after them with a specialized campaign addressing the potential defection.
You can either decide to attack certain groups or not spend the money because they are “already gone” and there will be no ROI. This doesn’t have anything really to do with any specific campaign, it is the more about the aggregate, overall decision to spend on any specific customer or group of customers. The fact you did a campaign to a certain segment has no bearing on this, because if you are successful with your campaign, those targets will have higher RF scores – and quite frankly, that is what you want, right? The higher the RF score, the more likely they are to respond and the less likely they are to defect.
Now, that said, the business of database marketing is about creating test programs, looking at the results, and realizing that certain segments respond better than others to certain campaigns. A classic example is the “discount ladder,” where you set the discount by RF score in order to maximize response and ROI. It is certainly OK and desirable to break the customer base into sub-segments and score these segments against themselves as well as all customers.
For example, score everybody who bought a lamp as their first purchase by themselves and everybody who bought a chair as their first purchase by themselves, or everybody who came from Google as a group and everybody who came from MSN as a group.
What you get from this approach is new insight and uncovery of new, profitable segments. So, for example, you find a customer segment with an average RF score of 35 within the overall scoring of the entire customer base, but it has a score of 55 (highest possible) when just scoring people who came from MSN. Though these people are not a “best customer” segment overall, they are “best” within the MSN segment and through testing you find they are generally responsive. They are best customer segments in terms of all segments from MSN, and as such, are probably worth targeting.
Another common use of multiple scoring groups comes into play for 1x buyers (see page 98 in the book). One time buyers by definition all have a Frequency of “1,” and online the percentage of 1x buyers in the database tends to be huge. So if you score the whole database together you get a very warped view of the world – if 80% of the database is 1x buyers, some of these poor quality customers will get fairly high scores.
A better approach is to split the database in one-time and multi-buyers. and score each group individually. This way, you have created segments which have very similar members, and a relative ranking score like RF becomes much more meaningful. When scoring the one-time buyers, you dump the frequency variable since it is the same for all, and use just Recency or perhaps Recency and Monetary. Then you test and see which approach is the best predictor of response and defection.
Your ability to formulate this question means you are on the right track. It’s not an either / or situation, it’s more like a “both” depending on what you are trying to do!
Download the first 9 chapters of the Drilling Down book: PDF