Ordering in the chat window

Every distributor I have worked with has the same shape to their customer base. A few hundred outlets justify a regular visit. Several thousand do not — they order small, order irregularly, and cost more to call on than they return. They are not unprofitable customers; they are customers the coverage model cannot reach economically.

The usual answer is a portal, and the usual outcome is that nobody logs in. A shopkeeper is not going to remember a password for a site they use twice a month.

But every one of those outlets is on a messaging app for most of the working day, and many of them are already sending orders that way — as a photo of a handwritten list, to a salesperson's personal number, at nine at night. The channel is not a proposal. It already exists. What is missing is that the order lands in someone's private chat instead of in the system.

Two modes, and the choice actually matters

Chat ordering is usually sold as one thing. It is two, and picking the wrong one for a given customer is the most common way these projects disappoint.

Guided ordering — the outlet taps through a structure

The conversation is a fixed navigation: cards the outlet selects, category by category, into a cart, to a confirmation. No free text is interpreted, so nothing can be misunderstood.

This suits the majority of the long tail better than people expect. A shopkeeper reordering the same fifteen lines every fortnight does not want a conversation — they want their usual order in four taps. The mode is predictable, it works for a user with limited literacy or a language the model handles poorly, it costs nothing per interaction, and it cannot invent a product that does not exist.

Its limit is equally clear: anything the structure did not anticipate is unreachable. "Do you have the smaller pack in stock?" has no card.

Conversational ordering — the outlet just talks

Free text, interpreted. The outlet writes what they want in their own words and the assistant does what a good tele-ordering desk does: resolves the item, confirms quantities, and — the part that earns its keep — volunteers what a person on the phone would.

Stock position at the moment of asking, so an order is not placed against something that will short-ship.
Scheme recommendations, including the sentence that matters commercially: you are five cases below the next slab, and closing that gap is worth more than the five cases cost.
Substitutes when something is unavailable, instead of a silent shortfall discovered at delivery.

That is not chat as a novelty. It is the tele-order conversation a distributor used to staff, available at nine at night without staffing it.

The cost is a model call per turn and a genuine ambiguity surface — which is precisely why the order lands as a draft the outlet confirms, and why pricing, credit and scheme calculation are enforced server-side exactly as in every other channel rather than being whatever the conversation concluded.

How to choose

Segment it. Guided for the high-frequency repeat orderers and anyone in a language or literacy context where interpretation is risky. Conversational for outlets whose orders vary, who ask questions, and where the upsell conversation is worth having. Most distributors should run both, and the honest starting point is guided — it is cheaper to operate, and it teaches you what people actually ask before you pay a model to interpret it.

Beyond taking the order

Order capture is the obvious use and the least valuable half. The channel's real return is that it absorbs the phone calls that currently occupy your office.

Order status. "Where is my order?" is the single most common inbound call in distribution, and it is answerable without a human — confirmed, picked, dispatched, out for delivery.

The invoice, as a compliant document. The outlet receives a GST-compliant, e-invoiced PDF in the same thread as the order. That matters more than it sounds: the invoice arrives where the shopkeeper already keeps their records, rather than in an email nobody opens or a paper copy that gets lost before the accountant sees it.

From there the pattern extends naturally, and these are the uses I would plan for in the same programme rather than as a second project:

Collections. Outstanding balance on request, a statement of open invoices, and a payment link against a specific document. Collections is a conversation distributors have constantly and awkwardly; doing it in the same thread as the order removes most of the awkwardness.
Reorder prompts. An outlet on a fortnightly cycle that has gone quiet gets a nudge with its usual list attached, ready to confirm. This is the highest-return outbound message in the channel.
Scheme announcements targeted to outlets that can actually qualify, rather than broadcast to everyone.
New product introduction with an image and a price — a listing conversation that currently requires a visit.
Complaints and service requests captured as cases rather than as messages someone promises to pass on.
Returns and damage claims, with a photo attached at the point of complaint, which is the evidence these disputes usually lack.
Delivery coordination — a heads-up before the vehicle arrives, and proof of delivery back into the thread.
Onboarding. A new outlet submits its details and registration documents through the same channel instead of waiting for a salesperson with a form.
Regional languages. The conversational mode's most underrated advantage in this market is that it does not require the shopkeeper to transact in English.
The primary leg. The same pattern works between distributor and brand, not only between retailer and distributor — which is where the order values are larger and the process is often still a phone call.

What has to be true underneath

The chat window is an interface. Whether it produces a real order or an expensive mess depends entirely on what it is talking to.

An order arriving through chat must price against the same customer-specific price list, apply the same trade schemes, check the same credit limit and allocate the same stock as an order taken by a rep in the shop or entered at the counter. If chat becomes its own order path with its own rules, you have not added a channel — you have added a reconciliation problem, and it will surface as disputed schemes at month end.

The same applies to identity. A message arrives from a phone number; the system needs to know which account that is, which price list they get, and whether they are credit-blocked. Getting that mapping right is unglamorous and it is what makes the rest safe.

The operational realities to plan for

Three things surprise people, and all three are better known before the pilot than during it.

Business messaging is a permissioned channel. Platforms require customer opt-in, and business-initiated messages must use pre-approved templates; free-form replies are only allowed inside an open service window after the customer writes to you. This shapes the whole design — reorder prompts and scheme announcements are template messages, and the rich conversation happens in the window the customer opens. Anyone promising unrestricted outbound messaging has not read the rules.

Know when to hand over. A channel that cannot escalate traps people. Any conversation that has gone twice around without resolving should reach a human with the transcript attached, not start again.

Somebody still owns the account. Chat ordering does not remove the salesperson; it removes the low-value order-taking visit and leaves the relationship visit. That distinction needs saying inside the sales team early, because the first assumption on the floor will be that this is a headcount exercise, and adoption dies quietly if the field believes that.

How to pilot it

Pick one distributor and their long tail — the outlets receiving a visit less than monthly. Start in guided mode with the top hundred SKUs, because it is cheap and it generates the transcript of what people actually ask. Measure three things: orders placed through the channel, the reduction in inbound status calls, and the order value versus the same outlets' previous average.

Then turn on conversational mode for the segment that has been asking questions the cards could not answer. By that point you will know who they are, because they will have tried.