Behavioural archetypes vs RFM: reading the shopper, not the receipt

She lands on the store at 9:14 on a Tuesday night. Boots, size unknown, no account, no cookie worth the name. She reads the product description twice. She opens the size guide. She scrolls to the reviews and stays there. She adds to cart, then backs out to compare a second pair. She reads the returns policy. She hovers on the checkout button, and closes the tab.

Two systems watched that session. They saw two entirely different things, and the difference between them is the whole argument of behavioural archetypes vs RFM.

Nothing. RFM saw nothing at all, because she did not buy.

RFM (Recency, Frequency, Monetary) is the segmentation model most Shopify stores still run on, usually without naming it. It scores customers on three axes: how recently they purchased, how often, and how much they spent. Steelman it fairly, because it earned its longevity. RFM is cheap to compute, legible to a board, and genuinely predictive of repeat behaviour among people who have already bought. In the direct-mail era it was the elegant answer to the single largest controllable cost in the business, postage. Score the house file, mail the top deciles, skip the bottom. Recency predicted response better than almost anything. Catalogue empires were built on it, and for the job it was built to do, it remains reasonable.

But notice what it requires. It requires a purchase. Until the sale closes, an RFM system has no row for this shopper, no score, no segment, no opinion. She was invisible for the entire session, and she left invisible. Had she checked out, she would have become exactly one thing: a new row in a table, recency equals today, frequency equals one, monetary equals the order value. A record written in the past tense about a decision already made.

What RFM is. RFM segmentation is a scoring method that ranks known customers by how recently, how often, and how much they have purchased. It is backward-looking by construction, because every one of its inputs is a completed transaction, and it can say nothing about anyone who has not yet bought.

The second system watched the same session and classified her before she reached checkout.

It read the sequence, not the sale. Two reads of the description, the long dwell in reviews, the add-then-retreat, the pause on the returns policy, the hesitation at the button. It did not need her name, her email, or a third-party cookie to do this. It needed her behaviour, which she was broadcasting freely, and which is legible if you are built to read it. By the time her cursor was on the checkout button, she had been classified as a Hesitant Abandoner: high intent, high consideration, one unresolved friction away from buying. Not a Price Checker bouncing between tabs for a discount. Not an Impulse Buyer who would have converted on the first screen. Someone who wanted the boots and needed a reason to trust the purchase.

What a behavioural archetype is. A behavioural archetype is a real-time classification of a live visitor by the intent their in-session behaviour reveals, inferred from what they do rather than from anything they have bought. It applies from the first visit, before any transaction and before any identity, and it updates as the session unfolds.

Hold those two definitions side by side and the fault line is immediate. One is a ledger. The other is a lens.

The instinct is to call these tools complementary, one for the past, one for the present. They are, and a mature store runs both. But the framing that matters for behavioural archetypes vs RFM is coverage, and on coverage it is not close.

Here is the fact that should reframe the whole comparison. On a live storefront, the anonymous, pre-purchase visitor is not an edge case. It is the majority. Most people on any given day have never bought from you, are not logged in, and will decide whether to convert without ever entering the population RFM is allowed to score. RFM cannot compute for any of them. Not badly, not approximately: it has no inputs. Recency of what? Frequency of what? A first-time visitor comparing two pairs of boots has a purchase history that is the empty set, and an empty set has no percentile.

So the model does the only thing it can. It goes silent on exactly the people whose decision is still live, and speaks confidently only about the minority who have already decided at least once before. This is the structural blind spot, and it is blind by design, because RFM was built for a world where the only data you had was the receipt.

There is a second flaw underneath the first: RFM runs on the wrong clock. It was designed for the monthly mail drop, so its natural verb is a batch job that scores the file overnight and hands marketing a list. But a shopper's intent lives and dies inside a single session. When RFM finally has an opinion about our shopper, the tab is closed, and the only lever left is a retargeting ad chasing her across the internet. The archetype had its opinion while her cursor was still on the button, which is the one moment a reassurance about free returns could have changed the outcome. Intent before identity is not a slogan. It is the difference between acting during the decision and reporting on it afterwards.

Behavioural segmentation on Shopify inverts every one of RFM's assumptions. It does not wait for the receipt; it reads the session as it happens, the depth of scroll, the dwell on price, the pogo between variants, the return to the same product. Because intent expresses itself the same way whether or not the browser is carrying a name, behavioural archetype classification runs on the anonymous majority from the first meaningful action, cookieless and server-side, and yields a predictive CLV from session one, an estimate of what a visitor is worth before they have spent a penny.

The credible objection is that "reading intent" sounds like astrology. It is not, because it is disciplined and it is few. There are five canonical archetypes in this model, and only five:

• Researcher, methodically accumulating detail, reading specs and reviews before committing.
• Comparison Shopper, moving laterally between options, holding two open at once.
• Price Checker, whose attention keeps returning to cost and the signals around it.
• Hesitant Abandoner, who reaches the threshold of purchase and steps back.
• Impulse Buyer, who compresses the whole journey into a short, decisive arc.

None of these requires a purchase history to detect, because none of them is about purchase history. Every visitor is scored against all five as a live probability, not stamped with one label for life. Our shopper drifted from Researcher (the careful reading) towards Hesitant Abandoner (the add-and-retreat) inside a single visit, and the classification moved with her. A Hesitant Abandoner and an Impulse Buyer might carry identical RFM scores, or far more often no RFM score at all, and yet they need opposite treatment in the next few seconds. RFM cannot tell them apart. Intent-based classification is built to.

This is the ground Auraflow is built on: that you can read intent honestly without knowing who anyone is. The intelligence runs server-side on per-store models, because a Comparison Shopper on a fashion label behaves nothing like one on a supplements brand, and a single global model would flatten exactly the distinctions that make the read worth having. It is pseudonymous and consent-respecting by construction, and it lives at kosmatic.com. That is the last we will say of it, because the argument stands on its own.

RFM is not wrong. It is just late. It answers its question, of the customers I already have, who was valuable, accurately and cheaply, and that question still has a place in a retention calendar. But a record cannot change an outcome, and our shopper did not need to be remembered. She needed to be recognised while her cursor was still on the button.

The direct-mail era optimised the receipt because the receipt was all it had. The storefront era has the whole session. Reading only the receipt now is not caution. It is choosing to keep your eyes closed for the part of the visit that actually decides the sale. RFM tells you who was valuable. Behavioural archetypes tell you what this visitor intends, right now. Only one of those lenses is pointed at the future.

Is RFM segmentation obsolete?

No. RFM remains a sound way to rank and reward existing customers by the value of their purchase history, and for loyalty tiers and win-back flows it still does honest work. Its limitation is not accuracy but scope: it cannot see or score the anonymous, pre-purchase majority of your traffic, and it can only act after a sale. Behavioural archetypes cover exactly the population RFM structurally cannot reach.

Do behavioural archetypes need cookies or a customer's identity?

No. Behavioural archetype classification is cookieless and pseudonymous. It infers intent from in-session behaviour, the sequence of what a visitor does, rather than from a third-party cookie or an identified profile, and it runs server-side on per-store models. A visitor can be classified and acted on without the store ever knowing who they are.

Can you combine RFM and behavioural archetypes?

Yes, and they are complementary rather than redundant. RFM describes the value of the customers you already have; behavioural archetype classification describes the intent of the visitors you are about to win or lose. Used together, one governs how you retain and the other governs how you convert, but only the archetype layer can act during the live session.

How is this different from just tracking events in analytics?

Event analytics tells you what happened, in aggregate, after the fact. Behavioural archetypes interpret the same raw behaviour live, as a classification of one visitor's intent while the session is still open, and support predictive CLV from session one. The difference is between a report you read tomorrow and a read you can act on tonight.