Your customer base, segmented and set in motion
One RFM-BT model reads every customer's recency, frequency, value, breadth and tenure. This is the model, the plan one cohort actually receives, and the proof it moved anyone, measured against a held-out control.
Where it sits: The Understand stage of the loop. Every other model reads its output: churn, pLTV and engagement all start from where a customer sits on this matrix.
A strategy, a stack of campaigns, and nothing in between
The strategy says grow complementary range penetration. The execution says here is a journey. Nobody records the bit in the middle: what one cohort actually experiences over a quarter, what they deliberately do not get, what would have to change for it to have worked, and what we already tried.
A cohort, not a send
The plannable object is a group of customers over a quarter, not a campaign in a calendar slot. That is what makes the quarter plannable.
A recorded state, so there is a baseline
Without a snapshot per customer you can only measure response, never whether anyone's trajectory changed. The snapshot is what makes movement measurable.
A written statement of what they do not get
The half nobody writes down, and where most of the coherence lives. Writing it down is what keeps the quarter coherent.
The model gives you the map to plan on: 9 segments worth targeting, each one dialled up or down by value, plus 4 groups that need their own treatment entirely. See how it reads →
Four steps happen without anyone, and one deliberately does not
The usual objection to a model like this is that it sounds like a lot of work. Almost none of it is work, and the part that is, is the part you actually want a person doing.
What runs without a human
- 1Weekly, upstream: recency and lapse are recomputed against each customer's own rhythm, and the state snapshot is written.
- 2The at-risk audience updates itself in Braze off the synced attributes. Nobody pulls a list.
- 3The holdout holds. It is assigned on a stable hash of the customer id, so the same people stay out with no list to maintain.
- 4The scorecard regenerates on the model's schedule, reading movement against the quarter-start snapshot.
- 5A human reads the result and changes the plan. That is the one step that is not automated, on purpose. Everything above re-runs on the new plan.
The refresh cadence sets the maximum learning rate of the programme. Persisting tier and state on the weekly refresh, rather than only the monthly recompute, is one line in the specification and four times the reads.
Not a funnel. A loop that keeps learning.
Score, segment, act, measure, re-score. The loop runs continuously, so a customer is never stuck in the wrong journey for long.
- 1Learn from the last cycleData platform + Braze
- 2Understand the customer, RFM lives hereData platform
- 3Decide across every signalData platform + Braze
- 4Predict what matters next, Churn · pLTVData platform
- 5Activate the next best action, RFM segments drive the sendBraze
- 6Measure the causal liftData platform + Braze
- 7Interpret, and change the planData platform
Your data platform is the intelligence engine, Braze the activation layer. Each revolution makes the next interaction smarter. RFM spans Understand to Activate: it gives every customer a tier and a segment, then pushes them straight into Braze to pick the message, feeding the Predict models along the way. And the loop only turns at Interpret: a human reads the measurement and changes the plan, and everything upstream re-runs on the new plan.
Follow it through
How it reads your data, who it turns them into, what that is worth, and how you know any of it worked. In that order.
That is RFM working on sample data. The service page has the rest: what it needs from you, how long it takes to build, what it pairs with, and how the lift gets proven.