All writing

Where the decision actually happens now

Where the decision actually happens now

Discovery on social. Verified by search. Checked against reviews. Closed on a marketplace. Collected in a store.

Which channel gets the credit?

The question is the problem. Somebody has to answer it, because budgets are split by channel and people are paid by channel, so an entire discipline has grown up to produce an answer that everyone involved knows is fiction.

What omnichannel actually means, stripped of the word

It does not mean being present in many channels. Every company is present in many channels and most of them are not omnichannel in any sense that matters.

It means the shelf is no longer a place. It’s a condition that follows her around, and it includes things you don’t own and can’t buy: a review, a comparison video, her sister’s opinion, an autocomplete suggestion, a marketplace’s ranking algorithm deciding you are the fourth result.

That last one deserves a sentence of its own. On a physical shelf you negotiate position with a retailer once a year. On a marketplace, position is recalculated continuously by a system whose rules are not published, and which is optimising for the marketplace’s margin rather than yours. Shelf position used to be a commercial negotiation. It is now, in large part, an input into somebody else’s model.

The evidence this is worth money

There’s a study I keep returning to because it measured behaviour rather than asking about it. Researchers tracked just over 46,000 shoppers at a large US retailer across fourteen months and sorted them by how many channels they actually used.

Seventy-three percent used more than one. Twenty percent were store-only and seven percent were online-only, which is worth noting on its own, since most strategy decks are written as though those two groups were the whole market.

The multichannel shoppers spent about four percent more per store visit and ten percent more online than single-channel shoppers, and customers using four or more channels spent around nine percent more in store. Within six months they’d made twenty-three percent more repeat trips (Sopadjieva, Dholakia and Benjamin, Harvard Business Review, 2017).

Now the caveat the headline version drops. It is one retailer, in one country, over fourteen months, and the direction of causation is not settled by the design. Shoppers who already like you more will naturally use more of your channels. Some of that nine percent is the effect and some of it is the selection, and the study cannot tell you the split.

I still think the finding holds, because the pattern repeats in client data I have seen that was never going to be published. But quote it with the caveat, because somebody in the room will know.

Attribution is a symptom

Here is my actual position, and it costs me work to hold it.

Attribution modelling is mostly an attempt to make a loop legible to an accounting system that only understands sequences. Last click is obviously wrong. First click is obviously wrong. The multi-touch models are less obviously wrong, which makes them more dangerous, because they produce a decimal place and a decimal place ends arguments.

Think back to the blender. Forty touchpoints, four of them visible, and the one that turned it was a colleague at lunch. Run any attribution model over that purchase and it will confidently assign credit among the four it can see. Every model will be wrong by roughly the same enormous amount, and they will disagree with each other in ways that look like insight.

What I do instead is cruder and more useful. Stop asking which channel caused the sale. Ask which channel was load-bearing, meaning: if you removed it, would the purchase still have been possible?

Reviews are usually load-bearing and almost never credited. Search is usually load-bearing for anything above a certain price. The store visit in the blender case was load-bearing and produced no transaction, which means in a channel P&L it reads as a cost centre and somebody is proposing to close it.

Where I was wrong about the store

For a stretch around the middle of the last decade I argued that physical retail was becoming a showroom and that the interesting question was how to monetise a visit that ends in an online purchase elsewhere.

That was half right in a way that made it useless. Showrooming is real, but I had the direction backwards for most categories. The larger flow was the other way: research online, buy in the store, because she wanted it today, or didn’t want to pay delivery, or didn’t trust the returns process, or needed to be certain it was the right one before money left her account.

The client where I made this argument had already started cutting store hours on the strength of it. What saved them was a category manager who pointed out that their own data showed the store’s busiest hours tracked the pattern of someone finishing work and collecting something she had already chosen. We were looking at a collection point and calling it a showroom.

The lesson I took is narrower than “I was wrong.” It is that the direction of the flow between channels is a measurable fact about your category and your market, and I had substituted a trend I had read about for a measurement I could have made in an afternoon.

The test that changes the conversation

For your category, write down the five things that have to be true for a purchase to be possible at all. Not helpful. Possible.

She has to be able to find you when she searches a phrase you don’t control. There has to be enough review volume that the absence of reviews isn’t itself a signal. The price she sees on the marketplace has to not contradict the one in store. The thing has to be in stock in the format her trip allows. Somebody she trusts has to not have told her otherwise.

Now compare that list to how your budget is split. In most companies I have walked through this, the load-bearing items are owned by nobody, funded by nobody, and measured by nobody, while three teams argue about which of them converted her.

That gap is the whole opportunity, and it’s available now, because your competitors are having the same argument.


The distributed shelf is chapter twelve of Shopper Marketing in the Era of AI. This follows The funnel is false the way an old map is false. Next, the thing she reads before any of this: Price is not a number, it is a signal.

More writing

5 min read

Price floors before the event, not during

Auto-buy means a promotion fires the instant you change a price. There is no pause button, no overnight reflection, no account manager ringing to check you…

5 min read

The number that keeps moving

AI-referred shoppers convert better than shoppers arriving from anywhere else. The figure is real, large, and not a plan. Here is the catch, and it matters…