Using Chat Data to Decide What to Stock

Sales data tells you what you sold. It cannot tell you what customers asked for and did not get.

That information sits in your inbox: the size you did not have, the colour you stopped stocking, the service you do not offer, the brand three people asked for last week. Most businesses never capture it, and so keep buying what sold last season.

The four signals in your inbox

1. Out-of-stock requests. Someone asked for a specific item you did not have. The clearest demand signal there is.

2. Questions you could not answer with a yes. “Do you have it in XL?” “Do you do home service?” “Do you stock [brand]?”

3. Fallback messages. The questions your automation did not understand are often about products or services you do not offer, phrased in ways you never anticipated.

4. Lost enquiry reasons. Why the customer did not buy: price, size, colour, timing, availability.

Each needs a capture mechanism, or it disappears.

Capture it as it happens

Out-of-stock: make it a required tag when an agent tells a customer an item is unavailable.

Agent marks: OUT-OF-STOCK → item → size/variant → customer

Unmet requests: a tag for “asked for something we don’t have”, with a free-text note.

Lost reasons: a mandatory field when marking an enquiry lost, from a short list: price, availability, size, timing, chose competitor, no response.

Fallback export: your automation already logs these. Export weekly and read them. See our chatbot metrics guide.

Keep the capture to one tap plus an optional note. Anything more and agents stop doing it during a busy hour.

What the data tells you, monthly

The size curve example

A boutique sells out of M and L every season, and ends with unsold XS and XXL. Sales data confirms what sold. Only the inbox data shows how many customers asked for XL and were told no.

Counting those requests for one season changes the next purchase order, and the change is measurable.

The same logic applies to colours, materials, pack sizes, price points and service timings.

Separate real demand from noise

Not every request is a signal.

• Repeated by many people: real demand

• Asked once by one person: note it, do not act

• Asked by price-shoppers who never buy: check whether the people asking have ever purchased

• Seasonal: a request in November may not mean annual demand

• Competitor checking: occasional, and usually identifiable

Set a threshold: act when an item is requested [X] times in a month by distinct customers, at least some of whom have bought before.

Use it before you buy

Before placing a purchase order:

1. Pull the out-of-stock request report for the last [3] months

2. Rank by frequency

3. Cross-check against sales data for items that sold out early

4. Check lost reasons for pricing signals

5. Adjust quantities and the range

This takes twenty minutes and is better information than any trend report, because it is your own customers asking for specific things.

Tell customers when it arrives

The best part: you already have the contact.

• Tag customers who requested an out-of-stock item

• When it arrives, message exactly those people

• “You asked about [item] — it’s back in stock”

This converts extremely well, because the customer already decided they wanted it. It also costs almost nothing.

Keep it to people who asked, within a reasonable period, and respect opt-outs.

Test demand before committing

For a product you are unsure about:

• Post it as a question in your channel or story: “Should we stock this? Reply YES”

• Offer pre-booking: “Pre-book and we’ll order it”

• Run one small batch and measure

• Ask your top customers directly

Pre-booking is the strongest test, because it asks for commitment rather than opinion.

See our chat surveys guide.

Also use it for content and ads

The same data tells you what to post and what to advertise:

• Frequently asked questions become content

• Frequently requested products become ad creatives

• Common objections become the copy on your product pages

• Fallback themes become blog topics

Your inbox is the cheapest keyword research you will ever do, and it is specific to your actual customers.

Measure whether it worked

If out-of-stock requests for the same items keep recurring after two buying cycles, the data is being collected and not used.

Frequently asked questions

How do we get agents to tag consistently? One tap, a short list, and show them the monthly report so they see it being used. Tagging stops when nobody acts on it.

Is a request as good as a sale? No. Weight it by frequency, by distinct customers, and by whether those customers have purchased before.

What if we cannot stock what customers ask for? Record it anyway. It may justify a supplier conversation, a different range, or simply a better answer than “no”.