The story you keep hearing is the wrong story
Every week another article warns that AI is coming for a marketer’s job. Copywriters. Junior media buyers. Content strategists. The framing is always the same: AI as substitution. Machine replaces human. Somebody loses.
That framing is misleading for how agency work is actually changing in South Africa.
The observable pattern is not substitution. It is leverage. A single operator now does the work of what used to be a small team, and takes home the margin that used to fund the salaries. Nobody at a big agency gets replaced by AI. The small competitor gets ten times better without hiring, and the big agency notices that the market it thought it owned is being eaten from below by someone with a laptop and a well-built stack.
If you run or work at a small agency in South Africa right now, this is the story that matters. Not the substitution one.
What this looks like from inside
GKnect Digital runs performance marketing, SEO and content for about twenty active clients across financial services, e-commerce, hospitality and industrial. On any given Tuesday I might be looking at Meta underperformance on a retirement product at nine, a GA4 attribution question from a wholesaler at eleven, and a new-business pitch to a hospitality group after lunch.
The honest answer to “how” is not that I work harder than everyone else. Effort is table stakes. The differentiator is that maybe thirty separate pieces of work in my week that used to require me sitting down and doing them now happen through a scripted or AI-augmented workflow that I set up once and now maintain occasionally.
Some concrete examples from the last six months.
Ad operations across four platforms. Meta, Google, LinkedIn and Microsoft Ads for the client mix. Historically a two-person job at agency scale. Now: a single command centre pulls performance nightly from all four platforms, flags underperformers, computes budget reallocations, and drafts client-ready weekly reports with the numbers, the interpretation and the recommendation. I spend my time on the interpretation and the client call. Everything upstream is done for me by the time I open the laptop.
Lead sync between CRM and ad platforms. For a financial services client, every qualified lead status in the CRM flows back to Meta the next morning as a conversion event, so the algorithm optimises for actual qualified leads and not for raw form-fills. This used to require a growth engineer at an early-stage startup. Now it is a scheduled task that runs at 04:00 in dry-run mode and posts a summary to me in the morning.
Weekly client reports. Every client that gets a weekly performance digest used to require logging into six dashboards, pulling screenshots, writing a summary, formatting the whole thing. Now a single skill reads the raw exports and produces a branded Word document with charts, commentary and recommendations, in about ninety seconds. I still write the “and here is what I think” section myself. The plumbing that used to consume Sunday nights is gone.
Compliance screening. Every AI-drafted piece for a regulated client passes through a structured FAIS / POPIA / CPA screen before it goes to the client for review. That screen used to require a lawyer, which meant it did not happen, which meant we were exposed. Now it happens on every piece.
New-business research. A new prospect asks for a proposal. Their public data (site, GBP, ads library, GSC-inferred signals, competitor set) is assembled into a briefing pack in about ten minutes. First call is on the strength of a full picture, not a Google search done in the Uber.
None of this is exotic. All of it is boring plumbing that used to require either a person or an expensive SaaS subscription per seat. I built most of it in evenings over the last twelve months.
Why this pattern compounds
When you replace one job with an AI tool, that is an incremental efficiency gain. Fine, but limited. When you replace ten pieces of work in one person’s week with ten separate small automations, something different happens. Each one gives back a few hours. Together they give back a day. That day gets spent on the next automation, which gives back another day next month.
The gains compound because the capacity freed up gets reinvested into building more capacity. This is why a small agency with the right mindset now outruns organisations twenty times its size on specific playing fields. Not because AI is smarter. Because the small operator has no meetings to sit in about approving the automation.
The reverse is also true. A large agency with the same tools and a functional governance layer will move slower than the small operator, every time. That is not a shot at large agencies. It is what happens when your surface area for governance grows faster than your surface area for output.
Three moves that separate the operators pulling this off from the ones who are not
Move 1: see plumbing everywhere. The operators who compound fastest are the ones who look at a manual process (pulling a client report, writing a follow-up, invoicing, drafting a monthly retainer summary) and ask “why am I doing this by hand.” Most people accept the plumbing as the cost of running an agency. The compounding operator treats it as a bug.
This is a mindset shift more than a skill. Once you see plumbing as fixable, you fix it. Six months in, your default question when you catch yourself doing repetitive work is “how do I make this the last time.”
Move 2: ship ugly automations. The operators who never get past step one are the ones who insist their first automation must be perfect before they use it. That is a killer. My first ad-performance script emailed me a report with a formatting bug for six weeks. It still saved thirty minutes a day. I fixed the formatting on a lazy Sunday later, once I had the time the script itself had bought me.
Ship the ugly version. Improve it later. The gain is the freed-up time, not the elegance.
Move 3: build your own, do not just buy. There is a whole SaaS market selling you the automations described above. You will pay per seat, forever, for a version that does not know your business. And you will be limited to what the vendor thought to build.
The alternative is using AI to build your own thin wrappers over the APIs you already have access to. Three years ago that was mostly out of reach for non-engineers. It is not out of reach now. If you can describe what you want a script to do in plain English, you can build a working version of it in an evening with a competent AI assistant. It will not be beautiful. It will do exactly what you asked. That combination is a superpower for a solo operator.
What this does not mean
It does not mean every agency should try to run solo. There are agencies where the value is in the team, the culture, the reliability of process. The solo-operator model is one shape. It is not the only viable one.
It does not mean staff are unnecessary. It means the shape of what you hire is changing. You are less likely to hire someone to do repetitive reporting. You are more likely to hire someone to interpret, decide, sell, or care for a client at the moments the automation cannot.
And it does not mean AI is a magic wand. Every one of the workflows above took real work to set up. Real debugging. Real re-runs of things that broke. The upfront cost is not zero. The compounding return is.
What to do this month
Three concrete moves for any agency operator this month.
Write down every repetitive process you personally do in a normal week. Pulling client reports. Drafting follow-ups. Chasing suppliers. Assembling status updates. The list will be longer than you expect.
Pick the one that costs you the most time and give it one evening. Not a project. One evening. Whether you can shrink it by half. Most of them, you can.
Repeat this rhythm monthly. Not weekly. Monthly is enough to compound and slow enough not to consume the agency you are trying to run.
Do that for a year and you will be running an agency that would have needed a small team to run at the same output five years ago. Not because you replaced anyone. Because you built the leverage yourself.
That is the actual AI story for agency-land. It is quieter than the substitution one, and much better news.
Want to see this stack in action?
If you are curious what an AI-augmented agency looks like from the client side (weekly digest, live dashboard, compliance screen and all), we can walk you through what our clients get. Reach out via gknectdigital.co.za.
Part 3 of a 4-part series on how Marcos Diez uses AI to run GKnect. Coming next: stop asking ChatGPT questions, start giving it your context.
