Last Tuesday a procurement manager in Rosebank had six pitch decks on her desk and a budget of R420,000 for a twelve-month automation build. Three decks were from Sandton consultancies with Fortune 500 logos down the right-hand margin. One came from a Fourways founder who had built the demo himself. One from a Durban shop. One from a Cape Town studio that had shipped the same build eight times. By Friday she had to pick one.
She did what most buyers do at that point. She called me because her auditor had mentioned our name, asked for an outside read, and ran through what each pitch had promised. Three were lying. Not maliciously, in the way a sales deck lies when a working engineer never wrote it. One was close to right but overpriced for her problem. Two were honest fits for different sides of her business. She still had to pick one, and the metric most buying guides tell you to use was not going to help her.
That is the problem with picking an automation partner in Johannesburg. The city has more vendors than the rest of the country combined. The pitches blur. The budgets are large enough to attract people selling nothing in particular. And the thing you are actually buying, a working automation that will still be running in eighteen months, is not what the demo is showing you.
What Joburg buyers get wrong that coastal buyers do not
Working with businesses in all three major SA cities for the last few years, a pattern repeats itself. Johannesburg buyers come to the vendor conversation with more procurement reflex than their Cape Town or Durban counterparts. The RFP is longer. The scorecard is spread across more criteria. There is often a board paper in the back that nobody will ever read again. None of this is bad. The problem is that none of it predicts whether the automation will actually work.
A Cape Town buyer will usually spend three meetings on scope and sign on a one-page email. A Durban buyer will try to pay in cash. A Sandton buyer will send a forty-page tender response template designed in 2012 for buying industrial compressors. The vendor who wins the Sandton tender is not necessarily the vendor whose builds last. It is the vendor whose proposal-writer plays the tender well. Those are rarely the same person.
The second thing Joburg buyers do that costs them money is buying on logo weight. A consultancy that worked with a listed insurer or one of the big-four banks must know what they are doing. In my experience that is wrong more often than it is right. A junior shipped the big-bank project. The partner on your call has not personally built an automation in four years. The retainer rate quoted assumes the same senior team you will not actually see after sign-off. The right question is not whose logo is on the slide. It is who writes the code when the deadline slips in month five.
Three partner profiles, three different Joburg businesses
Most Joburg automation buyers fall into one of three groups, and the partner that fits each one looks different.
The first group is the Sandton, Rosebank or Melrose Arch mid-corporate. Eighty to three hundred staff, a real IT function, a procurement team, an auditor who will ask about data lineage. These businesses need a partner who can sit in an audit meeting and answer POPIA questions without flinching, write proper documentation, and run a change-management rhythm their IT director recognises. The partner who runs their work out of a single Discord server will not survive the first audit cycle. Equally, the Big Four consulting arm will overrun the budget by a factor of four and ship something built by three contractors who left halfway through.
The second group is the Fourways, Bryanston, Woodmead or Rivonia owner-managed SME. Ten to fifty staff, the founder still signs every invoice, no in-house IT. These businesses need a partner who will build the thing, train the office manager to run it, and respond on WhatsApp when something breaks on a Friday evening. They do not need a procurement process. They need somebody they can phone. A R2m consultancy proposal with a steering committee and a Jira project is wrong for this buyer, whatever the Sandton vendor says.
The third group is the Isando, Jet Park, Midrand or East Rand manufacturer or distributor. Fifty to five hundred staff, often a second-generation family business, legacy systems the vendor must plug into without breaking. These buyers need somebody who understands Sage Evolution, Pastel Partner, or the older SAP B1 instances the business actually runs on. The Rosebank studio that only knows Shopify and HubSpot is the wrong fit here, however slick the deck.
The first mistake almost every Joburg buyer makes is picking a vendor from one group to serve a business in another.
Five questions that cut through the pitch
Most RFP scorecards weight the wrong things. These five questions, in our experience, actually predict whether an SA automation partner will deliver.
- Who personally will write the code? Not the team structure. The name. If the partner on the call cannot answer that, or the name changes between the proposal and kickoff, you are buying a bait-and-switch.
- Walk me through a build you shipped eighteen months ago that is still running. Not a case study. The actual client, the actual system, what has gone wrong since launch, and who fixed it. A vendor who cannot do this on the first call has either never shipped something that lasted that long, or they are not comfortable with you phoning the reference.
- What will you refuse to automate? The honest answer names something specific. "We will not touch client-facing credit decisions for FAIS reasons." "We do not write SARS submission bots because the Commissioner has not blessed one and we do not want our client in the Voluntary Disclosure queue." A vendor who says "we can automate anything" is selling you the demo, not the production system.
- What happens on the day somebody files a POPIA complaint against the workflow you build us? Who gets the Information Regulator letter. Who sits in the meeting. If the answer is "that is your problem", that is a defensible position, but you need to know going in.
- Price the first three months separately from the next twelve. A vendor who will not quote the build without an eighteen-month retainer is either covering a thin build budget or they intend to charge you R35,000 a month in year two for login support. Both are expensive answers.
Red flags most RFPs do not score
A handful that have cost SA buyers real money this year.
The team page shows ten faces. The LinkedIn trace shows two of them actually work there. We check this before every subcontract. You should too.
The pricing is a single flat monthly fee that covers "unlimited support." It never does. By month four they are either angling for scope reduction or they go quiet. Either way the build drifts.
Every answer about data residency starts with "well, actually AWS Cape Town" and ends with the data sitting in Virginia. If your workflow touches employee or customer records that fall under POPIA Section 72, the honest answer either names a South African data centre or walks you through the cross-border transfer paperwork. The hand-wave is the warning.
They demo the chatbot. They do not show you the admin console, the error logs, the retry queues, or the dashboard the operations manager will live in for the next three years. A production system is fifteen percent what the customer sees and eighty-five percent what the operator sees. A vendor who only shows the customer screen has only built the customer screen.
The lead developer is "offshore." Sometimes that is fine. Often it is not. Ask who in SA is accountable, in which time zone, and under which company's professional indemnity cover.
Where locality actually matters, and where it does not
Being a Johannesburg-based vendor is not, on its own, worth paying extra for. We are in Cape Town and we ship Joburg builds that run well. Being able to be in a Sandton boardroom on a Tuesday morning without a R6,000 flight does matter for one specific thing: discovery.
The first two or three sessions, where the vendor is working out what you actually do, benefit from being in the room. Not because video does not work. Because you walk the floor, you see the sales team shouting across the open-plan, you notice the whiteboard in the breakroom with the three systems nobody mentioned in the brief. After that, the build phase is work that happens on a screen. The vendor's postal code stops mattering.
The exception. Businesses in heavily regulated spaces, a financial adviser practice in Sandton under FAIS, a credit provider in Fourways under the National Credit Act, a healthcare group in Hyde Park under HPCSA, benefit from a partner physically able to attend the compliance meeting when the regulator eventually looks at the file. That access is worth paying for.
Everyone else is paying for a Rosebank office rent they are not using.
The first project worth funding
For most Joburg buyers, the right first project is small enough to be visibly working in six to eight weeks and valuable enough that somebody will notice. A big-bang eighteen-month programme funded as one block is almost never that. The failure mode is predictable. Month eight, two of three deliverables are late, the sponsor has moved roles, and the vendor is renegotiating scope.
The projects that pay back in the first year, from what we have seen across Johannesburg, are narrow. After-hours lead response on WhatsApp for a Fourways services business. Quote-to-draft automation for a Woodmead distributor with ten reps on the road. An intake and FICA flow for a Sandton financial advisory. A SARS-deadline tracker for a tax practitioner in Illovo. None of them changes the business overnight. Each of them removes between fifteen and forty hours of admin a week. They pay for themselves. They give the next project credibility inside the business, which is what earns the second round of budget.
Pick the partner for that project, not for the fantasy. If the vendor is the right fit for the modest first thing, they will still be the right fit for the second. If the only thing they can talk about is the eighteen-month plan, they are not building. They are selling.