A one-brewer, three-staff outfit off Albert Road in Woodstock. Friday, ten past three, taproom opens at four. The head brewer is sanitising the mash tun after tomorrow's brew day, which starts at half past five because the double IPA takes two hours to hit temperature on the direct-fire kettle. His phone has thirty-one WhatsApp messages queued since yesterday afternoon. Eleven of them are from customers asking whether the Table Bay Pale is on tap for the weekend (it is; the pallet came back from the co-packer on Wednesday). Six are booking enquiries, three of them for parties of eight or more, which need the standing table and a confirmed pizza order from the vendor next door. Four are trade. A Constantia bottle store asking if the ten-case order made it onto Monday's delivery; a Gardens gastro-pub asking whether the sour has come off tank yet; a Sea Point liquor store confirming they want an extra six-pack case of the lager for this weekend's rugby; and one from a new bar in Salt River that has said hello twice and is waiting. The rest are the usual: one supplier chasing malt payment, one former colleague, three photos of foam his sales rep is claiming as proof of clean pours in Newlands.
None of this is hard on its own. All of it, on a Friday, on top of a brew day tomorrow, is a lot.
That is what small-brewer life in South Africa looks like. Not the big-tank breweries at scale, not the once-a-fortnight craft taps you find in most restaurants. The ones under 5,000 hectolitres a year who make the beer, run the taproom, invoice the trade and drive their own kegs on a Wednesday afternoon. There are perhaps eighty of these in the country, from Cape Town round to the Midlands. This is where AI actually pays back, if you put it in the right places.
The three jobs a small SA brewer is doing at once
Every independent craft brewery in the country is running three businesses on the same balance sheet. Production, which is capital-intensive and mostly non-negotiable in terms of who does it (the brewer, and only the brewer, ends up on a mash tun at six in the morning). A retail operation, in the form of the taproom, running one or two lively nights a week with a Sunday-afternoon session for families. And a wholesale operation, chasing trade accounts across the metro, plus the odd out-of-town order that has to be palletised and put on a courier.
The revenue mix has moved noticeably in the last two years. Taproom direct sale is the highest-margin channel. Distribution to on-trade sits second. Bottle stores and the Takealot listing sit third. In practice, most brewers I have looked at get somewhere between 35 and 60 per cent of their gross from the taproom, and the temptation to grow it further collides with the fact that the taproom needs weekend evenings the head brewer would sooner spend not working.
That mix decides what AI is worth doing and what it is not. There is no automation for pouring a mash tun. There is significant, well-scoped automation available for the WhatsApp queue.
Taproom bookings, growlers, and Friday-afternoon triage
The single highest-volume repeat task is the Friday-afternoon booking queue. Groups of six-plus who want a table, mixed with growler-fill enquiries, mixed with people asking whether the food truck is coming this weekend (yes, both nights), mixed with people asking whether they can bring a dog (usually yes, on the outside deck, not inside). The right answer to almost all of these is a two-line reply and, for the group bookings, a card-link to hold the table with a per-head minimum spend.
An AI helper that reads the incoming WhatsApp, checks the calendar (or the paper list stuck to the fridge behind the till), drafts a reply in the brewery's voice, pre-fills the card-link where relevant, and hands it to whoever is on the taproom shift for a review click. Not autonomous. Autonomy fails on this specific problem because the taproom regulars — the ones who come in twice a week and ordered the entire second batch of the Milk Stout when the fermenter dropped a chill haze — recognise a script within one message, and losing that trust costs more than the automation saves in six months.
A specific note on growler and squealer fills. The reply flow needs to know whether the fill is a taproom collection or a licensed delivery, and the draft should hold on the delivery flow if the address is a residential one in a zone where the brewery does not itself hold delivery rights. This is the small print that trips most generic booking bots and it is worth wiring correctly from day one.
Trade orders, and the case for a lightweight trade WhatsApp
The wholesale side is where AI earns a different kind of keep. Trade orders — the case-and-a-half from a Gardens gastro-pub, the six kegs to Roeland Street, the pallet to Franschhoek — come in on WhatsApp, on email, and sometimes on a scribbled slip from a rep's Wednesday visit. Very few small breweries have a proper trade portal. Vinocurrent-style tools are priced for wine and most brewers cannot justify a full ERP for eighty accounts.
What actually works: a small trade WhatsApp number, separate from the general one, that accepts an order in plain English, drafts a proforma referencing the specific SKUs and current pricing, checks stock against a sheet the brewer updates weekly, and hands the draft to the salesperson for a review click. The order lands in Xero or Sage One as a draft invoice. The delivery gets slotted onto whichever van run covers that area.
Two things the flow must get right. It must confirm the buyer is trade-licensed (a bottle store, a licensed on-trade venue) and refuse residential wholesale, which is a WCLA problem waiting to happen. And it must hold on any credit-line question. A new trade account without a set credit limit is a decision the founder makes with the salesperson looking at the actual figures, not a message a model handles at eleven at night.
The distribution reality small brewers do not talk about
This is the section where the honest paragraph has to go. Most SA independents are not competing against other independents for shelf space. They are competing against SAB InBev's rep network and Heineken's on-trade contracts. The trade WhatsApp will not fix that. What it will fix is the specific administrative slowness that costs an independent the third case in a month, because the bar manager could not reach the brewer on Sunday to place a top-up order and went with the SAB rep who was there on Monday.
Fixing that is a real number. On the trade accounts we have wired properly, average order frequency has lifted by roughly a fifth inside three months. Not because the automation makes anyone buy more beer, but because it removes the friction that quietly cost the reorder.
Where AI does not help. If the big-brewery reps land in your CBD accounts with a rebate structure you cannot match, no WhatsApp automation is going to close that gap. The right call there is a targeted taproom investment or a strategic on-trade deal, not more distribution admin.
WCLA, POPIA, and the parts most brewery bots get wrong
Any SA brewery selling out of a taproom is licensed. The Western Cape Liquor Authority — and the equivalents in Gauteng, KZN and the Eastern Cape — audit, and the fines for a mistaken online sale to an unlicensed address are not trivial. Two things any AI flow touching a purchase has to do.
Confirm age at the door of the flow, in the way the platform actually accepts. A checkbox is not enough. A captured self-declaration with a timestamped record on the order is the baseline, and for delivery orders the driver still checks ID on handover. Confirm that delivery routes through a licensed fulfilment channel, not directly from a chatbot to a residential doorstep, unless the brewery holds the delivery licence for that address zone. Almost no small brewery does. Most partner with a licensed courier who does, and the AI flow needs to know which zones the courier covers.
On POPIA. The taproom mailing list — the one people give an email address to when they buy a taster paddle — is a POPIA responsible-party dataset the same way a Sandton law firm's client list is. Consent for marketing is separate from consent to be booked for a tasting. Every automated release-email flow needs a clear opt-out in the same language it was sent in. Deletion has to mean deletion, not a hidden archive flag on the CRM.
What not to hand to a model
The taster-paddle conversation with a new customer. The reason someone stays a regular is the ten-second story the pourer tells about the batch — the hop that arrived late from George, the reason the ale is a little maltier this month. A model that produces a competent paragraph on the porter is a model that has quietly removed the reason people come back.
The recipe question from a home brewer who wants to know your mash temperature for the Saison. The brewer answers that, in his own voice, in his own time, or does not. It is not a customer-service ticket and it will not be improved by templating.
Any QC concern from a customer. A beer that arrived hazy, an off-flavour, a leaking can. That escalates to the head brewer in writing immediately. A polite AI apology in that context reads as a cover-up on a Friday and does more damage than the original problem.
The sales call to a new venue. The rep drives out to a place in Woodstock, meets the manager, tastes the tap he wants placed. That is a conversation about a relationship, not a message that gets drafted for review.
Where to start
One thing.
For every small SA craft brewery I have looked at, the first project is the taproom-booking WhatsApp on the main direct channel. Wire it so a new enquiry produces a drafted reply, in the brewery's voice, with correct availability, food-truck details, and where relevant a card-hold link, in under ninety seconds. Use it on a busy Friday. See what breaks. If nothing breaks in a fortnight, run it as standard.
The trade WhatsApp second. It is quieter, higher-stakes, and it moves a real number on reorder frequency inside a quarter.
The release-note mailing list third, with the same caveat that applies to any content the brewer's voice is doing the heavy lifting on. The draft comes from a model. The final paragraph on why the hopped-saison tastes like it does comes from the brewer, or it is not worth sending.
None of this changes what a small brewery is. It gives the head brewer a Friday afternoon where the thirty-one WhatsApp messages are drafted before doors open at four, and a Saturday morning where the mash-in at half past five is not shadowed by an inbox that needed him at ten the night before. That is what the automation is for.