RFM: The Gateway Drug to Buying Stronger Insight

RFM is one of those Trojan horse propositions for insight and analytics agencies when working with businesses new to data-driven customer analysis. It is an extremely valuable piece of work in its own right, while providing ample opportunity to showcase the value an insight consultancy brings, possibly more so than would otherwise be noticed.

While one could consider RFM to be a gateway drug that gets businesses hooked on insight consultancy, like any drug, it can have devastating side effects.

What is an RFM Segmentation?
RFM groups customers based on how they buy. It tracks:

Recency: When was their last purchase?
Frequency: How often do they buy?
Monetary Value: How much have they spent?

By sorting customers into segments based on these three factors, revenue is linked to different sources of spend. These segments can then be used to prioritise engagement efforts and shape retention strategies.
What follows is why I like it, why it can be dangerous, and how to fix it.

Why I like it

1. Easy to Sell-in & Deliver
Unlike segmentations based on survey research, RFM projects run smoothly and can be costed appropriately. This is because the role of the analyst is more about description than creation or explanation. The inputs are already valued, understood and fixed. In other words, “It is what it is”, easy to grasp, with very little room for debate. The analyst takes on no risk, nor is asked to make additional efforts to sell the idea in.

2. Tactical
Coupled with this, RFM either sits within a single CRM system or can be merged with email and web analytics. RFM can then be overlaid with the content, features and product lines each customer engages with. Stakeholders appreciate this value, its direct link to supporting retention strategies, more efficient customer acquisition and upselling.

3. It calls for more
Once a business becomes comfortable with the above, they’re open to exploring the breadth of analytical options. The real need to research the customers beyond mining a CRM becomes clear and pressing. The many avenues for customisation will generate enthusiasm for deep-dive analyses and marketing automation.

It is the perfect springboard from which businesses enter a mature, full-circle view of their customers, helping them understand why customers find and move between segments.
Even basic things, like overlaying segments by region, can be eye-opening.

Why I don’t like it
A lazy RFM setup backfires silently. Naïve marketers who fail to push their insight journey along quickly enough become fixated on RFM and will kill off their business in a doom spiral.
– It’s only as good as the data. Most companies can’t track all relevant customer transactions. They are partially or fully blind to the research–purchase funnel. Some marketers are comfortable operating as if they always have the full story. Some data feels better than none. Quick data feels better than slow. RFM is both things. Marketers are lulled into assuming they know what is happening underneath dashboard metrics.

1. A lot of data on a customer is treated on par with having almost no data.
RFM encourages leaning in too heavily, too quickly. Newer customers, despite being allocated their own segment, are misclassified. Even at this early stage, they ‘could be anything’. In most retail situations, many customers buy once and never return, or if loyal, take a while to establish a pattern of behaviour. RFM assumes context, dumping new customers into low-value segments and alienating potential high-value customers who would otherwise be nurtured more appropriately. Collecting more information before making up your mind where a customer belongs runs counter to what the framework is asking you to do.

2. Can be seasonal.
If a business runs discount campaigns or changes its spend or strategy, RFM will shift dramatically. Segments have no stickiness. Notwithstanding, any RFM analysis conducted in January will look completely different by May or November. Yet treating this as more than a snapshot, as some marketers do, makes for expensive decisions.
Sensitive to instability. Similarly, if a proposition is affected by instability not generated by the marketers themselves, such as competitor action or financial markets or exchange rates, RFM becomes unknowingly distorted. As this is almost always true, any RFM is permanently glitchy to some and varying degrees.

3. It’s backward-looking.
Because RFM is descriptive, it only tells you what already happened. It assumes past behaviour will continue, which is never true over the long term.

4. It’s self-perpetuating analysis.
It leads to biased marketing spend (that is what it is meant for). If high-RFM customers tend to be from a specific geodemographic, a company may unintentionally neglect other areas of growth. The efficiencies it brings are not necessarily a net positive.

5. It creates a hierarchy that can backfire.
People talk. Customers know other customers. If high-RFM customers get better service or discounts, others may breed resentment and destroy trust overnight.

6. It can be creepy.
Over-personalised marketing can make customers feel like they’re being watched. Conversely, if a customer is in the wrong segment for any of the above reasons, messaging creates disconnect. (It is easy to be overzealous when setting up platforms which co-ordinate email marketing with baskets and RFM.)

Fixes
There are ways to improve RFM segmentation and ameliorate its many pitfalls, but this calls for investment and full-time attention.

1. Automate cleaning to remove anomalies.
Remove gifting, returns, and bulk purchases programmatically.

2. Keep validating your segments.
Always use holdout sets, past vs current data, and check if segments correspond to initiatives as intended. Without continually testing the framework against past or future data, RFM can conceal faulty assumptions.

3. Treat RFM as a dynamic system rather than a segmentation. Focus on trends rather than snapshots in time. Build in segment ‘stickiness’.

4. Use Mixture Models and Trees.
I personally believe that the binning is too rigid. The all-or-nothing allocation lies at the heart of the problem. I like to combine the above progressive (sticky) approach with Gaussian Mixture Models. I want to avoid strict segments, apply confidence metrics, and be more considerate to ‘clumpy’ distributions. Mixture models work well with other measures and remedy many of the above failings by smoothing out arbitrary cut-offs. We no longer need to pretend the axes are equal and meaningfully linear. It can also reflect the fact that some segments will vary in coherence over time. I want to be able to allocate a segment, but then also score how confident I am about its allocation, how internally coherent that segment is, and its direction of travel in relation to others. I hope this is then used operationally to ensure bolder interventions are deftly targeted. This should be combined with decision trees.

5. Pair with surveys and experimentation.
Overlaying or developing the segments further with behavioural data is a start. Using qualitative research helps but is unlikely to catch a customer in time to be relevant, or may be too specific, to flesh out underlying drivers. A handful of customers from each segment won’t reveal anything systematic, because of how the segments are constructed. Primary surveys, however, especially ones bringing their own ad hoc segmentations into the fray, offering different lenses, can explain why customer groups behave the way they do.

6. Monitor and update.
RFM isn’t a one-time analysis. If the world stood still, habits would still shift. Product life cycles would still wax and wane. Here, a simple dashboard showing RFM segment trends over time, with warning flags at both the customer level and across the segmentation solution itself, catches drift before it skews strategy.

RFM segmentation is useful and widely accepted but riddled with problems due to its simplicity. The closer one looks at it, the more it is lacking. Used in isolation, it’s a silent killer. These are the very ingredients of the ultimate gateway drug, an invitation to seek out the more potent services of insight agencies who know how to put in the above checks and balances, to leverage and evolve their client’s insight journey.