The chat that sells instead of answering

Almost everybody treats chat as a help desk: a question arrives, an answer is given, the conversation closes. Efficient, measured in tickets closed, and completely blind to the fact that a good share of those questions are not support at all.

They are the last doubt before buying. Somebody who writes "is this jacket warm?" does not have a problem: they have a card in hand and an uncertainty to remove.

The difference between answering and selling

Take the most common question in fashion: "what size should I take?".

The help desk answer is a link to the size guide. Technically correct, closes the ticket, and hands the customer the work of measuring themselves and interpreting a table.

The sales answer is: "for your height and the way this model fits, the M. If you want it roomier, the L." Same information, one step instead of four, and a decision made rather than postponed.

The second requires whoever answers to know two things a help desk does not have: how that product fits, and what that person asked. Those are data, not courtesy.

The three moments a conversation can become an order

When a fact is missing. The most frequent and the easiest: the page does not say whether it is compatible, washable, or arriving before Saturday. Answering with the real fact closes the sale in that same moment.

When what they want is not there. Here the help desk says "sorry, out of stock" and the conversation dies. An assistant that sees inventory offers the nearest alternative or says when it returns, and a share of those conversations becomes an order anyway.

When they are buying part of something. Somebody buying a jacket is often assembling an outfit; somebody buying a printer will buy toner. Suggesting the complement at the right moment is not pushiness: it is what a shop assistant does, and nobody finds it intrusive there.

Why almost no chatbot does this

Because doing it requires seeing the catalogue, the stock and the orders. An assistant that cannot see them can only answer generically, and generic advice about a specific product is obvious immediately.

The result is that chatbots get judged on how many tickets they deflected, and nobody looks at how many sales they closed, because that figure exists in no dashboard. Measure only one half of the value and you end up building only that half.

An example with numbers

At a womenswear brand the assistant does not just answer about size: it builds the complete look and recommends the fit. The numbers for the period: chat conversion +22%, items per order +15%, average order value +19%, returns -18%.

The first three measure the selling, the fourth measures the quality of the advice. It is the pair that matters: selling more with more returns is not selling more.

At a cheese producer, where the subject was not style but shipping a fresh product, the same approach gave -60% repeat questions and +18% chat conversion. Both cases, with all the numbers.

The line to respect

There is a line, and it is worth writing down before crossing it: offering a product while a customer has a problem with an order is the fastest way to turn a customer into a former one.

A return in progress, a late delivery, something that arrived broken: those conversations get resolved, full stop. The sale comes afterwards, if the resolution was good.

Frequently asked questions

How do you measure whether chat is selling?

By comparing the conversion of people who used chat with those who did not, over the same period. It is not a controlled experiment, because people who write are already more interested, but the gap is still the most useful number available without making life complicated.

Are automatic suggestions not annoying?

They annoy when they are irrelevant. A suggestion that follows from the question just asked reads as service; one that arrives regardless reads as advertising. The difference is in the data, not the tone.

Do I need inventory integration?

To answer, nearly always. The moment chat sells most is when what the customer wants is not there: without real-time availability that moment is lost every time, and with it the most profitable part of the conversations.

Does it work on expensive products?

It works differently. On a large purchase chat rarely closes the sale in the same conversation: it removes the doubts that otherwise cause postponement. The signal to watch is not the immediate order but how many come back later.

What if it recommends the wrong product?

It comes back, and it costs you a return plus trust. Which is why advice has to come from the product's data rather than a generic similarity, and why an assistant must be able to say it is unsure instead of recommending anyway.