AI made your people faster, not your business
AI tools make individual people faster. They do not change how work moves between people, and in a business where the delay lives in the handoff, a faster typist buys you nothing. The AI use cases worth money in South Africa are the ones that close a gap between two systems.
This argument is currently everywhere, and it is being made with numbers you should not repeat.
The evidence is sponsor content, and that matters
The claim that enterprises adopted AI and sped up individuals without fixing how teams work arrived this week inside a paid placement. Atlassian and AWS surveyed 500 executives, and the finding sits in a whitepaper promoted in a newsletter's sponsor slot. A second survey, of 1 500 early-stage founders, ran the same week in the same format for a different sponsor.
Neither underlying report has been read here, and both were paid to appear where they appeared. That does not make the numbers wrong. It makes them marketing, and a figure lifted out of marketing and quoted as research is how a business case gets built on sand.
None of those figures appear below. The argument does not need them, which is worth saying plainly, because the argument is older than the surveys and you can test it against your own week.
Where the delay actually lives
Time a job in your business from the customer's request to the invoice going out. Then subtract the minutes anybody spent typing.
The remainder is the answer. It is the quote sitting in a sales inbox waiting for a price confirmation. A completed job card stays in a technician's bag until Friday. Somewhere on a desk, a supplier invoice waits for a second signature, next to a delivery note nobody matched to a purchase order.
None of those are typing problems. They are handoffs, and a handoff is where information stops moving because the next system does not know it exists. Speeding up the typing at either end leaves the gap exactly where it was.
The other direction the problem is arriving from
The same point is now being made by people building for agents rather than for teams.
CIO.com's argument is that enterprise data stacks were designed around human access patterns while agents need reliable real-time access across systems, which turns data integration into a prerequisite for anything agentic actually working. Anthropic's own account of AI-native software development says teams need to redesign workflows around agents rather than bolt them onto existing processes.
Read those alongside each other and they say the same thing from opposite ends. A system built for a person to log in, look and retype cannot be automated by putting a model in front of the same login. The retyping was the integration, and it was being done by a human.
Ramp released the model routing infrastructure it uses internally and says the product saves customers 40% on cost by choosing models on cost, quality and latency. The newsletter that carried the launch reported 30% and credited it to Ramp's own bill, which is a small demonstration of the point above. The part that generalises is that the saving came from instrumenting a process nobody had measured.
What this looks like in a 40-person business
A distribution business in Germiston buys three AI subscriptions. The sales manager writes quotes faster. Letters come out of the bookkeeper in half the time. Long emails get summarised for the operations manager before he opens them.
Six months later the quote-to-invoice cycle is the same length it was, because the delay was never in the writing. It was in the two days a quote waits for stock confirmation, which lives in a different system that the sales manager cannot see.
The three subscriptions were not a mistake. They were just aimed at the fastest part of the process, which is the part where a person is already typing at full speed and there is nothing left to win.
The local version of the point
Kara Fox is chief of staff at Cape Town ecommerce agency Ecommerce Counsel, where automation is part of the day job. She has written a book called Human Still Wins, arguing that trust, judgement, taste and courage become more valuable as AI makes competent execution easier to produce.
That is the same observation from a third angle. When execution gets cheap, the value moves to the decisions between the pieces of execution. In an operating business, the decisions between the pieces are the handoffs.
Where to point the money instead
The staged version costs less and works. Find the single handoff where information is retyped from one screen into another most often, fix that one properly using the systems already in place, and only then connect the rest. That is the work Zorah does, and the differentiator is that it joins up the tools a business already runs rather than replacing them with something new to learn.
An AI capability bolted onto a broken handoff inherits the break. A distributor moving quotes to invoices discovers this in the first week.
What to do on Monday
Pick your longest-running open job and walk it backwards. At each step write down the date it arrived and the date it moved on.
You will find one gap that is measured in days while every other step is measured in minutes. That gap is your AI use case, and it will not be the step where anybody was typing.