Cross-Sell in a DM: A Suggestion, Never a Squeeze

Somebody messages a studio pottery account: "do you ship the blue mugs to Hyderabad?"

There are two replies waiting to be written, they are about the same length, and one of them is fine.

The one on the left is more useful than the answer they asked for. The one on the right is a different message, wearing the answer as a hat.

The test that actually works

Read the added sentence on its own and ask whether it makes the reply a better answer to the question that was asked.

The saucer sentence passes, because someone buying mugs to be shipped is better off knowing they tip. It is the thing a person behind the counter would say. It survives being read by somebody who did not want it, because it is information rather than a request.

"Reply BUNDLE now" fails, because it is not an answer at all. It is a demand for another action, addressed to somebody who came for one piece of information, and the shipping question is now a doorway.

That test also tells you when to add nothing. Most replies should carry no suggestion whatsoever. If the honest answer to "are you open on Monday" is yes, the correct reply is yes.

Never invent scarcity, and understand why you cannot avoid it

This is the rule that gets broken by accident more than by intent, because the sentences are so standard that they have stopped reading as claims.

"Only two left." "Selling fast." "Last day." "Almost gone."

Here is the structural problem. There is no inventory integration in this product. None. It does not connect to a store, a spreadsheet, a marketplace or a stock system, and it never knows how many of anything you have. So a template that says two are left is not a stock figure that has gone stale — it is a number that was never connected to anything in the first place, repeating itself to everyone who triggers that automation, for as long as the automation is on.

Same with a deadline. Nothing in a template counts down and nothing switches itself off. "Offer ends tonight" is true on the evening you wrote it and is a lie every day afterwards, silently, at scale, in writing, with your name on it.

If you run a real deadline, put the actual date in the wording, and put switching the automation off in the same calendar entry as the campaign. If you have real stock scarcity, say it in a person-typed reply where you have just checked.

You get one message, so stop planning a sequence

Most advice about upselling assumes a series: the answer today, the nudge on Thursday, the offer on Sunday. That entire shape is unavailable here and it is worth being blunt about it.

There are no scheduled sends. No follow-up sequences. No drip, no broadcast, no reminders, no abandoned-cart chase, no way to message someone who has gone quiet. The automation replies to what arrives, and that is the end of it.

The DM window makes it stricter still. You have twenty-four hours to reply inside a thread, and that clock restarts only when the other person writes again — not when you do, and not because you would like more time. If they stop typing, the conversation is over regardless of what your funnel diagram says.

So the only moment available is the reply they actually asked for. That is not a limitation to work around. It is a fairly good constraint, because it removes the option of pestering somebody and leaves you with the option of being useful once.

Where an automated suggestion should never appear

Any thread with a complaint in it. Somebody chasing an order, reporting a fault, or asking for a refund is not a cross-sell opportunity, and an automated bundle offer under a complaint is the screenshot that travels. Complaint vocabulary belongs in the negative keyword list.

Any thread where the purchase is emotional or sensitive. A condolence order, a medical enquiry, anything involving a child. The suggestion may be perfectly relevant and it will still read as someone trying to make money out of a bad afternoon.

Any thread where a human has already replied. Takeover handles this for you — the moment somebody types by hand, automation stands down in that thread permanently — but it is worth knowing that this is the mechanism protecting you from an automated pitch landing under your own carefully judged reply.

And first-time senders deserve a lighter touch than the setting implies. Somebody's first message to a business is a bad place to be sold a bundle.

Make it sound like the person who owns the shop

The suggestions that work are the ones a customer reads as a recommendation rather than a placement, and that difference is mostly voice.

Voice DNA is built from replies the account owner has actually written, which for this purpose is the useful property: your real recommendations already sound like you, hedge where you hedge, and mention the thing you genuinely think pairs well. A suggestion generated in generic ecommerce copy sounds like a plugin, no matter how relevant the product is.

The short version

Answer the question. If you know something that makes the answer more useful, say it in the same breath. Do not claim scarcity you cannot see. Do not put a call to action on a reply that was only ever a fact. And do not build a plan that needs a second message, because there is not going to be one.

Templated replies are free and unlimited; free tier is 100 credits with no card, packs from ₹499. No inventory, no payment links, no CRM sync — leads leave as CSV. Official Graph API only.

The ethics post covers the wider version of this argument, writing the reply itself is the practical companion, and what actually moves a conversation forward is worth reading before you add a single sentence to any template.