Skip to content
All blog
Retail Operations AI

Answering a customer who bought somewhere else

Masni 7 min read

"Hi, where is my order?"

No order number. No channel. Possibly a different name from the one on the order, because they checked out under a nickname. Sent to your WhatsApp number, which they have because it is on your shop receipt, while the order itself was placed on a marketplace whose chat they have never opened.

Whoever picks that message up has to work out who this person is, what they bought, which of four channels it came through, and what state it is in. Then reply, ideally within the couple of minutes before the customer asks again.

This is the customer-service shape of the same problem that shows up in stock and in money. The channels do not know about each other, and the customer has no reason to care.

Customers do not use the channel they bought on

That is the assumption most setups are built on, and it is wrong.

Marketplace chat exists and works, but it lives inside a seller centre somebody has to be logged into. WhatsApp is where Malaysian customers actually are, and it is where they will contact you regardless of where the money changed hands. Your storefront may have a contact form nobody uses. Email gets the formal complaints.

So the inbound is scattered across places that have nothing to do with how the order arrived, and the person answering is switching between screens to assemble context that should already exist in one.

What "context" actually means here

It is a short list, and none of it is exotic.

Who this is. Matched to a customer record, not a phone number floating free. If the same person has bought at your counter and on two marketplaces, that should be one record with three orders against it, not three strangers.

What they bought, and when. Across every channel, not just the one you happen to be looking at.

Where it got to. The delivery state on the order itself, whichever courier carried it, which is exactly what three couriers, one delivery promise is about.

What has already been said. The previous three messages, so the customer is not re-explaining and your team is not contradicting itself across shifts.

Assemble those four and most enquiries answer themselves in a sentence. Fail to assemble them and every enquiry becomes a small investigation.

Where AI helps, stated honestly

Retrieval is the genuinely useful part. A message arrives, the customer is identified, their recent orders are pulled up, and the delivery state is there before anyone has typed a search. That is not clever, it is just fast, and speed is most of what customer service is.

Drafting is the second part. Most retail enquiries are variations on a handful of questions, and a suggested reply that already contains the right order number and the right status saves the reply being composed from scratch forty times a day.

What it should not do is pretend to be a person, or answer confidently about something it cannot see. If the delivery status has not updated for four days, the honest reply says so and offers to chase it. A reply that invents reassurance is worse than a slow one, because the customer finds out later and now has two problems.

The judgement calls stay human. Whether to refund, replace, absorb the shipping or make an exception for a good customer is a commercial decision, and it is precisely the kind of decision that depends on knowing this person has bought from you eleven times.

The bit that costs money quietly

Enquiries feel like a service cost. Some of them are a reconciliation cost wearing a disguise.

"Where is my order" for something that shipped a week ago is often the first sign of a delivery that stalled. "I was charged twice" is sometimes a genuine double charge and sometimes a settlement that has not landed yet. "I returned this three weeks ago" is frequently the first anyone knows that the return arrived and was never processed against the original sale.

Each of those is a customer-service ticket at the front and a money problem at the back. If the reply is written from a courier portal and the case then closes, the front is resolved and the back never is. If the conversation sits against the order, the pattern becomes visible, because it stops being one annoyed customer and starts being eleven orders with the same courier that all stalled at the same depot.

What to fix first

Put the conversation next to the order. Almost everything else follows from that, because a person who can see the order can usually answer without asking anyone.

Then unify the customer record, so repeat buyers are recognised across channels rather than appearing as a new stranger each time.

Then, and only then, look at automating replies. Automating a reply that draws on incomplete context is a way of being wrong faster.

Common questions

Can customers message us on WhatsApp about marketplace orders?

They already do, whether or not you have set anything up for it. The question is whether the person answering can see the marketplace order while they are in the WhatsApp conversation, or has to go and find it.

Does the AI reply to customers on its own?

It retrieves the context and can draft a reply, but the judgement stays with a person. Anything involving a refund, a replacement or an exception is a commercial decision, and a system that answers confidently about a delivery it cannot actually see does more damage than a slower human reply.

We use marketplace chat already. Why change?

Marketplace chat is fine for marketplace orders, but it only covers one channel, and it is not where most customers start. The aim is not to abandon it. It is to stop the history of a single customer being split across three inboxes that cannot see each other.

How does this connect to reconciliation?

More closely than it looks. A significant share of enquiries are early warnings about stalled deliveries, unprocessed returns and settlements that never arrived. Handled in isolation each one closes as a service ticket. Handled against the order, the pattern behind them becomes visible.

One conversation, whichever door they came through

The customer messaging you does not know which of your channels took their money and would not find the distinction interesting.

Making that true on your side is not about being on more platforms. It is about the person answering being able to see one customer, one history, and one order, without first working out which of four systems is holding it.


See what you could build

Start a free trial and describe what your business needs in plain language — SmartB Studio builds the module for you.

Start free trial
Get started

No credit card · Cancel anytime · Your data stays yours