The RFM model: what it is, and why it falls short for hospital and clinic CRM in the AI era

The RFM model segments customers by three aspects of purchase behaviour: recency (when they last bought), frequency (how often) and monetary value (how much in total). It is a hugely popular CRM concept because it is easy to understand and needs only sales data, but in healthcare it often gives the wrong answer, because every service has its own care interval.

A LASIK patient immediately becomes a "champion" to RFM, thanks to high spend and several follow-up visits, and keeps getting promotions, when in fact they finished treatment as planned and need not come back. A dental patient who has not visited for seven months may score in the middle, when they are actually overdue for a cleaning. This guide explains what the RFM model is, where it came from, how it works, its strengths and limits in healthcare, and why CRM that follows each patient’s care pathway, automated with AI, fits better today.

Where RFM came from

RFM grew out of US catalogue and direct-mail businesses, which had to decide who should receive an expensive catalogue. Marketers found that customers who bought recently, often and for larger amounts responded best to the next one. The idea spread through database marketing in the 1990s, notably through Arthur Hughes’s book Strategic Database Marketing, and became a standard segmentation method in retail and e-commerce that is still used today.

The original context matters. RFM was designed for products bought again and again, in an era when one-to-one messages were impossible, so grouping customers into a handful of segments was the only practical option.

How RFM works

Each customer gets a score such as 5-4-3 and is placed in a segment, such as champions (high R and F), loyal (high F), promising new (high R, low F), at risk (used to visit often but not recently) and lost (low R and F). Marketing then sends each segment a different message.

What RFM does well

• Easy to understand: leaders and sales teams get it immediately, with no statistics needed
• Uses data you already have: customer ID, date and amount
• Works in Excel: no system or data team required
• Better than no segmentation: it ends sending everyone the same message

Where RFM falls short in healthcare

The core problem is that RFM measures every service with one ruler, while in healthcare when a patient should return depends on the service.

Other limits follow:

• Frequent is not loyal: in healthcare, frequent visits may mean frequent illness or treatment not working, which calls for clinical attention, not promotions
• High spend is not high long-term value: one operation pushes M up at once, though the patient may never need to return
• Segments, not people: two patients scoring 3-3-3 may have come for completely different reasons
• A snapshot: scores change at each recalculation but never say whom to contact when
• It can push over-selling: chasing low scorers with promotions can invite people who do not need more treatment

Compare RFM with care pathways yourself

Use the sample data from a made-up multi-service clinic and pick one service at a time. See how RFM segments those patients, against their real status on that service’s schedule, and the monthly return rate for the same service. Adjust each service’s return interval, or open your own file with a service column.

Watch the two yellow boxes: they are RFM’s two kinds of mistake. The first counts patients RFM calls at risk or lost who, on their service’s schedule, are not due yet or have finished treatment, so an RFM campaign would message them too early or needlessly. The second counts patients RFM calls champions or loyal, who would get promotions, although they have finished treatment; pick LASIK to see this most clearly. The blue box shows the people who really should hear from you now.

A better approach: 1:1 CRM by care pathway

Instead of three combined scores, ask the same question about every patient: which service did they have, and when, for medical reasons, should they come back?

1. Set each service’s interval with your doctors, such as dental cleaning every 6 months, health checks every 12 months, and post-surgery follow-ups at 1 week, 1 month and 3 months
2. Work out each patient’s due date from their last visit for that service
3. Message each patient as they come due, with content matching the service, such as "it has been 6 months, time for your cleaning", with a booking link
4. Stop when treatment is complete or the patient has booked, and hand over to the clinical team when behaviour suggests things may not be improving
5. Measure by service, using the on-schedule return rate; see patient cohort analysis

Why AI makes this possible

RFM was born when one-to-one messages were impossible, so a handful of segments was the only way. That limit has gone.

• The system knows every patient’s schedule from their visit and booking history, without anyone working through spreadsheets
• Messages can be personal: AI drafts messages matching each patient’s service, language and channel, such as LINE or WhatsApp, from the clinic’s real information
• Replies are handled at once: when a patient answers, AI answers general questions and books the appointment, handing over to a person for clinical questions
• It runs every day: no waiting for a monthly campaign; each patient hears from you on the day that suits them

Personal data needs care. Appointment reminders that are part of care are different from promotions: marketing messages need consent under PDPA, and messages must never reveal health information where others may see it.

Sources

• Arthur M. Hughes, Strategic Database Marketing (McGraw-Hill, first edition 1994)
• Wikipedia, RFM (market research)
• HDmall, Patient cohort analysis