Your customer is not
moving through stages.
They are building a case.

What is the ecommerce
customer journey?

The ecommerce customer journey is every encounter a person has with a brand before, during and after a purchase. Most maps draw it as five stages in a line. Google stopped drawing it that way in 2020, after testing 310,000 simulated purchases with 31,000 in-market shoppers across 31 categories.

That study, Decoding Decisions: Making sense of the messy middle, describes the middle of the journey as a loop rather than a descent. Shoppers move between two modes. Exploration is expansive and adds options to a consideration set. Evaluation is reductive and narrows them. People loop between the two until they are ready to buy.

A loop with no fixed length is awkward to plan against, so most teams keep the linear map and quietly use it as a scheduling tool instead. That is where the damage starts. A stage tells you roughly where somebody is. It tells you nothing about what they still refuse to believe, and the second question is the one that decides whether they buy.

The corpus is The Handoff, a running collection of ads captured first-hand alongside the page each one points at. As of today it holds 253 published entries across 207 brands. It is the evidence underneath most of what follows.

Do you still need an
ecommerce customer journey map?

Yes, for operations, and no, for deciding what to make next. A stage map is an ownership document: it records which team touches a buyer at which point, and where a handoff happens. That is worth having. It just cannot tell you which argument is missing, because no stage holds a belief.

Customer journey mapping came out of service design, where the stages are real events a person moves through in order: book the appointment, arrive, wait, be seen. A purchase journey is not that shape. Google's own testing put shoppers in a loop, adding options and cutting them back until one wins, and a loop drawn as a line loses the only thing worth knowing about it.

Keep both documents if you like, and give them different jobs. The stage map answers who owns what, so use it for staffing, service levels and handoffs. The belief map answers what is still unproved, so use it for briefs, budgets and what gets built next. Trouble starts when one document is asked to do both, because the stage map always wins that argument. It is older, and it has a box waiting to be filled.

Stop asking which stage a buyer is in. Start asking which of four things they still do not accept. The rest of this post is that question checked against real ads.

Why does running the same
claim everywhere feel
like it is working?

Because repetition does move something, just not the thing you need. Robert Zajonc showed in 1968 that repeated exposure to a stimulus raises liking for it with no new information involved at all. Familiarity is real, it is measurable, and it is not belief.

A second effect sits on top of it. In 1977, Lynn Hasher, David Goldstein and Thomas Toppino found that people rate repeated statements as truer than ones they have seen once. It holds whether or not the statement is true (Frequency and the conference of referential validity, Journal of Verbal Learning and Verbal Behavior). So the fifth run of your claim does buy you something. It buys a small lift on the one belief that claim addresses.

The problem is the other three, which got nothing, because nothing in the plan was ever assigned to them. The dashboard cannot show you that. It shows frequency going up and a claim getting warmer, which looks exactly like progress.

This is also roughly where the rule of seven comes from, and it is worth knowing the rule has no study behind it. The numbers attached to it have moved for a century: Thomas Smith put it near twenty in 1885, a 1930s Hollywood story put it at seven, and Jeffrey Lant fixed the B2B version at seven inside eighteen months in 1989. The serious research argument has run between one exposure and three.

Herbert Krugman's 1972 paper is titled Why Three Exposures May Be Enough, and his three are not three repetitions of one message. They are three different jobs. The first exposure answers what is it. The second answers what of it. The third is where the decision happens. Anything past that, in his reading, is reminder. Read that way, the oldest frequency research in advertising is already making the argument this whole post rests on: different exposures, different jobs.

What do 253 teardowns show
brands actually running?

Identity, mostly. Across the 253 published entries in The Handoff, the most common mechanic by a wide margin is identity appeal, in 110 of them. Objection handling of any kind appears in 51. Risk reversal appears in 7. The pile is built to be recognised, not to be believed.

The ads behind that table were captured first-hand from the Meta Ad Library between 9 February and 30 August 2026, across 207 brands and 10 categories, so this is what was observable there, not a random sample of DTC advertising. And the mechanic labels are mine, assigned on review. Each label records what an asset does, and none of them claims to know what the brand meant to do.

Entries carry more than one mechanic, so the column does not sum to 253. The median entry carries three. Twenty five distinct mechanics appear across the corpus and all twenty five are in the table, so nothing has been filtered out to make a point.

The mechanics that win a scroll sit at the top of that table and the mechanics that answer what if this does not work for me sit at the bottom, and the distance between them is the article in one column. Risk reversal, the cleanest way there is to retire that objection, shows up 7 times out of 253.

The rarest mechanic is also the one with the oldest evidence behind it. Two-sided messaging, admitting a real flaw, appears 6 times. Elliot Aronson, Ben Willerman and Joanne Floyd showed the opposite in 1966. A competent performer who commits a small blunder is rated more attractive afterwards (The effect of a pratfall on increasing interpersonal attractiveness, Psychonomic Science). Six entries out of 253 are willing to try it.

Does the job change when
the buyer is further along?

It does, and the corpus shows the shape of it. Objection handling appears in 35 of the 208 cold prospecting entries, which is 17%, and in 15 of the 37 warm consideration ones, which is 41%. Identity appeal runs the other way: 47% cold against 24% warm.

The bottom two rows carry counts rather than percentages on purpose. Six entries and two entries cannot support a rate, and printing 17% beside n=6 would hand the thinnest row in the table the same authority as the thickest.

Two readings of the warm row are available and only one of them is a finding. Either warm assets really are doing a different job, or warm assets are simply harder to catch in an ad library and the 37 that made it in are not representative. I lean to the first, because the direction is the one the mechanism predicts. Thirty seven entries are not enough to settle it.

The pile changes shape
by category

Which belief gets the work depends heavily on what you sell. Home and household entries run mechanism reveal in 29% of cases and objection handling in 26%, the highest of any category with enough entries to support a rate. Apparel, the largest category in the corpus at 70 entries, runs mechanism in 9%.

There is a readable logic in the mechanism column. Where the product does something physical a buyer cannot see happening, a cleaning tablet dissolving or a shoe holding a shape, the brand explains the mechanism. Products judged on taste or fit get no such explanation, because there is nothing to explain. Food and beverage sits at 4% for exactly that reason.

The footwear row is the one I would not build a plan on. Zero objection handling across 13 entries is either a real category habit or a thirteen entry accident, and nothing in the data separates those two. Read it as a prompt to go and look at your own category, not as a finding about footwear.

Risk reversal never breaks two entries in any category. Diapers and trainers have almost nothing in common as purchases. Neither spends its paid impressions on what happens if I am wrong. Every category in the table leaves that belief for the buyer to resolve alone.

Which beliefs does a buyer
have to hold before they buy?

Four, in most consumer categories, and they are not interchangeable. Google's messy middle work isolated six behavioural biases that shift choice inside the loop, social proof and authority bias among them. Strip those back to what a buyer is actually asking and you land on four questions.

1. Is this problem real, and is it mine? Proved by a specific description of the situation, not of the product. The asset best placed to do it is the one with native reach: paid social, a creator, an organic post. This is the only belief a cold ad is well positioned to move, and 110 of 253 entries are working on it.
2. Does this thing actually work? Proved by mechanism. What does it physically do that produces the result. The product page owns this, and so does a comparison page. Mechanism reveal runs in 44 entries, which makes it the fourth most common mechanic and still leaves 209 entries skipping it.
3. Does it work for someone like me? Proved by distribution, never by an average. A 4.8 star average answers nothing. The review histogram, the filter by skin type or foot width or use case, the photo from somebody with the same problem: those answer it. This belief cannot be proved by the brand in its own voice, which is exactly why it belongs to reviews and customer content.
4. What happens if I am wrong? Proved by the guarantee, the returns policy and the shipping terms, stated near the buy button rather than buried in the footer. Seven entries out of 253 carry risk reversal, which makes this the thinnest belief in the corpus by a distance. Where on the page it goes is its own question, covered in the ten-element product page audit.

Assign each belief to exactly one owner. A belief with two owners gets argued twice and proved once, because neither asset commits. A belief with no owner is the gap your buyer walks into, and it is why a pile that looks full still does not close.

One brand, three assets,
three different jobs

Coterie is the clearest example of the belief map running in production that the corpus holds. Three of its entries were captured on 19 August 2026, all three rated Continuous, and each one is built to prove a different belief. None of them repeats another's argument.

Belief four, run warm. A hand-drawn flowchart with the objection as the headline: are Coterie diapers worth the price? It branches through have I ever thrown out a onesie after a blowout, and leaves the sceptic on a flat two word terminal node, you will. There is no price on the tile, no discount and no product claim. It never argues the diaper is cheap. It swaps the currency from dollars to a ruined onesie and a 3am sheet change, and in that currency a premium diaper wins without a single spec. Four different wrappers ran the identical logic between 19 May and 11 August 2026, so format was the only variable being tested.

Belief two, run cold. A creative styled as an order summary for the Newborn Starter Kit: the diaper at $39.98, trial packs at $9.00, three wipes at $26.22, a skincare set at $40.00, a discount of minus $5.00, and a total struck from $115.00 to $110.00. The five dollar discount is not the point and was never meant to be. The $115.00 anchor is. The reader reaches the conclusion by adding up, which is a great deal harder to argue with than an adjective.

Belief three, run cold. A credential stack that borrows other people's liability rather than making a claim in its own voice. The number one pediatrician-recommended TCF diaper, footnoted on the tile to the IQVIA ProVoice Pediatrician Survey of August 2025. Beside it, EWG Verified, an OEKO-TEX Standard 100 certification and a Best of the BUMP Awards 2025 roundel. A parent cannot test absorbency before buying. Every seal is a promise somebody else is on the hook for.

Three assets, three beliefs, one brand, and the same marks reappear on the product page after each click, which is why all three rate Continuous. That last part is the rarer achievement. Across the 226 rated pairs, only 54 keep the argument intact from ad to page, and what those clean handoffs have in common turns out to be a short list.

I have no idea whether Coterie planned it this way. What is visible from the outside is that three assets are doing three different jobs and none of them wastes an impression restating another's. Whether that came from a belief map or from good instinct, the output is the same and it is the output you can copy.

How do you audit your
store's evidence pile?

From the buyer's side, in one sitting, with the belief map open. Of the 226 rated ad and page pairs in the corpus, 172 lose part of the argument somewhere between the click and the page. That is the same failure one level down: the surfaces are not being read in the order a buyer meets them.

Run this for one product before you run it for a catalogue. Roughly ninety minutes. You are not auditing what each surface says. You are auditing which of the four beliefs each surface proves, and which belief nothing proves.

1. Write the four beliefs down in the buyer's words rather than your own. If you cannot write the third one without naming a feature, you do not yet know what your buyer is afraid of, and everything after this step will be guesswork dressed as a plan.
2. Open your own Meta Ad Library page and read every live creative in one pass. Against each, write the belief it proves. Assets that prove nothing and only interrupt are fine in small numbers. Count them anyway, because the count is usually higher than anyone expects.
3. Click your own ads. Not the campaign URL in the ads manager. The actual link a person taps, on a phone, cold. Note where it lands and whether the argument the ad started is still on the page. Five destination shapes cover almost every paid click in the corpus, and which shape you land on decides how much of the argument survives. This is the cheapest check on the list and almost nobody runs it.
4. Read the product page top to bottom on a phone, at the width a real visitor uses, with the reviews collapsed the way they ship. Which belief does the first screen prove? In most stores nothing on it proves any of the four, because the first screen is a hero image and a price.
5. Open the review distribution, not the average. Find the most common complaint in the two star reviews. That complaint is belief three, unproved, written in your buyer's own words. It is free research and it is already sitting on your own domain.
6. Search your own brand name and read the first page as a stranger would. The brand SERP is the last surface before purchase for a large share of buyers, and most of it is assembled by other people. Whatever it proves is where your pile ends.
7. Read your welcome email and your abandoned cart email against the same four beliefs. Most of them repeat a discount and prove nothing. Those two slots are the best placed in the whole stack to carry belief four, because the buyer has already told you they are hesitating. Most of them are still running a bare percentage, which is what the welcome offer capture found across the brands it sampled.
8. Fill each gap with the surface that can actually prove it, then stop. One new asset per unproved belief, and no campaign around it. The point of the audit is to spend less, on fewer things, aimed at named gaps.

Two of those steps have longer treatments already. The click-through check is the entire subject of what breaks between an ad and its landing page, and the product page pass has a screen by screen version in the product page element grading.

Why does the last click get
credit it did not earn?

Because a brand search is a retrieval event, not a decision event. Somebody types your name because you are already in their consideration set. Of the 253 teardowns here, 208 are cold prospecting assets, and that is precisely the work a last-click model cannot see. The model credits recall and ignores whatever built it.

Measure anyway. Just stop reading one number as a verdict on a whole pile. The major tools disagree with each other for structural reasons, and none of them measures causation, which is the subject of the attribution tool comparison. The same blind spot is what makes a prospecting channel look like a loser next to a branded one, which is the Meta against Google trap in one line. A holdout does measure it, and how to run one is a separate piece of work.

The practical move is to stop asking which touchpoint got the click and start asking which belief was the last one to land. No dashboard has that. Five post-purchase survey responses do, if the question is what almost stopped you from buying. Ask it for a month and the same missing belief will come back in five different phrasings.

What to change first

Pick one product and one unproved belief. In this corpus the thinnest belief is the fourth: 7 entries of 253 carry risk reversal and 51 handle an objection of any kind. If you have no idea where to start, start there, on the grounds that almost nobody else is.

1. This week. Click your own live ads on a phone and write down, for each, which belief the landing page continues. Stop when you have done ten.
2. Next week. Put the guarantee and the returns terms above the fold on one product page, in plain language, and leave everything else alone so you can read the result.
3. The week after. Add the post-purchase question. What almost stopped you from buying. One field, no incentive.

None of that is a campaign, and that is the point. The stage map keeps producing more of the asset you already have too many of, because a stage is a schedule and a schedule wants filling. A belief map produces one asset and then goes quiet, which feels like underworking and is usually the cheaper answer.

If you would rather not run the pass yourself, it is most of what the first week of The Read covers: the same four beliefs, read across your ads, your pages, your reviews and your lifecycle email, with the gap written down rather than guessed at.

Questions operators ask
once the map stops
matching the buyer.