How can you use RFM in your marketing?

RFM is a popular model for retailers that profiles shoppers’ behavior. It segments them into easy-to-target personas; from your best customers to those who have stopped purchasing.

These RFM personas are created by giving each customer three scores. The scores are based on the recency (R), frequency (F), and monetary value (M) of purchases made. Recency is scored between 1 (worst) and 3 (best). Frequency and Monetary values are scored between 1 (worst) and 5 (best).

The RFM scores are relative and appropriate to your customers: a customer scoring three recency for you may not score three for another retailer.

How scores are created

Scores are created by a ranking method. As an example, let us look at how the Recency (R) score is calculated:

• For each customer, get the total number of unique days since last purchase and rank them (high to low).
• Split them into 3 equal groups (tertiles).
• Give each customer a score based on which tertile they belong to: 3 for the top tertile, 1 for the bottom tertile.

This ranking and scoring process is repeated for Frequency (F) and Monetary (M) values. However, F and M values are split in to 5 equal groups (quintiles), and are scored from 1 (worst score) to 5 (best score).

A better RFM model

The RFM model we have created at Dotdigital provides six standard personas to help target customers.

Whilst you can also create your own granular personas, those provided by the standard model do not overlap.

Non-overlapping personas are important and helpful. If you choose to send a coupon to a particular persona, you want to be sure you don’t accidentally send an additional coupon to the same people because they are part of another persona.

Better quintiles

Another way Dotdigital’s model is different is how we treat quintiles. We use a dynamic range rather than a fixed range.

Quintiles in FM are commonly made up of an equal number of customers (a fixed range). They are straight 20% slices of F, and M dimensions.

Fixed-sized quintiles are potentially flawed if you consider that a customer may be scored as a 5 or a 4 Monetary score based on small variances in their data. For example, someone who spent $673.10 should have the same Monetary score as someone who spent $673.11. With a fixed range model they may become different personas.

To create dynamic range quintiles we normalize the data. “Normalize” here just means we make the data easier to compare.

• Recency is grouped by unique days since last purchase
• Frequencies are grouped by unique values
• Monetary amounts may be subject to rounding and grouping

This approach leads to quintiles that are not evenly sized. This is a good thing. Customers are scored accurately by appropriately dealing with small data variance that could make them jump a whole persona.

Combining F and M keeps it simple and accurate

The final part of the model is to add up the F and M dimensions. This gives us an R dimension with a potential score of 1 – 3 and an FM dimension with a potential score of 2 – 10.

F and M are added together to help ensure segments do not overlap and to allow for easy visualization of your customers by RFM persona. This is done with a treemap.

Other RFM models treat this problem differently (such as throwing out a dimension). We think a combined FM score is a good compromise between accuracy and simplicity.

Conclusion

RFM is an easy way for retailers to extend their current behavioral targeting and reporting.

Key benefits:

• Data-driven personas with little effort.
• Easy segmentation and context for your marketing activities.
• At-a-glance overviews of your entire customer base.
• Risks and opportunities are visual and actionable.

When combined with retailer reporting KPIs, RFM gives you new and interesting ways to slice your data and find out where the money is, so that you can optimize your strategy and increase conversions.