A builder's merchant on the R44 out of Somerset West, twenty past seven on a Tuesday. The trade counter is already three deep. A plumber wants a case of 15mm copper, five metres of Cobra plate, and a proforma emailed to a GC in Roodepoort by ten because his job will not get paid without one. A subbie on an Airport Industria fit-out is on WhatsApp for the third time asking whether the AfriSam 42.5N is back in stock. A retail customer is holding a broken Everite ridge tile in one hand and a phone photo of his roof pitch in the other. The credit controller has printed a statement for a painter in Century City whose account went to sixty days last week and is asking the manager whether to release the R38,600 PO that just came through.
Anyone who has run a hardware store in South Africa knows this morning. The yard carries thousands of SKUs across cement, brick, timber, roofing, plumbing, electrical, ironmongery, paint, and tools. The customer base splits three ways — trade account holders on thirty days, cash tradesmen paying weekly, and DIY walk-ins on a Saturday. Each group asks different questions in a different tone on a different channel. Each expects an answer inside the hour.
AI does not mix concrete. It does not know whether a 6mm mild-steel plate cut to 1.2m will bear the load a customer describes in a badly-lit WhatsApp photo. What it does is take the paperwork tail off every transaction — the proformas, the "any stock?" pings, the statement chases, the delivery-ETA messages — and hand the counter back to the people who can actually help the tradesman in front of them.
Where the hours actually go on the trade counter
Not on the shop floor. That is where the margin sits, and interrupting the counter person mid-conversation with a tradesman is how sales get lost. The hours go into everything wrapped around the counter. Break a normal week down honestly.
- Trade-account queries. Statement chases, credit-limit questions, "please email me a copy of last month's", "why is this invoice on my statement, I returned it in April". Twenty a day at a mid-size branch.
- Stock queries by phone and WhatsApp. "Do you have any of the 90mm PVC in stock?" is the most repeated question in every SA hardware WhatsApp. Half the time the answer is on the POS. The other half means a walk to the bin.
- Proforma quotes for site orders. A tradesman submits a list, the counter person keys it into the POS, applies the trade tier, prices it, emails a PDF. Fifteen minutes each. Thirty on a heavy day.
- Delivery scheduling and site comms. A five-tonne cement load to a slab pour in Diep River at seven the next morning. Access, dropping point, who is on site to sign. Phones ring three times per delivery.
- Warranty and returns. A drill that packed up at week six, a bag of cement that "went off" in the yard, timber that arrived warped. The Consumer Protection Act, the manufacturer's warranty, and the store's own goodwill policy all pull in different directions.
None of it is hardware work. All of it is paperwork wrapped around the sale — the kind of structured, high-volume task AI handles reliably when the scope stays narrow.
Trade-account queries: the statement conveyor
The single most valuable place to put AI in an SA hardware store is the trade-account queue.
Every mid-size builder's merchant has one person, sometimes two, whose day is largely swallowed by statement queries. A quantity surveyor at a Sandton fit-out wants April's statement re-sent. A painter in Muizenberg wants to know why last week's return credit has not landed. A GC's admin lady in Randburg wants the balance broken down invoice-by-invoice before she pays. Every request means opening Pastel, exporting a PDF, checking whether the credit note has posted, drafting a courteous email, sending. Six minutes done well. Two hundred a week.
An AI-supported flow works differently. The account holder messages the store's WhatsApp or emails accounts@. The system verifies the sender against the account file, pulls the current statement from Pastel via a nightly export, and drafts a reply with the PDF attached plus a plain-language summary — "your balance at month-end is R47,283.12; the two open invoices are #24817 and #24902, both from 12 July." The credit controller reviews and clicks send. Ninety seconds per query.
Two things matter here, in my experience. The AI must never take a credit decision — never release a held PO, never authorise a top-up on a limit, never write "your account is in good standing" without a human eyeballing it first. And the drafted reply must always carry the actual PDF from the accounting system, not a summary the model has produced. Statements go to auditors. They need to be the real thing.
Stock queries: the "do you have any..." trap
The other high-volume WhatsApp category is stock. Most SA hardware stores have not built a live POS-to-WhatsApp link because the last quote came in at R48,000 and did not include the stocktake accuracy problem underneath.
AI does not solve the stocktake problem. It cannot tell a customer whether the 12mm rebar in bin 47 has actually been counted this week. What it can do is triage the query so the counter person only handles the ones that need them.
The pattern that works: a WhatsApp comes in, "do you have 90mm class-9 PVC in stock?". The system checks the POS SKU and, if the last export shows a positive figure, drafts a reply — "POS shows six lengths of 90mm class 9 in the yard as of yesterday's cutoff; happy to hold two for you if you are coming through this morning?" The counter person clicks send. If the SKU shows zero, or if the POS confidence is low (item has not been counted in six weeks, last stocktake flagged shrinkage), the AI escalates to the counter person immediately, marked "please verify — POS uncertain".
That triage changes the store's Monday. The counter person is not answering thirty pings on cement bags they already know the answer to. They are answering the four that need a walk to the bin.
Project quotes: draft, don't decide
Proformas for site orders are where hardware stores lose the most time to the least skilled work.
A subbie sends through a bill of quantities from a Camps Bay renovation. Nineteen line items, several vague — "6m of decent-quality skirting", "enough cement for a 40sqm slab, 100mm", "3 boxes of the usual screws". The counter person spends twenty-five minutes decoding, matching to SKUs, applying trade discount, and checking two items with the yard because the description is loose. The proforma goes out at half past four. The tradesman opens it at seven the next morning and asks for three changes.
An AI-assisted flow compresses the first pass. The bill of quantities lands on WhatsApp; the model parses it into structured line items, matches SKUs against the catalogue (this is worth the setup work — a mapping from tradesmen's loose vocabulary to real SKUs), and produces a first draft. Items the model was not confident about are marked "confirm at counter". The counter person reviews, walks to the yard for the two uncertain items, applies any relationship-based flex, and sends the proforma inside the hour.
The trick is what the model does not do. It does not commit the store to a price on an item it could not match, and it does not send anything to the customer directly. The output is a draft on the counter person's screen. The relationship stays between the counter and the tradesman.
Site delivery: the day the concrete goes down
Hardware deliveries are unforgiving. A slab pour in Constantia at seven the next morning does not tolerate a two-hour late arrival — the crew is on site, the pump is on site, and the whole day loses money if the cement truck goes to the wrong address.
A well-scoped delivery flow starts the day the order is placed. The customer gets a WhatsApp with three asks: the site address with a shared-location pin, the name and cell number of the person on site to sign, and any access details — a boom code, a security check-in, a keyholder. The morning of delivery, an automated message goes to that on-site contact at six with the driver's name, the truck's ETA window, and a live share-my-location link. When the driver arrives, they photograph the load against the delivery note before offloading, and the image is filed on the customer's record.
None of that is complicated. All of it is the difference between a driver reversing into the right driveway on the right morning and a slab pour that misses its window. On a busy day with fifteen deliveries, the automation saves the yard supervisor about ninety minutes of ETA phone calls.
Where AI has no business on your trade counter
A short honest list.
Anything credit-related. The National Credit Act is not friendly to automation that gets it wrong. A credit-limit increase, a release of a held PO, an account opening for a new contractor — each is a human decision, made with the FICA file open and the trade references verified.
Load-out advice. A retail customer at the counter with a photo of his warped rafter, asking what he should buy — a bot will guess plausibly and often wrongly, and the customer will come back angry when the fix does not fix. That is a conversation for the counter person who has seen a hundred of those rafters.
Warranty judgement calls. A drill that failed at week six sits on the border between manufacturer defect, mishandling on site, and the store's own goodwill. That is a manager's call, made with the drill on the counter in front of them.
The relationships. The GC in Diep River who has spent R2.1m at your store this financial year deserves your manager's voice on the phone, not a WhatsApp template. The plumber who has bought from your shop for eighteen years gets the extra ten percent because he asked politely, not because a system flagged him for retention.
Most vendors will not tell you any of this. They will sell you a full "hardware AI concierge" that fails quietly at the exact moments the store's reputation is being made.
Where to start
The temptation with a project like this is to automate the whole trade counter at once. The stores that get real value do the opposite — they pick the single loop that hurts the most, build it properly, and leave the rest of the counter alone until that loop is quietly working for a month.
For most SA hardware stores and builder's merchants, that loop is trade-account queries. Not because it is glamorous. Because it removes a two-hundred-a-week task from the credit controller's inbox without touching anything commercially risky, it pays for itself inside a quarter, and once your Pastel or IQ Retail sync is running for statements, the same plumbing extends naturally to the proforma workflow and the delivery loop.
Hardware retailers across South Africa who have run this first loop report the same change. The credit controller stops working through their lunch hour on statement PDFs. And the manager stops fielding the third statement query on the walk from the yard to the office, which is where the store actually makes its money.