Sell the Right Size to the Right Shopper, the First Time

The ledger fashion runs on is written in returns

Other industries lose margin to platform commissions. Fashion loses it twice, once at the moment of the lost sale and again in the returns pipeline. The numbers are not small. The National Retail Federation and Happy Returns put overall online return rates at 19.3% of online sales in their 2025 returns research, with total retail returns worth $849.9 billion. Clothing runs well above that average: Coresight Research measured US online apparel returns at 24.4%, roughly $38 billion in a single year, and returns industry benchmarks compiled for 2025 and 2026 place apparel at 20 to 40% depending on the store.

The cause is not a mystery. In Coresight's research, 53% of apparel sellers named sizing and fit as their top reason for returns, and a widely cited McKinsey analysis traced about 70% of apparel returns to poor fit or style. Read those figures together and the conclusion is uncomfortable but useful: most of your returns pipeline is made of questions nobody answered before checkout. Does this run small. Is there stretch in the waist. Will the fabric crease in a suitcase. What does the model's size tell me about mine.

Every one of those is answerable, from information you already publish. A size chart, a fabric description and a fit note answer them in theory. In practice, a shopper between sizes does not want a chart, she wants someone to tell her which size to order. That is a conversation, and until now the only way to have it at scale was a fitting room.

Your checkout is being used as a fitting room

When shoppers cannot get a fit answer before buying, they take both sizes and let your returns desk sort it out. Narvar's shopper research found 58% of consumers admitting to bracketing, ordering multiple sizes or colors with the intention of returning some, and returns industry compilations for 2025 and 2026 describe the habit as mainstream and rising. Each bracketed order looks like revenue on the day it is placed and turns into double shipping, inspection, repackaging and a marked down or unsellable garment three weeks later.

The evidence that pre purchase fit guidance moves this number is published. Researchers at Zalando reported in a peer reviewed ACM SIGKDD paper that algorithmic size recommendations alone reduced size related returns by 3.8%. That is the measured effect of a static hint. A conversation goes further than a hint: it can ask what the shopper usually wears, in which brand, and answer with the specific garment's cut in hand. The point of an AI sales agent in fashion is not just to convert the doubtful shopper, it is to convert her in the right size, so the sale you win stays won.

What your shoppers are actually asking

Run any boutique for a month and the same questions arrive through Instagram DMs, email and the contact form, always faster than you can answer them. On your website they sound like this:

Shoppers still deciding ask:

• "I'm usually a medium but reviews say this runs small. Which size should I get?"
• "Is this dress lined? Is the fabric see through in daylight?"
• "What would you style this with for a rehearsal dinner?"

Shoppers comparing ask:

• "What is the difference between the wrap dress and the slip dress for a curvy fit?"
• "Is this the same material as the top I bought last spring?"
• "Do you have this in a petite length?"

Shoppers ready to buy ask:

• "If I order today, will it arrive before Saturday?"
• "What is your return policy if the size is wrong?"
• "Can I exchange for a different size instead of returning?"

Every one of these is a person holding a credit card. SparkGPT reads your product pages, size charts, fabric descriptions, shipping and returns policies, then answers each shopper's own version of each question instantly and steers the conversation toward a purchase in the right size, a styled add on, or a captured lead when the item they want is out of stock. The fashion demo on the SparkGPT homepage, StyleMaven, shows exactly this motion: a shopper says "elegant, for a wedding," and the agent answers with a specific dress and offers the matching shoes.

How fashion chatbot tools charge, compared

The tools ranking for fashion chatbot searches today are capable products, and it is worth being fair about that. Zipchat is built for ecommerce, handles sizing and availability questions well, and publishes attributed revenue case studies. Chatbase is the biggest name in the category. Tidio's tooling is mature. The difference worth studying is not features, it is what event triggers the bill, as of September 2026.

Per reply pricing deserves a moment of attention in fashion specifically, because fashion conversations are long. A shopper working out her size asks three or four questions before she trusts the answer. Under per reply billing, your most engaged shoppers, the ones closest to buying, are also your most expensive, and a chatty browser who never buys costs the same as a buyer. Under attributed sale pricing the long conversation is free until it produces an order.

From your store to a working agent in three steps

You do not need a developer, and most stores are live in about ten minutes, whether you run on Shopify, WooCommerce or a custom site.

1. Paste your store URL. SparkGPT crawls your product pages, size charts, policies and FAQ, and builds an agent that speaks from your own catalog.
2. Test it like your pickiest customer. Ask it the hard ones: the fit of your most returned item, the fabric question, the delivery deadline. Refine anything you want said differently. This step is free, with no credit card and no expiry.
3. Add one line of code to your theme. Paste a single embed line and the agent greets every visitor, on every product page, at 11 pm on a Saturday when your best traffic arrives. If you are on Shopify, the step by step Shopify guide walks through it.

What a conversation looks like when the agent is selling

Here is how SparkGPT handles the shopper from the top of this page, the one with the wedding in twelve days.

Shopper: I'm between a small and a medium in most brands. Which size should I order in the Aveline midi dress?

Agent: The Aveline runs true to size with a defined waist and no stretch in the fabric, so if you are between sizes I would take the medium for comfort through the waist. The size chart puts the medium at a 28 to 29 inch waist. What do you usually wear in a fitted dress?

Shopper: A 28 waist usually. It's for a wedding on the 14th, will it arrive in time?

Agent: Then the medium is your size. Standard shipping delivers in 3 to 5 business days, so ordering today arrives with about a week to spare, and exchanges for a different size are free within 30 days if you want the safety net. Would you like the matching wrap in the same fabric? Guests often pair them for an evening ceremony.

Shopper: Just the dress for now.

Agent: Done. Here is the checkout link for the Aveline in medium. Enjoy the wedding.

Notice what happened. The agent answered the size question with the garment's actual cut, confirmed it against what the shopper wears, resolved the deadline anxiety with concrete dates, offered one styled add on without pushing, and asked for the order in the right size. One sale won, one bracketed double order and one probable return quietly avoided.

Pricing built for how fashion actually sells

Fashion demand is seasonal and spiky. Drops, sales and gifting seasons stack traffic into a few loud weeks, and per conversation or per reply pricing bills you hardest exactly then. A flat subscription sends the same invoice through your quietest February.

SparkGPT is priced on results.

You only pay when SparkGPT makes you a sale. A sale is attributed when it is placed within 24 hours of a conversation with your agent, and every attributed sale appears in your ROI dashboard with its value, so the number you see is the number you are billed on. Billing runs on prepaid credits, where 1 credit equals $1, and if your balance runs out the agent pauses instead of surprising you with a bill.

The math fits fashion margins. If the agent helps close a $180 dress inside the attribution window, the fee is $9. If it answers forty questions from browsers who never buy, the fee is zero, where per reply pricing would have billed every one of those answers. Full details are on the pricing page, and if you are comparing against a flat subscription line by line, the SparkGPT vs Chatbase comparison walks through it.

Frequently asked questions from fashion stores

Can the agent really answer sizing questions accurately?

It answers from what you publish: your size charts, fit notes, fabric descriptions and product reviews. If your size chart says the medium fits a 28 to 29 inch waist, that is what the agent says, consistently, on every shift. It does not invent measurements you have not published, so the better your size information, the better its answers.

Can it make styling suggestions like a personal stylist?

Yes, from your own catalog. The agent knows every product you sell, so when a shopper says she needs something elegant for a wedding, it can recommend a specific dress and offer the matching pieces, the way the StyleMaven demo does on the SparkGPT homepage. It suggests what you actually stock, never items you do not carry.

If a shopper chats about a dress today and buys tomorrow morning, does that count as an attributed sale?

Yes, if the purchase lands within 24 hours of the conversation. The shopper's session is marked when they talk with your agent, any order placed inside the next 24 hours is attributed, and it appears in your ROI dashboard with its value. The 5% applies to those attributed sales and nothing else.

Will an agent actually reduce my return rate?

What the evidence supports is this: most apparel returns trace to fit and sizing, and Zalando's published research measured that even static size recommendations reduced size related returns by 3.8%. An agent answers the same fit questions interactively before checkout, which is the moment sizing mistakes are made. Your results will depend on your catalog and how good your size information is.

Does it work with my Shopify theme?

Yes. Deployment is one embed line pasted into your theme, and it works alongside your existing apps. The Shopify walkthrough on the SparkGPT blog covers the whole process, which most merchants finish in about ten minutes.

What does it cost me in the off season?

The Growth plan is $49 per month plus 5% of attributed sales, so in a quiet month the variable part shrinks toward zero and your cost floor is $49. Prepaid credits mean the agent pauses rather than overdrawing if the balance runs out. Your bill grows only in the months your sales do.

Your shoppers are between sizes right now

Paste your store URL at www.sparkgpt.ai and ask your own agent the fit question your customers ask most. It takes about ten minutes, with no credit card, no expiry and no risk.