In one page: the argument
Every retail executive team runs on two numbers: how many people came in, and what they spent. Both are good numbers. Neither is wrong. But both are taken at the beginning and the end of a store visit. Everything in between, which is where the purchase is decided, stays invisible.
Online channels never had this problem, and they did nothing clever to avoid it. A webshop was born decomposed: it sees the session, the category page, the product page, the seconds spent on it, the cart and the checkout. It does not know more about visitors than a physical store does because it is smarter. It knows more because the medium records it.
This essay makes one argument: the same decomposition is now available in the physical store, on infrastructure retailers already own, and without knowing who anyone is. Not as a forecast. Not as artificial intelligence. Not as a promise of return. As arithmetic, the kind that can be checked line by line by the person ultimately accountable.
It is an argument, not a pitch. The claim it makes is provable in a single store, and that is the standard it should be held to.
Revenue = Visitors × Reach × Engagement × Conversion × Average basket
The visitor count is measured at the door, conversion and spend at the checkout. The two terms in the middle have never been recorded, and those two are the ones that show which part of the store is falling short.
01 The two numbers that get the trust today
This argument does not begin from the position that the numbers used in physical retail are wrong. They are not.
Footfall is a good number. It is clean, continuous, comparable between stores and across years, and it is the only leading indicator physical retail has ever had. POS data is a better number still: it is audited, it reconciles to the bank, it drives replenishment, and it is the one place in the business where a claim is settled in cash. Almost everything defensible about how stores are run today rests on those two, and with reason.
The problem is not that they are inaccurate. The problem is where they sit. One is measured at the door. The other is measured at the checkout. Between them lies the visit itself: a visitor passes four departments, stops at one, picks something up, puts it back, looks for someone in a uniform, doesn’t find one, sees six people at the checkout, and leaves.
None of that appears in either number. Both record it only as an absence: one visitor in, no transaction out.
So when the predictable question arrives, why is this category down eight per cent, the honest answer today is a story and not a measurement. And as long as the only supporting number is that footfall is down too, the conclusion writes itself: more people have to come in. More campaigns, more promotions, more discount, more spend at the entrance.
It is a rational conclusion. It is also the most expensive conclusion available, and it is reached because there are only two numbers in the room.
The question is not whether the numbers are right. It is whether two measurements at the ends of a visit can ever show which part of that visit went wrong.
02 Online retail has never had this problem
Somewhere in retail, a share of revenue is being managed at a much higher resolution. And it is not the physical half.
Ask an e-commerce director why conversion dipped last month and the answer is a step, not a story. Traffic held, but the product page dropped out of the top navigation after a template change; far fewer people reached the page, add-to-cart fell with it. One day later it is fixed, and the fix is measurable.
The same question put to a regional manager of physical stores produces judgement, experience and anecdote. Often excellent judgement, from people who have walked those floors for twenty years. But not a step.
That difference has nothing to do with talent. It sits in the medium. From the very first webshop the record already existed: the session, the category page, the product page, time on page, the basket, the checkout. Nobody had to build it; it fell straight out of the technology. Analytics companies arrived later and made it presentable, but the data was there on day one.
The physical store was born the other way round. It records the beginning and the end and forgets the middle. Not because the middle is unknowable, but because nobody has ever recorded it.
The steps are not even different. They map almost one for one.
| Online | In the store |
|---|---|
| Visits the website | Comes in through the entrance: visitor |
| Opens a category page | Reaches the department: reach |
| Opens a product page | Reaches the aisle: reach |
| Time on page | Comes to a genuine stop at the shelf: engagement |
| Adds it to the basket | Picks it up, asks a colleague |
| Pays | Buys: conversion |
If a webshop reported only sessions and revenue, no executive team would accept it for a single quarter. It would be called unmanageable. That is exactly the reporting standard physical stores are held to.
03 What the online cookie actually did
Everyone surfing the internet knows what a cookie is, and most now know it mainly as a nuisance: the banner, the consent log, the regulator.
It is worth being precise about the job it did, because that job is far smaller and more mundane than its reputation suggests.
A browser has no memory. Every request arrives as a stranger. Left alone, a web server can see that a page was requested but cannot tell that the same visitor requested the next one and therefore cannot know that anybody travelled from the category page to the product page to the cart. The cookie is a small piece of text that lets the server tie those separate steps into one visit. That is all it is.
It did not sell anything. It did not predict anything. It gave a stateless medium a session. And with a session comes a funnel, with a funnel comes a drop-off, and with a drop-off comes a diagnosis instead of a total.
Twenty-five years of e-commerce optimisation rest on that one property. Not intelligence. Continuity.
Everything that followed, from testing to merchandising to layout optimisation, exists because somebody could finally see which step people left at. The practice followed the measurement. It has never worked the other way round.
04 This has happened before and it re-valued an industry
If the online comparison feels like a different business with different physics, take one from an industry with rooms, staff, rent and a fixed footprint.
Hotels ran for decades on occupancy. A full hotel felt like a good hotel. Everybody understood the number and everybody reported it. And it was, exactly like footfall, perfectly accurate and quietly useless on its own, because a hotel can fill every room by giving them away.
In the late 1980s the industry split it.
RevPAR = Occupancy × Average Daily Rate
The power was never in the new number. It was in the split. A full hotel earning too little is a rate problem. A well-priced hotel standing empty is a demand problem. The same disappointing revenue, opposite diagnoses, opposite levers, a different department accountable.
Two things followed, and the second is the one investors should notice. Hotels began to be run differently. And then hotels began to be valued differently: analysts stopped underwriting occupancy and started underwriting RevPAR, and a portfolio’s story became legible in a way it simply had not been before.
Retail has already done this once, incidentally, and thinks nothing of it. Nobody reports same-store sales growth as a single number any more. It is split into transactions × basket, because everyone accepts those are different problems with different owners. The decomposition of the visit is that same move, one level deeper, into the part that happens before a transaction exists at all.
Physical retail stands roughly where hotels stood in 1985. It has one abstract ratio, revenue divided by visitors, and no way to split it.
05 The arithmetic, and why there is no magic in it
Here is the entire proposition in one line, and it is deliberately not a model.
Revenue = Visitors × Reach × Engagement × Conversion × Average basket
It is a multiplication: an identity, not a theory. It has always been true for every store ever operated, in exactly the way transactions × basket has always been true. Nothing is being predicted, inferred or learned.
The first term is already measured. So are the last two, collapsed together, at the checkout. The two terms in the middle, how many visitors actually reach a given part of the store and how many of those come to a genuine stop, are the ones nobody records.
They are not exotic quantities.
Reach and engagement are measured within a zone. That zone can be an entire department, an aisle, or a single metre of shelf.
- Reach is: of the visitors who came into the store, what share arrived at that zone.
- Engagement is: of those who arrived, what share came to a real stop instead of walking through. A visitor who does that is engaged. The two words describe the same measure.
- Conversion is: of those who engaged, what share went on to buy.
- Average basket is: what they spend when they do.
Reach and engagement are measurements in a defined place, over a defined period, applied the same way every day. This is the point at which a CFO starts asking questions, and so precision is in order here.
| This is measurement, not prediction. | The same discipline the warehouse applies to stock and the online team applies to sessions, applied to shop floor. Consistency matters more here than cleverness, and consistency is the easy part. |
| Every number breaks into two smaller numbers. | A funnel that does not reconcile is visibly broken. There is no layer at which anything has to be taken on faith, and no step that cannot be requested. |
| Estimates are labelled as estimates. | On the screen, next to the figure, not in a footnote. Until the checkout data is connected, zone-level conversion remains the retailer’s own assumption, and it says so where the number appears. |
| Nothing here touches price. | No pricing, no markdown, no promotion optimisation. Those decisions stay entirely with the retailer, and there is no ambition to take them. |
The reason the middle of the store was never measured is not that the mathematics was difficult. The mathematics is the easiest part of this. It is that nobody had ever recorded it.
06 And a physical store does not need a cookie
This is usually the part that surprises the room.
Online had to invent an identifier because a browser is anonymous and stateless; it needed something to make the steps hold together. The store’s problem was never that. A visit is already a session. It has a beginning at the door, an end at the door, a duration, and a physical path through defined space. The continuity is inherent. Nobody has to invent it.
Which means the store can be decomposed without knowing who the visitor is. What is measured is movement and dwell: did somebody arrive at this zone, did they stop, for how long. Never a face, never an identity, never a person followed from one visit to the next. What is distinguished is behaviour: walking past, pausing briefly, or coming to a clear stop in front of something.
For an executive team, that lands in three places at once.
| Legal | This is not the programme with a two-year privacy runway where the impact assessment decides the business case. There is no identity to protect, no consent to collect at the door, no profile to erase on request. |
| Commercial | Online measurement has spent a decade under consent regimes, tracking prevention and browser policy that keep moving. The store version needs none of it, because there is no identity to consent to. The store’s funnel is not degrading; it has simply never been switched on. |
| Practical | The equipment is installed and already amortised. Cameras fitted years ago for loss prevention run every opening hour, produce exactly this record and discard it a few days later. This is not new hardware in the ceiling. It is a second use for what is already there. |
Online had to build an identity layer to see the middle of the funnel. The store already has the continuity and never needed the identity.
07 What changes on a regular Wednesday morning
Decomposition is not valuable because it is interesting. It is valuable because the diagnosis names both the lever and the owner.
Take the question every executive team asks, and follow the three possible answers: Category X is down eight per cent.
| What the numbers show | What it means | Who typically owns it | Where to look |
|---|---|---|---|
| Fewer people reach the zone. Stopping and buying unchanged. | They are not getting there at all | Store planning | Signage, the route from the entrance, what sits next to it, where it is in the store |
| They reach it but do not stop. Reach unchanged. | They walk straight through | Category and merchandising | The display, the price ticket, stock on the shelf, lighting, what it sits beside |
| They stop but do not buy. Reach and stopping unchanged. | Something loses them at the last moment | Store operations | Staff cover at that hour, product knowledge, the queue, stock in the back |
Today all three produce the same response, because with one number more traffic is the only lever available.
Decomposition turns a marketing-spend question into an operations question. Today footfall is the only visible lever, so footfall is what gets pulled, whether or not it is the cause.
08 What changes at chain level
A single store gets a diagnosis. A chain gets something considerably more valuable: variance that finally explains itself.
That the best store outperforms the weakest by a wide margin on comparable space is known. Which of the five terms accounts for the gap is not. And today the explanation defaults to the two least actionable answers in retail: the manager or the catchment.
Decomposed, the same gap becomes a specific statement. Store 14 receives the same visitors and the same reach as Store 3, but half the engagement in the same category. So it is not the location and it is not the traffic; something on that floor is different, it can be found, and it can be copied. Or Store 14’s reach is structurally lower because of a layout the shell imposes, in which case the honest conclusion may be that this is a store to re-fit or re-site. And that conclusion has become a number rather than an argument.
Three consequences follow, and all three are executive-level.
- The best store becomes a specification. Instead of the observation that every store should be a bit more like Ghent, there is the specific step at which Ghent differs, and therefore something a rollout can contain.
- A rollout becomes measurable rather than an act of faith. A change is recorded with its date, and the before-and-after is measured over equal windows against the trend the store was already on. So a rising tide is never claimed as a win, and a change made in a declining month is not condemned for the weather. That discipline matters far more than it sounds; it is precisely where most retail “proof” quietly fails.
- Capital allocation gains a second input. Refit, relocation and closure decisions are taken today on sales density and lease economics. Decomposition adds the question those cannot answer: is this store underperforming because of where it is, or because of how it is run? Only one of the two is fixable with capital. And getting it backwards is expensive in both directions.
For an investor the thesis is one sentence: a decomposed estate is one where operational upside and structural limitation can be told apart before the money is committed.
In closing: the middle is the only thing missing
Physical retail is not short of data. It is short of the middle.
The two ends are there and they are trusted. The arithmetic is already accepted: same-store sales growth has been split into transactions and basket for years without anyone calling it exotic. Online retail has been running on the split version of exactly this since the day it launched, and nobody finds that remarkable either.
The only thing that has ever been missing in the store is a record of what happens between the door and the checkout. It is produced every day, on equipment that is already installed, and thrown away every night.
Every store knows what sold. No store knows what almost sold, or why it didn’t.
That number is not out of reach, and it does not require knowing who the visitor is. It requires only the decision to record what is already happening.
