>RFM modeling is a powerful tool for marketers. It is a concept taken from direct mail that has been adapted to online, and as we're going to look at today, email. RFM analysis and segmentation takes the sum of past behavior to predict future behavior and adjusts messaging to target that future behavior.

The methodology has, previously, applied only to past customers; today, thanks to the ease of tracking email behavior, it also can be applied to any behavior through a tool I call progressive profiling. During the analysis, we take a look at each piece of the "RFM" puzzle separately:

R = Recency

When did the customer last buy from you? When did the customer or subscriber last open/click? Take the entire range of data, from most recent (usually day of or previous day) through a pre-determined cutoff point based on your buying cycle. Let's say that's 3 buying cycles back -- everyone that purchased/opened/clicked prior to that get lumped into an inactive file.

Then, based on your buying cycle, determine the number and ranges of segments for recency of behavior. It could look something like this:

1: 0-30 days: 5,100 customers/subscribers

2: 31-60 days: 12,300

3: 61-90 days: 32,800

4: 91-180 days: 75,000

5: 181-365 days: 123,400

As we will continue to explore, your "best" segment should always have the fewest number of customers/subscribers and your "worst" should always have the most. They are the best and worst for a reason, so don't modify the parameters to try to shoehorn data to even out the segments. It should look like an inverted pyramid. In rare instances, your middle segments are the largest, but this is rare.

F = Frequency

This stands for the frequency with which a customer buys or subscriber opens/clicks. Amazon, for example, can and probably does, have a very high frequency reorder rate. The best RFM segments for a company like Amazon are probably very quick turnaround times, very frequent and very high dollar values.

Much like recency, we'll want to take a look at the entire range from 0-1 through X number of purchases/behaviors. Then, we'll want to take both the median and the average to determine the ranges for each segment. Again, remember not to sell yourself short when building the modeling to ensure you have "statistically significant" segments. It's more important to have your contacts in the appropriate bucket.

The data also doesn't have to be in even amount differences. If there are peaks and valleys in buying behavior, the segments should reflect that. It could look like this:

5: 0-1 purchases/behaviors

4: 2-4

3: 5-11

2: 11-15

1: 16+

M = Lifetime Monetary Value

The "M" stands for lifetime monetary value. Over their lifetime, how much is each subscriber/customer worth to you? For non-monetary analysis, you can score each behavior and assign a value to each.

As before, take a look at the ranges of the data, the median and the average to build your segment ranges.

5: $0-200

4: $201-350

3: $351-600

2: $601-1000

1: $1001+

What's important to remember is to take a holistic approach to combining these segments as someone that's a very high "M" could have potentially only purchased from you once 7 months ago, in which case would be considered a 5R-5F-1M, which could indicate dissatisfaction with your products, service or another unknown.

When conducting this analysis you can look at just email subscribers or you can open it up to your entire database for a more holistic view on what impact your marketing channels have on purchases, which produce the most valuable customers, how to utilize the various channels to generate desired responses and so forth.

From here, we will want to group each of the individual ranges into their RFM combined segments to determine the best messaging for each. I will be covering messaging in the next post.

Kelly Lorenz
Email Marketing Strategist at Bronto