What an AI Agent Does for a Fashion Shopify Store

An AI agent for a fashion Shopify store spends most of its time on one question, asked a hundred different ways: will this fit me. Everything else, shipping dates, restock timing, whether the black one runs small, is a variation on the same theme. Get that question answered well before checkout and you sell more. Get it answered badly, or not at all, and you inherit the return.

That second half is the part general ecommerce advice skips. A generic support bot tries to answer questions faster. A fashion store needs the specific question answered correctly, because a wrong answer here becomes a return three weeks later.

Fashion returns are a different scale of problem

Every ecommerce category deals with returns. Fashion deals with a lot more of them. Online retail overall runs at roughly a 19% return rate, per the National Retail Federation's 2025 Retail Returns Landscape, but apparel sits well above that average, with multiple industry benchmarks putting clothing return rates anywhere from 20% to 40% depending on category and price point. Footwear and fast fashion tend toward the higher end; considered, mid-price apparel toward the lower.

A shopper buying a sweater can't touch the fabric, and a shopper buying jeans can't try them on, which is the actual reason, not quality control. Baymard Institute's usability testing on apparel sites found that 90% fail to let shoppers properly assess a product's appearance, size, or fit before buying, and separately that 83% of desktop apparel sites and 87% of mobile apparel sites don't provide sufficient sizing information to make a confident decision. That's the majority of the category, on every device, still asking shoppers to guess.

Guessing has a name now: bracketing. A shopper orders a medium and a large, plans to keep whichever fits, and returns the other. Estimates of how common this is vary by study and year, but multiple sources now put it at roughly six in ten online apparel shoppers doing this deliberately, up sharply from a decade ago. Some of that volume is unavoidable: people genuinely don't know their size across brands, and a bracket order is a rational response to that uncertainty. But a meaningful slice of it is a shopper who would have ordered one size if someone had actually answered their sizing question first.

The questions a fashion store's chat actually gets

Pull your own chat and email history before an order exists and the pattern holds across almost every apparel brand:

Between-sizes questions. "I'm a 10 in Everlane, a 12 in Madewell, what does that make me in your cut?" This is the single most common pre-sale message in apparel, and it's the one static content answers worst, because the right answer depends on the shopper's specific measurements against a specific garment's cut, not a generic chart.
Fabric behavior. "Does this stretch?" "Will this shrink in the wash?" "Is this see-through?" These aren't in most size charts at all; they live in the product description, if they're written down anywhere.
Fit across the line. "I bought your relaxed-fit tee in a medium, will the same size work in the fitted style?" A shopper loyal to your brand still has to relearn fit for every silhouette you sell.
Color and stock. "Is the sage green true to the photo?" "When is this restocked in petite?" Real-time questions a static page answers with whatever inventory count it had at the last sync.
Layering and occasion. "Will this coat fit over a blazer?" "Is this warm enough for a Chicago winter?" Judgment questions that require connecting two pieces of product information, not reciting either one.
Return risk. "What if it doesn't fit right?" Asked before the order exists, specifically to lower the risk of placing it.

Every one of these is answerable, in principle, from information the store already has: a size chart, a product description, a return policy, current inventory. The problem was never that the answer didn't exist. It's that the answer lived on a page the shopper didn't read, in a format that couldn't handle "I'm 5'4 and pear-shaped, will this work," which is how the question actually arrives.

Bracketing is the return you can partly prevent

Not all fashion returns are fixable by better answers. A shopper who dislikes how a color looks in person, or who simply changes their mind, is going to return the item regardless of how the pre-sale conversation went. That's a normal cost of selling apparel online, and no chat agent removes it.

But a real slice of bracketing is a proxy for an unanswered sizing question, and that slice does respond to a specific, grounded answer at the moment the shopper is deciding. "Based on our size chart, if you're a 10 in Everlane, our runs about half a size smaller, so you'd likely want a 12" is a different experience than a shopper ordering both a 10 and a 12 to find out. It won't convert every bracket order into a single-size order. It will convert some, and every one it converts is an order that never becomes an inbound return, a restocking cost, and a size the warehouse has to grade and shelve again.

The honest claim here is modest, and it's worth stating plainly: an agent grounded in your actual size chart and product content can answer the specific version of a sizing question a static chart can't. What it can't do is predict fit the way a dedicated sizing tool does. Apps built specifically for that job, like size-recommendation widgets that ask for a shopper's measurements and run them through a fit-prediction model, are a different category solving a narrower, more specialized problem. A chat agent grounded in your content and a purpose-built fit engine can coexist; they're not competing for the same job.

What happens after the order ships anyway

Some returns happen no matter what. When they do, the post-purchase conversation looks like this:

"This runs small, can I exchange for a size up?"
"The stitching came loose after one wash, is this covered?"
"I already know it doesn't fit, how do I send it back?"

This is support-agent territory, not sales-agent territory, and it's worth treating as its own workload rather than an extension of the pre-sale conversation. The distinction that matters operationally is between a fit-based return (routine, covered by policy, usually fine to move through quickly) and a defect claim (needs a different path, sometimes a photo, sometimes escalation). A store running high apparel return volume benefits from an agent that can look up the actual order, check it against the return policy, and start the return in the same conversation, rather than routing every "this doesn't fit" message into a queue a human has to open, read, and manually process. We've written more on what that automation should and shouldn't do on its own in our guide to automating a returns process; the short version is that returns can run on autopilot, but refunds should always wait for a human's approval.

What to check before you trust an agent with this

Fashion is a worse place than most verticals to have an agent that answers confidently and wrong. A wrong "yes, that runs true to size" doesn't just cost the return, it costs the shopper's trust in every other answer the agent gave. Before putting one in front of real customers, check:

It answers from your actual size chart, not a general impression of your category. Ask it your five hardest real sizing questions, the between-sizes case, the fabric-stretch case, the fit-across-styles case, and see whether the answer traces back to something your store actually says.
It sees current stock and color, not last week's export. A confident "yes, that's in stock in petite" that turns out to be wrong is worse than no answer at all.
It knows the difference between a fit question and a defect claim, and escalates the ones that need a human rather than guessing at a resolution.
It names a size instead of describing a policy. "Check our size chart" is what a bot says when it can't actually check. "Based on your measurements against our chart, you'd want a large" is what one says when it can.

Where Ernest fits

Ernest is one AI agent that plays three roles for a Shopify or WooCommerce store, and a fashion store touches all three in a single conversation. As a sales agent, it answers sizing, fit, and shipping questions grounded in your actual size chart and product content, and recommends specific products from your live catalog rather than describing a category. As a support agent, it looks up real orders and can start returns and exchanges, with the automation level you choose per action; refunds always wait for your approval. As a product agent, it answers fabric, care, and compatibility questions from what your site already says, rather than guessing.

Honest limits worth knowing before you install anything for this use case:

Ernest doesn't predict fit from body measurements. It answers from your size chart and product descriptions; it isn't a dedicated fit-recommendation engine, and stores that want algorithmic size prediction from shopper measurements need a purpose-built app for that, alongside or instead of a chat agent.
It's reactive, not proactive. It answers when a shopper opens the chat. It doesn't pop open uninvited or send a follow-up nudge to someone who abandoned a cart.
Channels are the storefront widget and email, with escalation to a human by email. No SMS, no Instagram, no phone line, for stores whose sizing questions mostly arrive through DMs today.

Pricing is conversation-based and volume-tiered: a free plan with 100 conversations that never expire, then $49/month for 500, $149 for 2,000, $299 for 10,000, with AI included at every tier and no per-seat or per-resolution charges. For more on how the sales side of this works across categories beyond fashion, see our breakdown of what an AI sales agent does for a Shopify store and our piece on answering pre-sale questions automatically.

How to tell if it's actually cutting returns

Set these up before you launch, or you'll be guessing at whether it worked the same way your shoppers were guessing at sizes:

Track fit-related returns as their own category, separate from color, quality, and changed-mind returns. If sizing questions in chat go up while fit-driven returns go down over the following weeks, that's the mechanism working.
Ask it your hardest real sizing questions before launch. The between-sizes case, the fabric-stretch case, the cross-silhouette case. A wrong answer here is disqualifying regardless of how well it handles easy questions.
Read the escalations. Every sizing question it couldn't answer confidently is a gap in your product descriptions or size chart, worth fixing whether or not you keep the agent.
Watch bracket-order rate, if your platform tracks it. A shopper who gets a specific sizing answer in chat has less reason to order two sizes and return one.

A two-week pilot against real traffic, with those four numbers written down first, tells you more than any vendor's claimed conversion lift. The transcripts, more than the dashboard, will show you exactly which sizing question your product pages still don't answer.