A Thursday morning in October, at a Woodstock catering kitchen that does maybe 180 events a year. The head chef is prepping a canape platter for a book launch in Bo-Kaap that night. The owner is on his second coffee with eleven quote requests waiting. A Newlands 21st for 80. A corporate breakfast off Adderley Street. A March wedding on a farm outside Franschhoek. A Sunday memorial in Athlone, halaal. A children’s birthday, gluten-free, no nuts. Three came in on Sunday evening. Two on Monday. The rest since Wednesday. The oldest is nearly a hundred hours old.

He knows, without looking at the numbers, that at least four of those enquirers have already gone with someone else.

This is the shape of SA catering outside chain-hotel banqueting. A small kitchen, a talented team, a client list built on referral, and an admin flow that lives in an inbox, a WhatsApp Business number, a Google Sheet of past events, and the owner’s memory. When enquiry volume is normal, it works. When peak season hits — matric farewells in October, weddings November through March, end-of-year corporates in December — it stops working, and the caterer loses either bookings or evenings.

Why speed matters more than price for a catering enquiry

I’ve sat with enough catering owners now to know a specific pattern holds. On a corporate or private-event enquiry sent to two or three caterers, the one that replies first with a coherent proposal wins about two-thirds of the time. Not the cheapest. The fastest with something the client can act on.

The reason is simple. The person sending the enquiry is usually an EA, an office manager, a mother of the bride, or a father organising a memorial. They are stressed, working under a deadline, and looking for a caterer they can stop worrying about. The first proposal that arrives, if it reads professionally, becomes the anchor. Every later quote gets compared to it.

Caterers who understand this instinctively still lose bookings during peak season, because the physical constraint is real: the owner cannot be chef at 10 am, quote writer at 2 pm, run-sheet builder at 6 pm, and also reply to a Sunday-evening enquiry before Monday afternoon. Something gives. Historically it has been the enquiry.

This is where AI has a clear job.

The intake form that changes everything

Before touching AI, most catering companies need one thing: an intake form that captures the fields a quote actually depends on. In my experience, the biggest reason quotes take two days is not the writing. It is the four rounds of WhatsApp back-and-forth to establish the event is a seated dinner for 55, not a canape reception for 80, and yes the venue has a proper kitchen, and no the client does not want plated dessert.

The form is not complicated. Nine fields:

Put the link on Instagram bio, on the website, and inside your WhatsApp Business autoresponder. A meaningful share of enquirers will fill it. The rest still message you the old way. That is fine. You have moved from zero structured data to sixty percent structured, and the sixty percent is the part where AI helps.

Quote drafting from the intake form

Once the form data lands, an AI-drafted proposal is a well-scoped engineering problem. Feed the form fields plus your last thirty priced events into a model, get back a draft proposal in your voice, with a menu suggestion, suggested per-head price, staff estimate, hire list if the venue is bare, and a note on which parts need a human eye.

Draft lands in the owner’s inbox or a WhatsApp review queue. Ninety seconds to read. Two minutes to adjust the menu. One click to send as a branded PDF proposal.

A quote that used to take an hour takes five minutes. The Sunday-night enquiry gets a Monday 07:30 reply. That reply closes the booking.

What the AI must not do: send the quote automatically. Every catering owner I’ve worked with who tried auto-send pulled it inside a month, because pricing has judgement in it. A Franschhoek destination wedding in March is priced differently from a Newlands corporate breakfast, and the difference is not just per-head. It is delivery distance, staff overnight, hire logistics, whether the venue takes commission, whether it is a repeat client. That judgement stays with the owner. The drafting does not.

Menu customisation that stays coherent

Dietary requirements in SA catering have got more complex over the last five years. Halaal, kosher, coeliac, vegan, no shellfish, no nuts, no dairy, low-FODMAP, banting. A wedding for 120 might have 4 vegan, 3 halaal, 2 coeliac, 1 nut-allergic, 6 vegetarian, and the couple wants none of that flagged on the printed menu so guests feel equally catered for.

Building a coherent menu that hits all of it without producing eight separate dishes at plating is a skill. Not one AI has by itself. What AI can do usefully is take the intake requirements and the caterer’s active menu library, then produce a first-cut menu that respects the constraints and flags where one swap covers multiple diets (a coconut-milk curry is halaal, vegan, dairy-free and nut-free at once, if the sambals are right).

Chef reads the draft, adjusts, and the version that reaches the client is coherent and buildable. An hour saved per event on menu design. Over a busy November, that’s a working day back.

Where I’ve seen this fail: caterers who let the AI pick from a static menu PDF instead of the real current inventory. If the AI does not know the langoustines are off the market this week and the burrata supplier is on leave, it will happily draft a menu you cannot execute. Feed it what the kitchen actually has, updated weekly. Otherwise skip this and stick to intake-and-quote.

Run sheets and kitchen briefs

The document that keeps a catering event on the rails is the run sheet. Arrival time, delivery van, setup and service crews, chef in charge, menu with allergens, dietary counts, service times, breakdown, client contact on the day. Most kitchens still build this in Word or Excel the day before, copying fields from the accepted proposal.

That copy step is where AI is at its most reliable. A model given the accepted proposal, the venue notes, the crew roster and the standard run-sheet template produces a first-draft that is 90 percent correct. Chef and ops lead read it once, fix the 10 percent, and it goes on the WhatsApp group by Thursday for a Saturday event. Two hours of Friday-afternoon admin, gone.

Event-day WhatsApp: careful, useful, limited

On the day of an event, WhatsApp is where everything happens. The client asks about arrival time. The venue asks whether the bar hire fits their loading bay. The florist asks if there is a fridge for the aisle petals.

AI has a narrow, useful role here. A WhatsApp assistant that answers the fifth-time-repeat questions — arrival time, dietary counts confirmed, dress code for staff, whether the caterer brings ice — drawn straight from the accepted proposal. Client sees an answer inside a minute. Owner sees a log. If the client asks something the assistant is not confident about, it flags to the owner and says exactly that: Let me check with the team and come back shortly. No hallucinated menu items. No made-up delivery times.

The part AI must not attempt: anything that changes the event on the day. Not a menu swap. Not a guest-count change after 48 hours. Not a refund promise. Not any conversation with a venue coordinator whose relationship with the caterer is the whole reason you got booked. That stays human, preferably one specific human.

I saw a WhatsApp bot at a caterer confirm a headcount increase from 80 to 110 the day before a plated dinner, because it was set to auto-acknowledge. The kitchen did not know. The event ran short of mains. The client left a review that took the caterer months to recover from. Do not let a bot do headcount.

Corporate accounts and repeat bookings

A meaningful slice of a mature SA catering business is repeat corporate work. The advocates’ chambers ordering sandwiches every Thursday. The law firm whose partner meetings run on platters. The advertising agency doing a monthly all-hands lunch. These clients do not want a full proposal every time. They want to WhatsApp “Same as last week for Thursday, headcount 24” and have it happen.

AI here is a light layer that reads the message, pulls the standing order, checks Thursday capacity, drafts a confirmation the owner reviews and sends. Monthly-invoice reconciliation against EFT payments — matching the reference to the invoice number, flagging the wrong ones — saves an hour a month per corporate account.

POPIA note that matters. The intake form and WhatsApp assistant both collect personal data (contacts, dietary requirements, sometimes children’s ages for kids’ parties). Two things to get right. Explicit consent at intake with a plain-English line about what you keep and how deletion works. And a retention rule that actually deletes event data after, say, 24 months unless the client is a live repeat account. Any AI provider that cannot tell you where their data lives is the wrong provider for a business holding dietary records for children.

A realistic build order

Setting up automation from scratch for a small-to-mid SA caterer today, the order would be simple.

First, the intake form and the WhatsApp autoresponder pointing to it. No AI needed. Two days of work, and structured enquiry data starts flowing in.

Second, the AI-drafted proposal on top of the form and the last thirty priced events. Two to three weeks of build. About ZAR 15,000 to ZAR 35,000 depending on the shape of your past-event data. Payback in one peak-season week.

Third, run-sheet generation from the accepted proposal. One week of build, once the proposal piece is live.

Fourth, only if you have real corporate repeat volume, the standing-order layer and the reconciliation piece. The event-day WhatsApp assistant is the last one I would build, and only for opted-in clients. Highest risk if it misfires, smallest reward per event.

What I would not build: an AI menu-designer running independently of the chef, a booking bot taking deposits without a human confirming the date, or any auto-poster to Instagram. Catering runs on a specific voice and a specific reputation. Those are the parts a bot cannot carry.

Caterers who’ve implemented in roughly this order in Cape Town, Joburg and the Winelands report the same pattern. Quote turnaround from 36 hours to under 6. Peak-season booking rate up 15 to 25 percent because the fast reply wins the client. Kitchen runs Saturday on a run sheet built Thursday, not scrambled on Friday night. Nothing about the food changes. That’s the point.