YundaDeskOpen the inbox for any apparel storefront and six or seven out of every ten messages circle the same question: what size should I get. Customers send height and weight asking for a fit recommendation, then come back after delivery to ask how to return it, and a separate crowd sits in the DMs asking when the next restock lands. These three categories eat up most of an agent’s day, and most of it is repeat work - same question, same answer, different customer and different channel.
Apparel support carries a different kind of pressure than support for standardized products. A hardware brand answers “does this work” and “how do I get it repaired.” An apparel brand answers “will this fit me” and “is returning this going to be a hassle.” The first kind of answer comes from a spec sheet. The second comes from trust - and trust builds slowly and breaks fast. One bad sizing recommendation and a customer may walk away from the brand entirely. That’s why apparel support deserves its own playbook.
Sizing: the single biggest chunk of the workload
Sizing questions look simple but carry several layers of complexity. The same “size M” can vary across cuts, fabrics, and even production batches. A customer’s stated height and weight isn’t always precise. And some customers aren’t really asking about size at all - they want to know if it’s slimming, or whether the fabric pills after a few washes.
Handling sizing well doesn’t mean asking agents to memorize a size chart for every SKU. It means feeding that information into the AI agent’s knowledge base:
- A detailed size chart per SKU or collection - not just S/M/L, but actual chest, waist, and length measurements.
- Fit notes: slim cut, relaxed fit, runs small or true to size.
- Fabric stretch and shrinkage behavior after washing.
- A running set of frequent sizing Q&As (e.g., “I’m 5’5” and 130 lbs - which size fits best?“).
The AI agent answers from the knowledge base first, and hands off to a person when it can’t answer or when the customer asks for one. That way agents stop repeating “what size should I get” over and over, and can spend their time on the cases that actually need judgment - unusual body types, customers who’ve already exchanged sizes twice and are frustrated, or custom sizing requests.
Apparel support conversation mix (illustrative)
Returns and exchanges: apparel’s natural high-frequency case
Apparel return rates run higher than most categories by default - wrong fit, color looks different from the photo, fabric feels different than expected. Support has to handle more than “how do I return this”:
| Scenario | What customers ask | What matters in the reply |
|---|---|---|
| Wrong size | Can I exchange for a different size instead of a refund | Confirm stock first, then give the exchange steps - don’t leave them waiting |
| Color or fit disappointment | This doesn’t look like the photo | Explain lighting/photo variance, offer the return path, no arguing |
| Tags removed / tried on | Can I still return it with the tag off | Answer clearly according to store policy, no hedging |
| Repeat exchanges | This is my third exchange already | Hand off to a person to decide if special handling is needed |
Most of these scenarios follow a fixed policy, which makes them a good fit for the AI agent to handle automatically. Put the return and exchange policy - return window, who pays shipping, tag requirements, whether exchanges carry a price difference - into the knowledge base, and the AI can answer accurately the moment the customer asks, without waiting for an agent to come online.
Fit and return answers are part of the product experience
What still needs a human: confirming a refund amount, approving an exception outside policy, or de-escalating a customer who’s upset. Refunds and compensation always route through human approval - the AI can collect order details and organize the reason for return, but the final approval stays with a person.
Restock questions: repeat work in a different shape
Before and after a product launch, DMs and comments fill up with the same handful of questions: when’s the restock, are other colors coming, when does the next batch arrive. These questions carry low information density but high volume - if every one goes through a human, agents spend a lot of time delivering “still waiting” answers.
The knowledge base can hold a running update of launch information:
- Expected restock dates for out-of-stock SKUs.
- Planned color or style releases (or “TBD - follow our official channels” when it isn’t confirmed yet).
- Standard answers for common “do you have this in a bigger/smaller size” questions.
Letting the AI agent absorb these high-frequency, low-risk questions frees agents up for conversations that actually need judgment - a customer wanting to place a bulk order, a potential collaboration inquiry, or an after-sales issue that goes beyond sizing.
Omnichannel apparel support: the answer has to follow the customer
Apparel brands naturally have scattered touchpoints - Instagram DMs asking for styling advice, TikTok comments asking where to buy the outfit, WhatsApp messages checking on an exchange, email for wholesale inquiries. YundaDesk covers channels including the website widget, custom API, email, WhatsApp, Telegram, Messenger, Instagram, TikTok, LINE, WeChat, VKontakte, Zalo, and YouTube, all flowing into one shared inbox and one customer profile.
That means a sizing question a customer asked on Instagram last week shows up in full context when they follow up about an exchange on WhatsApp this week - no need for the customer to repeat themselves. Which channels to connect depends on the brand’s target market: a brand selling into North America and Europe may lean more on Instagram and WhatsApp, while a brand selling into Russian-speaking markets will use VKontakte. That’s a matter of picking what fits, not a limitation.
Turning frontline experience into answers the AI can use
Apparel support carries a lot of “only the veteran agent knows this” knowledge - this supplier’s size M actually runs small, this batch of fabric pills so customers should be warned in advance, a particular influencer’s outfit gets asked about constantly. If that knowledge only lives in one agent’s head, new hires can’t use it and the AI can’t learn it either.
YundaDesk turns that kind of experience into a controlled learning loop. When the AI can’t answer, an agent writes the answer, or an agent clicks “correct AI” to flag a wrong fit judgment, the system generates a pending learning suggestion rather than changing AI behavior directly. An owner or lead reviews the suggestion in the review queue to check whether it’s accurate and whether it should apply to similar products. Only after approval does it become a knowledge base entry or an agent skill. Every suggestion stays traceable, testable, and reversible with one click.
This matters especially for apparel, because judgment calls about fit and fabric carry some subjectivity - they’re not as black-and-white as a shipping address. Keeping final approval with the owner lets the system absorb frontline experience without letting a wrong call get copied across the catalog. For the full mechanics of this loop, see teaching AI that gets smarter over time.
A knowledge base checklist before launch
When apparel sellers build out their knowledge base, it’s easy to miss content that doesn’t feel like “policy” but customers ask about constantly. Start with this list:
- Detailed size chart and fit notes for every collection or SKU
- Fabric composition, stretch, washing, and care instructions
- Return and exchange policy: window, who pays shipping, tag requirements, exchange price differences
- Restock and launch timing (mark “TBD” where needed, but keep an update process)
- Frequent sizing Q&A organized by height/weight range
- Handoff boundaries: complaints, large refunds, wholesale inquiries
The knowledge base isn’t a one-time project - it needs updates with every season change, new launch, and policy revision. For a full guide on building and maintaining it, see the knowledge base that feeds AI.
The hard part of apparel support was never the volume - it’s how many questions look simple but are actually judgment calls in disguise. Whether to allow an exchange, whether the color matches the photo, whether a batch of fabric holds up - all of it takes experience to answer well. Let the AI agent handle the standardized part, keep the judgment-heavy part with humans, and let a controlled learning loop slowly turn veteran know-how into something the whole team can use. That’s what a real support system looks like for an apparel brand.