Commerce Without Checkouts: Bryan House on AI, Composable Commerce, and Why Digital Commerce Is Becoming Intelligent

Bryan House, CEO of composable commerce platform Elastic Path, made one claim worth tracking for retail media operators: if high-intent buyers start product discovery inside ChatGPT instead of a retailer's search bar, sponsored-listing inventory loses its foundation. Sponsored product ads and onsite display are the profit engine of Amazon Ads, Walmart Connect, and Kroger Precision Marketing. They only pay off if people browse first.

The mechanism House describes is real. The timeline he implies is not supported by the evidence in his own episode.

What the numbers say

House concedes Amazon Ads kept growing quarter over quarter. Google search revenue is holding. If LLM discovery were already draining retail media, the largest retail media network on earth would show it first, and it hasn't. The revenue base that would have to crack has not cracked.

His headline stat, that 30% of LLM answers would not appear on Google's first page, is unsourced and, by his own admission, self-quoted from a prior podcast. His two agentic-checkout examples, OpenAI's Instant Checkout and Perplexity's Buy Now, both launched and were pulled within months. He blames a bad transaction layer, not weak demand. That is convenient for a man who sells the transaction layer. Read the incentive before you buy the forecast.

What operators should do anyway

You don't have to believe the thesis to do the work. Two moves are cheap and pay off regardless.

First, structure your product catalog so an LLM can read it: full specs, technical attributes, FAQs, everything SEO pages strip out. That data-hygiene project pays off whether House is right or not, because Google's own AI-generated answers eat the same feed.

Second, instrument LLM referral traffic now. If higher-intent buyers arrive pre-decided from an AI answer, last-click attribution and A/B testing start measuring noise. You won't notice until the numbers stop making sense.

Where the real opportunity sits

The agentic-checkout drama is a B2C story that isn't ready yet. House pegs global B2B commerce near $36 trillion (treat as order-of-magnitude) and calls routine reorders the actual near-term agentic use case. That is the customer already raising a hand.

The call

No major retail media network among Amazon Ads, Walmart Connect, or Kroger Precision Marketing will report a year-over-year decline in sponsored-product ad revenue attributable to LLM discovery through the Q4 2026 earnings cycle. Confidence: medium. The erosion mechanism is real but slow, and the base is still growing.

The more interesting move is what smart networks do before revenue dips. Separate your retail media exposure into browse-dependent inventory (sponsored listings, onsite display) versus data-and-measurement products (closed-loop attribution, purchase-signal targeting). Browse-dependent inventory is exposed to the shift. Purchase data and measurement get more valuable as discovery fragments, because when you can't see the shopper browse, the retailer's record of what they actually bought is the only ground truth left. The networks that see this coming will start pricing those two halves differently before the headline numbers move.

Your draft

Bryan House, CEO of composable commerce platform Elastic Path, sat down with Signal & Noise hosts Brett House and Rio Longacre and made a claim that should keep every retail media operator awake: if high-intent buyers start their shopping inside ChatGPT instead of a retailer's search bar, the sponsored-listing business that funds retail media networks loses its oxygen. Everything else in the episode is commerce plumbing. This one thread is the ad-tech story.

What's being decided (briefing frame): whether the shift of product discovery from search boxes and browse pages into LLM answers erodes the value of retail media inventory. Sponsored product ads and onsite display are the profit engine of Amazon Ads, Walmart Connect, Kroger Precision Marketing, and the rest. They only pay off if people browse. House is asking what happens when they stop browsing and just ask.

Reversibility: Type 1 for anyone rebuilding a catalog data strategy around it. Type 2 for watching the signal. Forcing function: none hard. This is a multi-year drift, not a quarter-end event, and House himself hedges on timing.

---

The Market Analyst. Watch what the money did, not what the panel fears. House himself notes Amazon Ads kept growing quarter over quarter. That is the evidence. If LLM discovery were actually draining retail media, the biggest retail media network on earth would show it first, and it hasn't. Google search revenue is also holding. The revenue base that would have to crack has not cracked. For an operator, the trade is not "sell retail media." It is "retail media that depends on browse impressions is more fragile than retail media that sells access to purchase data and closed-loop measurement." Those are different products inside the same network, and the market will start pricing them differently.

To a non-specialist: the ad business built on people scrolling shelves is shakier than the one built on knowing what people actually bought.

The Skeptic. Steelman House and the case gets thin fast. His headline stat, that 30% of LLM answers would not appear on Google's first page, is unsourced, and he admits he is quoting himself from a prior podcast. His OpenAI and Perplexity checkout examples cut against his own thesis: both companies launched autonomous checkout and pulled it inside months. He attributes that to a bad transaction layer, not weak demand, which is convenient for a man selling the transaction layer. And House sells composable commerce infrastructure. A world where LLM discovery upends everything is a world where his replatforming pitch lands. Read the incentive before you buy the forecast.

Plain version: the guy predicting a revolution happens to sell the shovels.

The Operator. Forget the thesis. What do I do Tuesday? Two real jobs fall out of this, and they are cheap. First, structure the product catalog so an LLM can read it: full specs, technical attributes, FAQs, the stuff SEO pages strip out for humans. That is a data-hygiene project, not a moonshot, and it pays off whether or not House is right, because Google's own AI answers eat the same feed. Second, instrument LLM referral traffic now so you can see the curve before it matters. What breaks at 90 days is measurement. House flags it and the summary cuts off mid-sentence, but the point stands: if higher-intent buyers arrive pre-decided from an LLM, your A/B testing and last-click attribution start measuring noise. You will not notice until the numbers stop making sense.

Plain version: fix your product data feed and start counting how many buyers arrive from AI, before it's a fire drill.

The Customer / End User. Two customers here, and they diverge. The advertiser buying sponsored listings wants browse volume, and House says that volume is the thing at risk. The retailer running the network wants the closed-loop data, which survives the shift because purchase still happens somewhere. Nobody is asking for autonomous checkout yet, and the OpenAI single-item-basket failure proves consumers were never the blocker: the product just wasn't good. The B2B buyer is the one genuinely underserved. House pegs global B2B commerce near $36 trillion (unsourced, treat as order-of-magnitude) and calls routine reorders the real near-term agentic use case. That is the customer actually raising a hand.

---

The tensions:

1. The Analyst versus House on timing. House says the discovery shift is happening now. The Analyst points at Amazon Ads still growing and says the revenue that would prove it hasn't moved. Both can't be right about "now."

2. The Skeptic versus the Operator on whether to act. The Skeptic says the thesis is a vendor talking his book. The Operator says the two prep moves are cheap and pay off regardless. This is where the decision lives: you don't have to believe House to do the work.

3. B2C hype versus B2B reality. The whole agentic-checkout drama is B2C theater. The actual near-term money, per House, is boring B2B reorders. The stakeholder chasing the headline and the stakeholder with the real opportunity are looking in opposite directions.

---

Synthesis. This hinges on one belief: does product discovery move off browse surfaces fast enough to starve sponsored-listing inventory before networks adapt? The council leans skeptical on speed and agnostic on direction. The mechanism House describes is real, the timeline he implies is not supported by the one hard number in the episode, which is Amazon Ads still climbing. What to de-risk: separate your retail media exposure into browse-dependent inventory (sponsored listings, onsite display) versus data-and-measurement products (closed-loop attribution, purchase-signal targeting). The first is exposed to the shift. The second gets more valuable as discovery fragments, because when you can't see the shopper browse, the retailer's purchase data is the only ground truth left. Don't sell retail media. Reprice the two halves of it differently.

---

Prediction: No major retail media network among Amazon Ads, Walmart Connect, or Kroger Precision Marketing will report a year-over-year decline in sponsored-product ad revenue attributable to LLM discovery through the Q4 2026 earnings cycle in early 2027.

Confidence: Medium. The erosion mechanism is real but slow, and the base is still growing today.

Why: House's own episode contains the counterweight to his thesis: he concedes Amazon Ads is still growing quarter over quarter, and Google search revenue is holding. LLM shopping is real but tiny, and the two most-cited autonomous-checkout experiments, OpenAI's Instant Checkout and Perplexity's Buy Now, both launched and were pulled inside months, which means the AI shopping funnel that would divert browse traffic barely functions yet. For a network's sponsored-listing revenue to actually fall year over year, LLM discovery would have to scale from marginal to material in under two quarters, and nothing in the transaction layer is ready for that. The opposite outcome, continued growth, is what the revenue base is already doing.

Revisit by 2027-03-15: We're right if none of Amazon Ads, Walmart Connect, or Kroger Precision Marketing reports a year-over-year sponsored-product ad revenue decline blamed on AI-driven discovery in their Q4 2026 results. We're wrong if any of the three reports such a decline and names LLM or AI search as a cause.

The interesting move, though, is not the headline number. It's that the smart networks will start selling the purchase-data-and-measurement half of their business harder while it's still valuable, precisely because they can see the browse half getting shakier. That repricing starts before the revenue ever dips.