Jim answers questions from fellow Drillers
(More questions with answers here, Work Overview here, Index of concepts here)
Topic Overview
Hi again folks, Jim Novo here.
Recency and the RFM model are both very powerful predictive models of customer behavior, But they’re not always the BEST models to use, because the nature of some business activities do not create the kinds of behavior these models are good at predicting. For example, any business – or segment of the business – that is likely to have a rhythmic repeating behavior (event occurs every week, or every month, or every summer, or every Christmas, etc.) will not benefit much from Recency analysis. But it is ideal for Latency analysis – when to expect an event, did it happen or not? Great example of both ideas in a single business would be the airline business – you have random flyers, and you have frequent flyers. Use Recency for random flyers, Latency for frequents. You dig? Great, let’s Drill it !
Q: Can you please direct me to specific information (in your site) regarding analyzing data in the Airline frequent flyer programs?
A: There isn’t anything specific to airline frequent flyer programs on the site, so I’ll create something though with this reply!
Q: Are there any “success methods” that proved to be the right way to define one flyer over the other?
A: Not sure what you mean by “define”… the triggers I have seen used in these kinds of programs usually have to do with changes in rate balanced against the value of the flyer. So, you look for slow-downs in frequency, for example, people who used to fly 3 times a week that now fly only 1 time a week. Their “fly rate” has dropped significantly and could be a flag for potential defection.
Q: I’m familiar with the RFM method, and wonder how to implement an RFM score considering that you have a:
* Flyer that is true loyal and doesn’t have any interest in flying to other parts of the globe (expensive long mileage ones) and yet,
* due to the method of RFM you will not find him at the “top customers”.
Well, I’m not sure RFM would be the right model for this kind of program. You want to look more at a “rate of change” kind of model since there are many levels of activity and Recency isn’t always a controlling factor.
So for example, you could create “Frequency buckets” based on deciles – divide customers into Top 10%, second 10%, third 10%, 4th 10% down to the bottom 10% based on their annual Frequency. Then track people based on how they are moving between the buckets. Somebody who was in the Top 10% that falls down to the third 10%, then falls lower in their annual rate would be a likely defector.
Q: Do you suggest modifying the basic RFM model, for example, R+F+M+(factor) instead of only R+F+M? (factor) could be anything that will enhance the information regarding the average-target destination-flyer behavior.
A: No, I don’t think a “factor” will do anything for you because RFM is not the basic model you want to be dealing with in this situation. Now, there are probably some **segments** where RFM would work – if you could identify them first. Affluent leisure travelers are such a group, because they have “free will” and the program is probably not very important to them in terms of deciding when / where they are going to fly.
RFM is best used in a “frictionless” situation where the population’s behavior is not influenced or controlled by external factors (such as would be the case with business travelers). There are too many behavioral cross-currents in a Frequent flyer airline program for RFM (in the classic sense) to be of much help.
So what you need to do is break the elements of RFM down and use them as appropriate to the business model. The “buckets” example above would be one way to do this, relying only on Frequency. You could make the buckets model more “sensitive” by adding a Recency or Latency component to the buckets to provide more predictive power.
For example, use the “average trip Latency” to provide another trigger. If on average, someone flies at least once every 30 days, and 45 days go by with no trip, that’s a trigger. At that point, you could then check the “bucket model” to see if Frequency is still in the same 10% range as it has been. If Frequency has started falling, then you know you have a potential defection and you could test what kinds of actions might influence the future behavior of the traveler.
Once you determine the behavioral segments, then you can look for any other “factors” you may have available to further flesh out the model (destinations, seasonality, average price, demographics, surveys) but you always want to do the behavioral segmentation work *first* so that your segments are actionable.
With that, now I’ll have something on my site about analyzing Frequent Flyer program data. Thanks for the idea!
Jim
Get the book at Booklocker.com
Find Out Specifically What is in the Book
Learn Customer Marketing Concepts and Metrics (site article list)
Download the first 9 chapters of the Drilling Down book: PDF