AI-01 · AI case study · Order management · FMCG distribution

A WhatsApp agent that takes shop orders between salesman visits

Concept case study · 2026 · Based on my order management work on a DMS rollout

ConceptA design based on my project experience, written during my 2026 sabbatical. It is not a deployed client solution.

24/7Shops can order at any time
Text or voiceOrders in the shop owner's own words
DMS-checkedPrice, scheme and stock confirmed
Salesman in the loopUnclear orders go to a person

The context

In general trade, most small shops order only when the salesman visits, often once a week. Between visits, urgent orders come by phone and are typed in by hand, or they don't happen at all and the sale goes to a competitor.

Most shop owners already use WhatsApp every day. This concept builds on my order management requirements work for a DMS rollout and designs an agent that takes orders on WhatsApp and checks them against the DMS before they reach the distributor.

The problem

Ordering is tied to the salesman's visit, and everything outside it is manual:

  • Shops can only place a proper order when the salesman visits.
  • Phone and paper orders are re-keyed by hand, with mistakes in products and quantities.
  • Shops don't see the final price, scheme or stock position until delivery.
  • Salesmen spend visit time writing routine orders instead of selling.

What I did

  • Mapped the current order journey from my DMS project experience.
  • Designed the conversation flow and the hand-offs between the agent, the DMS and the salesman.
  • Defined the rules the agent must follow for prices, schemes, credit limits and stock.
  • Wrote user stories with Gherkin acceptance criteria and a sample conversation script.

The process, before and after

Each column is a team or system. Numbered badges point to where time is lost today, and how the design saves it.

Scroll sideways to see the whole diagram →

Shop ordering before: tied to the salesman visitShopSalesmanDistributorDMSShop needsstockWait for thenext visit1Write orderon the appRe-key phoneand paper orders2Create orderand invoiceOrderdelivered
  1. 1Shops can only place a proper order when the salesman visits, often once a week.
  2. 2Urgent orders arrive by phone and are typed in by hand, with mistakes.
Start Step Decision End Each column is one team or system. Arrows show hand-offs.

What changed, step by step

StepBeforeAfter
Placing an orderOnly during the salesman's visit, or by phoneAny time on WhatsApp
Entering the orderRe-keyed by handCreated in the DMS from the confirmed chat
Price and schemeSeen at deliveryShown before the shop confirms
Unclear ordersSorted out at deliverySent to the salesman straight away

The approach

  1. 1
    Start from the shop's habitsShop owners already message on WhatsApp, in their own words and language. The agent works with that instead of asking them to install an app.
  2. 2
    Match free text to productsThe agent maps phrases like 'two cases of the 1 litre pack' to product codes, using the shop's past orders to resolve doubts.
  3. 3
    Let the DMS decide pricesPrices, schemes and credit limits come from the DMS, never from the agent, so the total matches the invoice.
  4. 4
    Confirm before orderingThe agent sends a summary with the total and any scheme applied, and creates the order only after the shop confirms.
  5. 5
    Keep the salesman in the loopUnclear orders, credit holds and complaints go to the salesman, who also sees every WhatsApp order on his beat.
  6. 6
    Measure what mattersTrack orders placed between visits, order accuracy and the share of orders that need a person.

Sample work: conversation design extract

StepShop ownerAgentBehind the scenes
1Send 2 cases sunflower oil 1L and 5 packs biscuitsLists the oil it matched and asks which biscuit packMatches products using the shop's recent orders
2Glucose biscuits, 100 gShows the total with the scheme applied: buy 5, get 1 freeGets price, scheme and stock from the DMS
3OKConfirms the order and the delivery dayCreates the order for the distributor
4Can I pay next week?Says the salesman will call about paymentCreates a task for the salesman

Illustrative script written for this portfolio. Products are generic.

Expected outcomes

  • More orders between salesman visits, without adding visits.
  • Fewer order errors, because nothing is re-keyed by hand.
  • Shops see the right price and scheme before they confirm.
  • Salesmen spend more of each visit on selling and display.

Design principles

  • The agent drafts; the DMS decides. Prices and schemes never come from the agent.
  • Always confirm before creating an order.
  • Hand off early. A wrong order costs more than a slow one.

Concept case study written during my 2026 sabbatical. It is not a deployed client solution. Examples are generic.