YundaDeskScroll through a cross-border support inbox and it feels chaotic - every message worded differently, every customer seemingly asking something new. Sort those same tickets by channel and category, though, and the opposite is true: the questions cluster hard into a small set of buckets - where is my package, can I return this, what size fits me, do I owe extra fees. The problem was never variety. It was that nobody had structured these buckets into something reusable, so every ticket got judged from scratch.
No invented statistics here, no borrowed authority. The answer already lives in your own ticket history - it just hasn’t been organized into a framework yet.
Why “every question is different” is usually an illusion
Support leads think their volume is chaotic because they’re reading raw wording instead of underlying intent. The same issue shows up in a dozen phrasings:
- “Where’s my package” / “Tracking hasn’t moved” / “Is it stuck in customs” - all shipping status.
- “Can I return this” / “I want a different color” / “Something’s wrong with it” - all returns and exchanges.
- “What size is S roughly” / “How does Asian sizing compare to EU” / “I’m on the smaller side, what should I get” - all sizing.
- “Why am I being charged more” / “Is this tax on me” - all duties and extra fees.
Based on our observations of cross-border customers, these four buckets - shipping status, returns and exchanges, sizing, and duties or extra fees - tend to account for most pre- and post-sale contact volume across independent stores and marketplace shops, with the mix shifting by category and target market.
If the team has not formally tagged tickets yet, start with one week of history as an illustrative sizing exercise. The goal is not to claim an industry average, but to prove that messy wording can collapse into a small set of intent buckets.
Four intent buckets in one week of cross-border tickets (illustrative)
Layer one: shipping status, the perennial number one
No matter what you sell, “where’s my package” is almost always the single highest-volume bucket. It breaks down into sub-questions that need different depths of answer:
| Sub-question | What the customer actually wants to know | Knowledge source needed |
|---|---|---|
| Has it shipped | Whether the order has been processed | Order system status |
| Tracking looks stuck | Whether it’s lost and whether to worry | Carrier tracking + common stall explanations |
| How long will customs take | When it will actually arrive | Destination-country clearance benchmarks |
| Can I fix a wrong address | Whether there’s still time to intercept | Carrier address-change rules + time windows |
What these have in common: the answer depends heavily on the real-time status of that specific order, not a generic policy line. If AI support only says “please be patient,” the customer escalates to a human immediately. What actually works is capturing the recurring patterns - common causes of stalled tracking, typical clearance timelines by destination, address-change windows - as knowledge, so AI can offer a grounded explanation first and escalate only when the specific order needs a human look.
Layer two: returns and exchanges, the messiest rules and the easiest place to slip
On the surface, returns questions are just “can I return this.” Underneath sits category rules, time windows, who pays return shipping, and the extra cost of shipping something back across a border. The common structure:
- Eligibility - is the item within the return window, does it meet conditions (unopened, unworn, non-final-sale, etc.).
- Process - where to file, whether original packaging is required, who covers shipping.
- Money handling - full refund or partial, original payment method or not.
The first two are rules-based and belong in the knowledge base for AI to answer directly. The third - the moment it touches an actual refund amount, compensation, or a price change - has to route to a human for approval. That’s not a capability gap, it’s a governance line: AI can help a customer understand exactly which rule applies to their case, but the button that actually issues a refund should always be pressed by a person.
Layer three: sizing and fit, a conversion killer and a support magnet
Apparel, footwear, and baby-product sellers know this one well - sizing questions are consistently underestimated as a share of total contact volume. And sizing isn’t a lookup-table problem. What the customer is really asking is “will this actually fit me,” tangled up with height, weight, fit preference, and worry about the cost of getting it wrong.
A well-structured sizing knowledge base usually needs to cover:
- Category-specific size charts, not one generic international conversion table.
- Fit notes (runs slim, runs loose, size up recommended).
- Guidance mapped to common height/weight ranges, rather than making the customer guess.
- A cost reminder for cross-border returns, since round-trip time is long enough that confirming before ordering matters.
This is exactly where “AI that gets smarter with use” earns its keep: a support lead feeds in what an experienced agent already knows - “this category tends to run a size small” - as a knowledge update or a correction suggestion. Once approved, AI support gives a sharper recommendation next time instead of escalating the same fit question over and over.
Layer four: duties and extra fees, the anxiety unique to cross-border
This bucket doesn’t exist for domestic sellers, but no cross-border seller escapes it. The real anxiety isn’t “do I owe tax” - it’s “what’s my total landed cost, who’s responsible for it, and am I about to get hit with a surprise charge.” Common questions include:
- Whether the destination country has a de minimis threshold and what triggers duties above it.
- Whether the price already includes tax (DDP vs. DDU).
- Whether being asked to pay more at clearance is a scam.
- Whether the fee rules differ between the brand’s own site and a third-party marketplace.
This bucket depends heavily on a knowledge base structured by destination country and by channel, because the same question has a completely different correct answer depending on where the customer is. A single generic “about taxes and fees” article will confidently give the wrong answer to some fraction of customers - and generate more escalations, not fewer. This is also where cross-border CRM fields for country, language, and time zone earn their keep: AI can identify which market a customer is in and match the right duty rules, instead of giving one answer that’s technically correct nowhere in particular.
Turn the four buckets into knowledge base structure, not a pile of tickets
Once the four layers click, the real work is reorganizing your knowledge base around them instead of continuing to patch things ticket by ticket. A workable sequence:
- Tag recent tickets into shipping / returns / sizing / duties and see the actual mix and sub-questions for your store.
- For each bucket, draw the line between “rules-based, AI can answer directly” and “needs a human judgment call.”
- Structure shipping-delay patterns, clearance timelines, sizing guidance, and duty rules by country and channel inside the knowledge base.
- Have agents flag “this should go in the knowledge base” in the moment, instead of relying entirely on a retrospective review.
The exact mix will shift by category (apparel vs. electronics vs. home goods) and by target market (US/EU vs. Southeast Asia vs. Middle East), so it’s worth tracking your own numbers by channel rather than borrowing someone else’s ratio. To build this structure properly, see how a knowledge base feeds AI support. To see where the line between AI answering and escalating to a human should sit, see the AI-first, human-backed boundary.
Cross-border support questions were never actually scattered - they only look that way when read one raw message at a time instead of sorted by intent. Structure shipping, returns, sizing, and duties into four clear layers, and AI support finally has something solid to answer from, while agents get their time back for the cases that genuinely need a human call.