This article is based on a Q&A with Paxon Senior Director, Product Management Robin Gomez and Senior Director, Head of Marketing Brendan Lease and Greg Kihlström for The Agile Brand Guide’s Expert Mode series.
For about twenty years, ecommerce marketing has had a fixed address. It’s the product page. We tuned the photography and rewrote the copy. Whole attribution models were built on the assumption that a person would land there before buying anything.
That address is getting less reliable. Meta’s Muse agent can now find products in Shopify stores and complete a Shop Pay checkout without the shopper ever loading the page, and Shopify merchants are enrolled by default. Robin Gomez, Senior Director of Product Management at Paxon (the logistics company formerly known in North America as Radial), and Brendan Lease, the company’s Senior Director and Head of Marketing, argue that brand still matters inside that flow. But the deciding factors in agentic commerce look a lot more operational than most marketing teams are used to: whether the agent can read your data, and whether you actually deliver what it promised on your behalf.
Brand Preference Picks the Starting Line, Not the Winner
We asked Gomez to argue both sides. Either agents fundamentally change brand strategy, or they’re one more interface and preference works the way it always has. He accepted both and then picked one.
I believe both arguments are true, but the first one is more consequential. Brand preference will continue to matter, especially in categories tied to identity, lifestyle, or self-expression. What changes is that preference may determine where the agent starts, while structured data and execution determine whether the brand ultimately wins. Marketing will not become irrelevant, but it will have to expand beyond influencing people to making the brand understandable and dependable to machines/agents.
– Robin Gomez
Lease’s version is more tactical. He thinks the new work is figuring out what each agent actually weighs. Is it reviews, the description, or the returns terms? Can the full cost of purchase, delivery fees included, even be scraped? “Learning how the different agent interacts and selects product for recommendation is what truly changes here,” he said. He’s also clear that the old fundamentals don’t go away. Strong descriptions and solid reviews still matter “regardless of the interface.”
For a CMO, the uncomfortable part is where those inputs live. Returns policy usually belongs to operations, and delivery fees to finance or logistics. If an agent is weighing those things as heavily as it weighs your copy, then the product detail page is no longer something marketing owns outright. One concrete starting point: ask whether your landed cost, shipping included, appears anywhere in a form a machine can read. For a lot of brands it only shows up in the cart.
Default-On Means the Decision Was Made for You
Shopify merchants are in Muse unless someone switches the channel off. So we asked Gomez what has to be true before staying in is a good bet, and what opting out really costs.
First, product data must be complete, accurate, and machine-readable. An agent can only recommend what it understands.
Second, your operational promises need to match reality. Inventory, pricing, delivery dates, returns, and cancellations must be consistent across every system.
Third, you need a clear exception process. Customers and agents need to know what happens when an item is delayed, cancelled, returned, or requires human intervention.
The cost of opting out is reduced visibility. As agents become discovery engines, brands that aren’t participating risk being excluded from consideration long before shoppers visit a website.
– Robin Gomez
Not one of those three is a marketing deliverable in the traditional sense. Inventory accuracy lives in the order management system, and exception handling usually sits with customer care. Marketing doesn’t own any of it, but it owns the result when an agent tells a shopper something the business can’t back up.
Here’s a test that takes an afternoon. Pull your ten best-selling SKUs and compare three things: the delivery date on the product page, the date in the product feed, and the date the warehouse actually hit last month. If those don’t match, the agent will repeat whichever one it found, and it probably won’t be the conservative one.
Opting out can be a reasonable call for a brand whose data isn’t ready. It just shouldn’t happen by accident, which is how most defaults get decided.
Don’t Answer Agent Volume With More Agents
The consumer data behind much of this conversation comes from two surveys Radial commissioned and fielded with US shoppers last December and January, before Muse existed. Paxon also sells the fulfillment, payments and fraud services those findings point toward, so read the numbers with that in mind. We asked Lease directly where vendors, his own company included, might be over-reading consumer caution. He didn’t think they were. In the research, 64% of consumers said they worry about giving payment information to AI agents, and 52% said the same about sharing personally identifiable information. Lease’s spending advice follows from there.
One, consumers want to be able to approve every transaction, and two, they want to have an easy path to fixing any potential mistakes that are made by the agent acting on their behalf. … The research indicates that CMOs should not spend a lot of money on Agentic Customer Service as buyers are very consistent in indicating that if something goes wrong with an agentic-driven order they want to talk to a human to resolve it.
– Brendan Lease
That’s a little counterintuitive. The reflex when a new automated channel shows up is to automate the support behind it too. Lease is arguing that the path to a person is part of what makes an agent purchase acceptable in the first place. Shoppers in the same research put human support near the top of their must-haves, next to refund guarantees for mistakes. In budget terms: if anyone’s planning to trim holiday contact center staffing on the theory that agents will deflect volume, the human-in-the-loop escalation path is the wrong place to find savings.
As for what’s moved since the surveys closed, Gomez’s answer is short. “What’s changed is the infrastructure,” he said, citing Muse’s Shopify integration and Google’s Universal Commerce Protocol. Consumer expectations haven’t moved. He credits the Muse checkout for being built around customer approval rather than opaque browser automation, which is a fair standard to hold any agent partner to.
Measure Whether the Promise Was Kept, Not Just Whether the Order Came In
We asked both guests what to finish before holiday peak, and what one number belongs on the weekly dashboard. Their answers differ, and both are worth taking.
Lease’s is reviews. Great reviews were the top reason shoppers in the survey gave for trying an AI-recommended brand, cited by 43%. He’d track reviews as a percentage of purchases, which is cheap to set up.
Gomez went further.
Before peak, every CMO should complete an agent-readiness audit. Validate that your top products, delivery promises, pricing, return policies, and inventory data can be accurately understood and represented by an AI agent. … The one metric I’d add is Agent Order Success Rate, which is the percentage of agent-driven orders that are fulfilled as promised without preventable cancellations, payment issues, inventory exceptions, or service failures. … Revenue tells you agents are generating demand. Success rate tells you whether your organization can deliver on what the agent promised.
– Robin Gomez
His metric is the more interesting of the two, partly because few dashboards can produce it today. In practice, it means tagging agent-originated orders when they come in, then tracking cancellations and payment failures and inventory exceptions against that tag specifically. It’s close kin to on-time in-full (OTIF), which supply chain teams already report, just cut by channel. And if the agent channel comes in worse than your own site, the agent is overpromising on your behalf. The shopper won’t blame Meta for that. They’ll blame you.
The agent readiness audit comes first. Without it, the success rate mostly measures how often bad data turns into a bad order.
The product page isn’t dead, and both guests said the fundamentals still hold. What’s changed is that a growing share of the persuasion now happens in a comparison the brand never sees, built from delivery windows and return terms and whatever the feed says about stock. So before November, pick your top ten SKUs, check that every system tells the same story about them, and settle with operations who owns the agent success number. It’s less glamorous than the holiday campaign. It’s also the part an agent will actually check.



