Most companies can say they use AI somewhere, and far fewer can show that a single workflow now runs differently. This guide gives South African mid-sized business leaders a way to score five dimensions with evidence, and to decide what to do first.
TL;DR
Most organisations that tell us they are using AI can name the tools. Few can name a workflow that now runs differently because of them. The gap between those two answers is what a readiness assessment is for.
This guide is written for managing directors and functional leaders at South African businesses of roughly 50 to 500 people. It covers how to score the five dimensions we use at Praxes, what evidence to ask for, what South African law adds, and what to do with the result.
An AI readiness assessment is a structured check of whether your organisation can get value from AI before you commit money to tools or pilots. It produces a score for each area and a view of which gap to close first.
We assess five dimensions: strategy, data and systems, people, process, and tooling. We use five because readiness fails at whichever one is weakest. A business with a clear strategy and clean data can still stall if nobody has changed how the work is done.
Readiness works like a production line: output is set by the slowest station, however fast the others run. Buying a faster machine for the station that is not the bottleneck changes nothing. The assessment finds the slow station.
Scores based on opinion drift upwards, because the people scoring are usually the people responsible. Evidence keeps the score honest: for every point you award, you should be able to show a document, a system, or a person who can confirm it.
The public numbers show why this matters. Only 36% of employees in AI-integrating organisations strongly agree that their manager actively supports their team's use of AI, and manager support is one of the two conditions most strongly linked to frequent use. A leadership team would likely score itself higher than that. Asking employees, and not only executives, closes the gap.
Score each dimension from one (nothing in place) to five (in place, evidenced, and working). For each, I give one question and the evidence that would justify a four or five. Treat the scale as a conversation tool and not a precise measurement.
A strong score means leadership has agreed two or three business problems to solve with AI, each with an owner and a measure. A weak score looks like a long list of ideas and no agreed order. The evidence is a one-page document that a manager outside the leadership team can explain back to you.
A model can only work with what your systems can give it. Ask whether the data for your first use case is accurate, reachable, and shared between systems, and whether the workflow is documented well enough to improve. The evidence is a named data source for one use case, with an owner who can say when it was last cleaned.
In South Africa, add one more question: do you know where personal information sits? POPIA section 71 restricts decisions that affect a person to a substantial degree when they rest solely on automated processing of personal information, with exceptions that require appropriate protections. If AI will touch hiring, credit, or performance, you need to know which data it will use and where a person reviews the outcome. I am not a lawyer, so take specific cases to one.
Adoption is where most programmes stall. A strong score means people feel safe to use approved tools, managers can say what will change in their team's week, and someone has asked staff, not only leaders, how it is going. The evidence is a short staff survey and two or three manager conversations. Shadow AI, where staff use unapproved tools because no approved option exists, usually signals that people have not been given a safe route, which is a weakness in this dimension.
A tool that is not built into the work becomes another tab nobody opens. Pick one workflow and ask whether the steps, handovers, and sign-offs have been redesigned around AI, or whether AI has been added to the old steps. The evidence is a before-and-after description of one workflow, with the time or cost of each. Some workflows need a better process and nothing more, so it is worth asking whether this one needs AI at all.
Platform and vendor decisions compound quickly and cost a lot to reverse. A strong score means you know which tools are approved, who decided, what data each can see, and what it costs in rand once the foreign-currency pricing of most software is counted. The evidence is a list of tools in use, including the ones staff bought themselves.
The table below is an illustration I built to show the logic. It is not data from a client engagement.
| Dimension | Score | Evidence found | What it tells you |
|---|---|---|---|
| Strategy | 4 | One-page priorities with owners | Ready to proceed |
| Data and systems | 2 | Customer data in three systems, no owner | Fix before building |
| People | 3 | Managers unsure what changes | Brief managers first |
| Process | 2 | Workflows undocumented | Document one workflow |
| Tooling | 4 | Approved list in place | Hold further purchases |
In this example the first move is a data owner and one documented workflow, and a new tool is not on the list. Two dimensions score two, and tooling already scores four, so more buying would add to the gap.
You can run a first pass in about a week with two or three people. Each step below is specific enough to do on its own.
A low score points to the work to do first. It is a starting point for the business and carries no verdict. The table below maps each weak dimension to a first move.
| Weakest dimension | First move | Does it need AI? |
|---|---|---|
| Strategy | Agree two or three problems, with owners | No, it needs a decision |
| Data and systems | Name a data owner and clean one source | No, it needs data work |
| People | Brief managers, then approve a tool | No, it needs conversations |
| Process | Document and redesign one workflow | Sometimes |
| Tooling | List what is in use and who approved it | Not yet |
Most first moves involve no AI at all. That is common, and it is a good result, because it saves you from a purchase the business was not ready for. Our view is that the best advice is sometimes to fix a data pipeline or a process and leave AI alone for a while. Where data or process is the bottleneck, adding AI amplifies the problem, so fixing the foundation first usually returns more value sooner. The distinction is explored further in automation versus AI.
Keep the wider picture in proportion. AI high performers, those attributing 5% or more of earnings before interest and tax to AI and reporting significant value, make up just 6% of survey respondents. Most organisations are still working out the foundations, and a low score puts you in the majority.
A self-assessment stops being enough when the people scoring are the people accountable for the result, or when nobody in the business has the technical judgement to test the data and tooling answers. At that point an outside view helps, and it can come in several forms.
Some organisations will do all of this well alone, and that is a fine outcome. Where decisions keep arriving and nobody owns them, senior outside judgement can help without a full-time hire. That is a view, and you are better placed than we are to judge your own situation.
One caution on the future. One Gartner survey of more than 700 CIOs found they expect AI to be part of nearly all IT work by 2030. I read that as a direction rather than a timetable, and I would not let it set your pace. Your own scores should.
It is a structured check of whether your strategy, data and systems, people, process, and tooling can support AI before you spend on tools. Scored with evidence, it shows which dimension is weakest and what to fix first.
A first pass for one function takes about a week with two or three people. Most of the time goes on collecting evidence and asking staff, not on scoring.
Yes. The steps above need no outside help. An outside view helps most when the scorers are also the people accountable for the result.
It does where AI uses personal information to make or inform decisions about people. Know which personal data a use case touches, and keep a person involved in decisions that affect someone substantially. Take specific cases to a lawyer.
Start with the lowest dimension and make one change to one workflow. A low score is common, and the first move often involves no AI. Review at 30 days to check the work changed.
If you would like a scored picture across the five dimensions to take to your leadership team, the AI Readiness Diagnostic is free and self-serve, and you are welcome to use it alongside the steps above. For the general method behind this post, our guide to running an AI readiness assessment inside your organisation covers it.
Ready to move from reading about AI to getting your organisation ready for it? Start with the diagnostic.
Take the AI Readiness Diagnostic