# Agentic commerce risks retailers should be thinking about now

**Published:** September 24, 2026 | **Updated:** September 24, 2026 | **Authors:** Gladly Team | **Categories:** Trends and expert opinions

> Agentic commerce risks for retailers go beyond checkout. See how AI shopping agents squeeze margins and loosen brand control, and what lean teams can do.

---

Most coverage of agentic commerce risks for retailers has focused on the shopper. Will they trust an AI agent with their card, and what happens when the agent buys the wrong thing? Those are real questions, and we've covered them in [The trust layer](/blog/trust-layer-ai-agents-commerce/) and in [Agentic checkout, explained for ecommerce teams](/blog/agentic-checkout-explained/).

This post is about the risks that land on your side of the counter. When an AI agent does the comparison shopping, it changes what your margin depends on, and it changes who gets to describe your products and your prices before a shopper ever reaches your site. Both show up slowly, as a thinner gross margin a few months out and a brand that reads a little flatter in every AI answer.

If you run a lean ecommerce team, you'll feel both earlier than a large retailer will. You don't have the supplier buying power to absorb a price war, and you probably don't have a media budget that buys visibility back when an agent leaves you off its shortlist.

#### Margin and price-comparison pressure from agent shopping

A person comparing two pairs of hiking boots opens a few tabs, reads a couple of reviews, looks at the photos, and makes a call that is part price and part feel. An agent runs the same comparison across far more sellers in a few seconds, and it builds its shortlist from whatever it can line up side by side. Price, delivery date, return window, and stock are easy to line up. Your brand story, your photography, and the care that went into your product copy are much harder to put in a comparison table.

McKinsey's January 2026 [analysis of agentic commerce](https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-automation-curve-in-agentic-commerce) says it plainly about everyday, low-regret categories like groceries and household essentials: "being agent-readable and dependable matters more than being distinctive." That logic is spreading well past the grocery aisle, and for a lot of catalogs it's a margin problem. Once a product is reduced to the fields an agent can compare, the cheapest dependable option wins the recommendation, and everyone else either matches it or drops out of consideration.

##### Your promotions travel further than you meant them to

The subtler version of this is about discounts. A first-order code, a code you gave an influencer, or a sale price meant for one marketplace used to reach the shoppers it was aimed at plus a handful of coupon-site regulars. An agent tasked with finding "the best price" goes looking for all of them, and it reports the lowest effective price it can find as your price.

So a promotion you designed for acquisition quietly becomes the price every agent-assisted shopper expects to pay. [Agentic checkout, explained for ecommerce teams](/blog/agentic-checkout-explained/) covers how promo logic behaves inside an agent-built cart. The margin risk starts earlier than that, when the agent is still deciding whether you're the best deal at all.

##### Why lean retailers feel it first

Our post on [how AI shopping agents are changing the ecommerce funnel](/blog/ai-shopping-agents-ecommerce-funnel/) touched on this briefly: instant comparison hits hardest when a retailer's only visible point of difference is price. A national retailer can negotiate better costs and run loss leaders across a catalog big enough to absorb them. A 20-person DTC brand selling 300 SKUs usually can't do any of that, and a price match on its best seller can wipe out the month.

Resellers make it worse. If you sell through wholesale partners or marketplaces, an agent can find a third-party seller listing your product below your own price and send the shopper there. You still make the wholesale margin, but you lose the direct customer and whatever you would have earned from their next order.

##### Where margin can still hold

The same McKinsey analysis offers a way out: "Margins are shaped by service guarantees, fulfillment reliability, and clarity of policies." Agents compare those things too, as long as they can read them.

That means your free-returns window, your delivery promise, your warranty, and your fit guarantee need to be as easy for an agent to extract as your price is. If a competitor is $6 cheaper but takes nine days to ship and charges for returns, an agent that can see all three facts has a reason to recommend you. If it can only see the price, it doesn't. [Preparing your product feed for AI agents](/blog/product-feed-ai-agents/) walks through getting that data into a form agents can parse.

Bundles and kits help for the same reason. A starter kit with a specific set of items has no identical listing elsewhere, so the agent has to judge it on what's in the box, which works in your favor if the kit is a good one.

**How trust breaks when AI agents shop**

The consumer side of agentic commerce risk, including the five failure modes HBR identified and what brands need to build first.

[Read the post]

#### Brand and pricing control risks across the shopping journey

Checkout gets most of the planning because it's where money moves, but control starts slipping well before it. By the time a shopper's agent reaches your checkout, it has already described your product, compared it against competitors, and summarized what other buyers think, and you had no say in any of those steps.

##### At discovery, the agent describes you before you say a word

A shopper asks an AI assistant for a rain jacket that packs small and costs under $150. The answer they get might describe your jacket with last season's price, a colorway you discontinued, or a return policy you changed in the spring. Nothing about that is malicious. The agent pulled what it could find, and some of what it found was stale.

Shoppers notice when it goes wrong. In a SmartCustomer survey of nearly 1,200 U.S. consumers [reported by Retail Dive](https://www.retaildive.com/news/shoppers-burned-bad-ai-purchase-recommendations/831013/) in September 2026, a third said they'd made a purchase they regretted based on an AI recommendation, and apparel and footwear made up 45% of those bad purchases. A shopper who gets a wrong price from an agent usually takes it up with the store that won't honor it.

##### At comparison, your product sits in a table you didn't design

When an agent compares you with two competitors, it picks the attributes, the order, and the summary line. It might call your product "the premium option" or "similar to a cheaper alternative," and it might summarize your reviews by leading with the three complaints about sizing out of 400 happy customers.

Pricing control gets tested here too. If you use a minimum advertised price policy with your retail partners, an agent reading every listing it can reach will surface the partner who ignored it, and that price becomes your product's price in the answer. The same goes for price inconsistencies between your site, your app, and your marketplace storefront. An agent finds the gap faster than your team does.

##### After the purchase, the relationship is up for grabs

McKinsey has a line for this too: "Loyalty becomes less about sentiment and more about policy." When an agent picks where to reorder based on price, speed, and return terms, the goodwill you built with a shopper counts for less unless it shows up in something the agent can compare.

The bigger loss is the relationship data. If the purchase happened in a third-party assistant, you may never learn why the shopper chose you, what they almost bought instead, or what they'd want next. That context is what you'd normally use to bring them back, and without it, every reorder is a fresh competition.

**Where liability lands at agentic checkout**

Payment delegation, authorization, and who handles a dispute when an agent's purchase goes wrong.

[Read the post]

#### A risk-aware approach to early agentic commerce adoption

Waiting this out isn't much of an option, because agents are already reading your product pages and quoting your prices whether you've made any decisions about them or not. What you can do is figure out where you're most exposed and decide a few things now, so you aren't writing policy in the middle of a margin problem.

> **A quick exposure check for lean teams:** - Which of your top 20 sellers can a shopper buy somewhere else, either the identical item or something close enough?
> - Is your price the same on your site, your app, and every marketplace listing today?
> - Which active promo codes would you be unhappy to see an agent quote as your regular price?
> - Can an agent read your return window, shipping speed, and warranty as easily as it reads your price?
> - When did someone on your team last ask an AI assistant about your best seller and check the answer?

##### Sort your catalog by how easy it is to compare

Some of your products are commodities to an agent: the same item, or a close substitute, is sold by dozens of stores. Others are hard to compare because they're exclusive, bundled, or made to spec. The first group is where margin pressure lands, so compete on the dependability facts covered above, like shipping speed, returns, and stock accuracy, and make sure those facts are machine-readable. The second group is where brand-control risk lands, so put your effort into accurate, detailed product content and reviews an agent will cite correctly.

Most lean teams can do this sort in an afternoon with a spreadsheet of their top 20 sellers.

##### Decide where your price lives

Pick one source of truth for price and make every channel match it, including marketplaces and partner feeds you don't manage day to day. Then make a deliberate call on which promotions should be public at all. A code that only makes sense for a specific audience belongs behind a login, an email, or a loyalty account, where an agent looking for the lowest public price can't treat it as your list price.

If you have wholesale partners, check whether your agreements say anything about price parity on AI-readable listings. Most were written before anyone had to think about it.

##### Keep the conversations you can own

Third-party assistants will keep sending you shoppers, and some of that traffic converts well. The customer relationship still lives where you can see the full history: your own site and support conversations, and whatever happens after the order ships. The more a shopper gets from talking to you directly, like fit advice based on their last return or a heads-up that the item they reorder is back in stock, the less every reorder has to be decided on price alone.

That's the job of the AI on your own site. Gladly's AI sees each shopper's order history and past conversations, so it can suggest sizing up after last month's return, answer an order question in the same chat, and hand off to your team without the shopper starting over. A third-party assistant can't do any of that for you.

**Keep the relationship in your own conversation**

In a live demo, see how Gladly's AI uses each shopper's history to recommend the right product, handle order and return questions, and bring in your team with full context.

[Schedule a demo]

##### Watch a few numbers, starting now

You don't need a new dashboard. Track gross margin on orders that arrive from AI referral sources separately from the rest, note how often those orders carry a discount code, and once a month, ask two or three AI assistants about your top five products and write down what they get wrong. For the fuller set of data, checkout, and measurement work, [An AI agent readiness checklist for ecommerce teams](/blog/ai-agent-readiness-checklist-ecommerce/) lays out where to start.

Each of these steps comes down to someone on your team deciding what your price is, which of your promises are worth publishing, and which products you'd rather compete on something other than price. Make those calls before an agent makes them for you.

---

*This content is provided by Gladly. Visit [gladly.com](https://gladly.com) for more information.*