Two weeks ago I read a chat that still annoys me. A customer asked about a beige linen shirt, got a straight answer, wrote "perfect, I'll take it," and paid. Ninety seconds, start to finish. Clean sale.
That same store sells a linen spray that stops the shirt looking slept-in by noon. It sells the trousers the shirt was photographed with. Nobody said a word about either one.
The customer left happy. The seller left money on the table and never saw it happen.
My honest read, after going through a lot of seller inboxes: most small stores don't have a traffic problem. They have a basket problem. Someone is already in the chat, already convinced, already opening their banking app. That is the cheapest selling moment you will ever get, and most sellers spend it typing "thanks, order noted".
Upsell and cross-sell are two different moves
Upselling is a bigger or better version of the thing they already picked: the 500ml instead of the 250ml, the two-pack, the heavier fabric, six months instead of one.
Cross-selling is a second, different item that makes the first one work: the case with the phone, the spray with the linen, the spare filter with the kettle.
The difference matters because of what each one does inside the customer's head. An upsell asks them to reopen a decision they just closed. A cross-sell asks them to add, which is a much smaller ask. In chat, where you get maybe one sentence of real attention, cross-sell wins more often than sellers expect.
So here's the rule I'd give anyone starting: a cross-sell needs a fit, an upsell needs a reason. If you can't say in one line why the bigger one is better for this specific person, don't offer it. "It's better value" is not a reason. "You'll finish the small one in three weeks" is.
Timing decides this, not wording
You can have a perfect suggestion and still lose the sale by saying it forty seconds too early.
Before the customer has decided, an extra product is noise. They're holding a question in their head. Does it fit? When does it ship? Is that the real colour? You just handed them a second thing to think about, and thinking is what stops people from paying.
After they've decided, the same sentence reads as help. They've already accepted the spend. Buying mode is real, and its window is short and generous.
Moments that work:
- Right after "I'll take it", before you send payment details.
- Right after payment lands, while the parcel is still open: "I can put it in the same box."
- When they're re-ordering something they bought before. They already trust the product.
- When they ask a question that reveals the gap themselves ("does it come with a charger?").
Moments that quietly cost you:
- While they're still comparing two options. You'll just add a third.
- While any question is unanswered: size, price, delivery, availability.
- Anywhere near a complaint, a delay, or a return. Never sell to someone who is currently annoyed with you.
- The second and third suggestion in one conversation. One is help. Two is a pitch. Three and they stop reading you entirely.
One suggestion per conversation. I'd put that on the wall.
What a suggestion that actually works sounds like
Most chat upsells fail because they're lazy. "Would you like to see our other products?" isn't an offer, it's homework. The customer has to go browse, decide, and come back. They won't.
A suggestion that works does three things in a line or two: it names one specific item, it gives a reason tied to what they just bought, and it makes saying yes a single word.
Compare:
- Weak: "We have accessories too, check our page."
- Better: "Do you want the care spray as well?"
- Strong: "Linen creases hard in the first week. Most people take the care spray with it, and it fits in the same box. Add it?"
For a size or quantity upgrade, the reason is usually time or math:
"At one pump a day the 250ml is gone in about three weeks. The 500ml lasts you two and a half months for 40% more. Want me to switch it?"
For a bundle, make the saving specific instead of vague:
"Two of them ship for the same delivery cost as one, so the second one effectively costs you 25,000 toman less. Should I make it two?"
Notice what's missing: no "special offer", no countdown, no "limited stock" unless it's true. You don't need urgency on someone who already decided to buy. You need clarity. If you want more phrasing you can steal outright, we've collected a batch of DM scripts that actually close sales.
Three numbers, and the third is the one nobody watches
Track these weekly. They take ten minutes.
- Attach rate. What share of orders included a second item? Most sellers who start doing this deliberately move from near zero to somewhere between 15% and 30%.
- Average order value. Compare the four weeks before and after. If your typical basket sits around the median DM basket of roughly 1.3 million toman, one accepted complement can move that line more than a month of new followers.
- Silence rate. What share of conversations go quiet immediately after your suggestion? This is the one nobody measures, and it's the one that catches damage.
The third number is the whole point. If your attach rate climbs six points but silence after a suggestion climbs four, you are not winning, you are trading finished sales for maybes. Cut back until silence goes flat, then push again.
Why good sellers still don't do it
Three reasons, and none of them are laziness.
They think it's greedy. It isn't, as long as the second item genuinely helps. Selling a linen shirt to someone with no way to keep it looking good is the less honest option.
They're tired. At message five you're sharp. At message eighty on a Thursday night you're answering in three words. The suggestion is the first thing to disappear, and it disappears exactly when volume is highest, which is when it would have been worth the most.
They overdo it and then quit. Someone reads an article like this one, offers three add-ons to everybody for a week, watches customers go cold, and decides upselling doesn't work in DMs. It works. Three at once doesn't.
Build the pairing map first
Before you automate anything, spend twenty minutes on paper. Take your top ten products and write three columns:
- The product.
- One complement and one upgrade. Only one of each. If you list four, whoever is answering will pick randomly.
- The reason line, in customer language. Not "related item". Something like "the strap sold with it is the one that doesn't stretch".
Two hard rules for that sheet. Never pair anything that's out of stock, because a suggestion you can't fulfil is worse than no suggestion. And re-check it monthly, because your stock moves faster than your notes do.
This sheet is the whole strategy. Everything after it is delivery.
The part worth handing to an assistant
Consistency is where software beats a human, and it isn't close. An AI assistant makes the same well-timed suggestion at 2am on the four-hundredth conversation as it does on the first. It remembers what this customer bought in March, so the re-order gets the right complement instead of a generic one. It reads the whole thread before it speaks, so it knows whether the sizing question was ever answered.
What matters more is what you tell it not to do. Give it boundaries in plain words: never suggest anything while a complaint is open, never suggest before the customer's own question is fully answered, never suggest something out of stock, never make a second suggestion if the first was declined. An assistant with those four limits will out-sell one with a longer product list and no manners.
That's roughly how we built this behaviour into Vardast: it answers your customers on Instagram, WhatsApp, Telegram and your site from your real product knowledge, in your voice, and it can carry your pairing map into every conversation without ever pushing a fifth message at a customer who has already said no. If you want the wider version of this playbook, start with how selling in Instagram DMs actually works end to end.
One sentence, one time, with a reason
The stores that grow their basket aren't the ones with the cleverest offers. They're the ones that make one relevant suggestion, in the right five seconds, and then shut up and take the order.
Pick your top ten products tonight. Write one complement each. Use it for two weeks and watch the silence rate. That's the whole experiment, and it costs you nothing but twenty minutes.