First Monday of the month, 08:20, at a twenty-two-person wholesale distributor off Marconi Beam Road in Montague Gardens. Six reps have already checked in from the road — one in Piketberg, one in Vredenburg, two working the Athlone and Mitchells Plain corridor, one in Worcester, one halfway to Beaufort West who won't be back on Wi-Fi until Wednesday. By lunchtime, forty new orders will land back at the office as WhatsApp voice notes, photographs of a duplicate order book, and the occasional typed message from the two reps who bought themselves a Bluetooth keyboard. The bookkeeper is going to type all of them into Sage. She will get halfway through, then stop to answer a Spaza owner asking why his invoice from three weeks ago has a R1,240 line he did not order.
Anyone who has run an SA distributor recognises this Monday. The reps are the business. They are also the bottleneck. Every order from the field has to be typed, priced against the customer's price list, checked against stock, invoiced, and packed for the next delivery run. Every rep drives home dreading the call report their sales manager will ask for.
AI is not going to replace the rep. In distribution the rep is exactly the thing that cannot be replaced — the person who walks into the Somerset West Spar on a Wednesday and knows the receiver only signs for pallets before 10:00, that the perishables buyer needs the seasonal listing conversation before the store manager will approve it, and that if the buyer at the family-owned Cape Flats wholesaler is having a rough month you skip the up-sell and take the standing order. That knowledge does not sit in a system. It sits in the rep's head.
What AI earns its place doing is everything around the rep. The voice note that becomes a Sage order without anyone typing. The call report that writes itself from the day's messages. The trade reminder in the buyer's voice, not a template.
Where a distributor's week actually goes
Not on selling. That is what the reps do. The office hours — the bookkeeper's, the sales manager's, the two admin people at the front — live somewhere else entirely.
- Order capture from the field. Voice notes, photographed order books, WhatsApp lists, the occasional email. Each one has to become a Sage or Pastel sales order at the customer's contracted price. Fifteen to twenty hours a week at a mid-sized distributor.
- Call reports and journey plans. The sales manager wants to know who was visited and what was discussed. The reps hate writing them. Reports arrive late and thin, or not at all.
- Trade-account queries. "Why is my statement R1,240 higher?" "Did you deliver the two cases of Mrs Ball's?" Half the front-office phone is this.
- Debtors. Thirty-day accounts that quietly become sixty-day, and the credit controller's Friday-afternoon problem.
- Returns and credits. The messiest workflow in distribution. The rep says the customer wants to return the case. The driver brings it back. Everyone forgets whose job it was to raise the credit.
None of this is selling. All of it eats the office. The reps stay in the field, the front office chases paperwork, and the sales manager writes a Sunday-night WhatsApp asking where everyone's Wednesday reports are.
Order capture: from voice note to Sage in two minutes
The highest-value place to start at an SA distributor is order capture. Every rep is already sending order voice notes into a shared WhatsApp group, or into the sales admin's personal number. The distance between "voice note lands" and "Sage sales order at the right price for the right customer" is where the office loses its Mondays.
A properly wired capture loop closes that gap. The rep sends the voice note against a specific customer — by tagging the customer's WhatsApp thread or naming the customer at the top of the note. The AI transcribes, extracts each line item against your product catalogue (matching brand names, pack sizes and case counts the way a human would), applies the customer's contracted price list, checks live stock, and drafts the Sage sales order. The admin gets a single review screen with the source voice note on the left and the drafted order on the right. She confirms, or she edits and confirms. Two minutes per order instead of ten.
Three details matter, in my experience, before this is worth switching on. The catalogue matching has to be good enough that "the 6-pack Robertsons black pepper" reliably resolves to the right SKU without the admin fishing for it. If the AI matches 70% of lines cleanly it is worse than useless, because the admin has to check every line anyway. Threshold is around 95%. Below that, do not ship.
Price must come from the customer's contracted price list in Sage, never from a "typical" price. Distribution sits or falls on the fact that different accounts have different prices, negotiated over years. A bot that quotes a different customer's price has just cost you a relationship.
The admin stays in the loop on every order. Not because the AI is not good enough, but because her eye catches things the model cannot: the buyer who always double-orders on the last Monday of the month and needs a phone call before you invoice, the customer whose account is on stop but whose voice note came through anyway.
Call reports that write themselves
The rep's call report is a solved problem, and most distributors have not noticed. Every rep is already narrating their day into WhatsApp: the Wynberg SPAR visit at 09:30, the buyer meeting at Food Lover's at 11:15, the Kraaifontein trader drop-in at 14:00. That narration — voice notes, customer replies, shelf-gap photos, "buyer wants a callback Friday" — is the report. Nobody needs to write it a second time.
The version that works is a nightly draft. At 18:00 the system takes the day's messages against each customer, drafts a two-line summary per visit (who was seen, what was said, what needs to happen next), and puts the whole thing in front of the rep as one WhatsApp message with an "approve" button. If the rep needs to add colour, they voice-note back and it gets folded in. The approved report lands in the sales manager's inbox by 20:00. One minute of rep time instead of an hour.
The knock-on effect is what makes it worth doing. When the sales manager can see who was visited last week, she can journey-plan against real data. When a rep leaves, the customer history is not walking out with them — it sits in the customer's file, ready for the next rep.
Trade-account comms in the buyer's voice
Wholesale distribution is a small-community business. In Cape Town, the same twenty buyers cycle through the same eight distributors over the course of a career. A statement reminder that reads like a bank collection letter will lose you an account faster than an actual bad debt would.
What works is the reminder that sounds like the credit controller herself. She has been dealing with the buyer at that Wynberg supermarket for seven years. She knows he prefers a phone call before an email, that his statements go to his personal Gmail because the shop email is chaos, and that he settles on the 27th when the shop's cash-up clears. "Morning Dennis. Your July statement went to your Gmail this morning — R48,270 on the 30-day, all good for a 27 August settlement. Give me a shout if you need me to resend anything. — Riana."
That reminder is drafted by AI. Riana reads it in five seconds and clicks send. She sends thirty in the time it used to take her to write two. Because every reminder went out in her tone, not a system's, none of her customers noticed anything changed.
The same logic applies to trade queries. "Why is my statement R1,240 higher?" gets a drafted answer against the actual account movement — a return that never landed, a delivery on 3 August, a small credit raised but not yet applied — and the credit controller sends it inside a minute. She stopped playing hide-and-seek with the OneDrive folder to find the delivery note.
What the bot must never do
A short list.
Change a rep's price. If a rep has agreed a special on twelve cases of Mrs Ball's for a customer for the quarter, that price sits behind the rep's judgement of the account. A bot that auto-adjusts because inventory is short, or a national list moved, is undermining the rep in front of the buyer. Price changes go through the sales manager, always.
Credit a return. The credit is where the money actually leaves. A bot that raises a credit because a voice note said "the customer sent back the case" is generating shrinkage. The credit controller signs off on every credit, or you find out in the year-end audit that you gave away six months of margin without noticing.
Sign off on a listing. New listings at a chain, or a large independent, are commercial negotiations — pricing, listing fees, promotional support, delivery terms. A bot has no place in any conversation whose outcome is a signed trading agreement. That is the sales manager or the MD, on a call, with the buyer.
Anything on stopped accounts. If an account is on stop, the credit controller decides what happens next. Not a template. Not an auto-reply.
Where to start
The temptation at a distributor is to buy a rep-management platform and roll it out to everyone in one go. This never works. Reps either resist the app because it is one more thing to update while driving, or they nominally adopt it and then quietly go back to WhatsApp within a fortnight.
The distributors that get real value do the opposite. They pick the one loop that costs the front office the most hours — almost always voice-note-to-Sage order capture — and wire it against the tools the reps already use, meaning WhatsApp. No new app for the rep. No new login. The office picks up the drafted order in Sage and confirms it. That is the whole change.
Give that loop a month. Measure it against the Monday it was designed to fix: order-hours the bookkeeper spent typing, mispriced invoices, customer calls about "why is my statement wrong". If those numbers move, add the call-report layer. If the call reports get honest, add the trade-comms drafting. Each layer earns its place on the one before it, not on a promise.
The small-to-mid SA distributors that build order capture first report the same shift. Monday stops being a scramble. The reps stay out longer because the office is not phoning at 11:00 to double-check what they meant by "six of the small Milo". And the sales manager gets Wednesday-morning journey plans built off Tuesday-night reality, not off a spreadsheet she updated in her head over the weekend.