Most organisations overestimate their AI maturity. They call themselves Stage 3 or 4 while missing the actual markers of Stage 3: a measured baseline, a named owner and cross-functional ownership. This four-stage AI maturity model shows why progress typically stalls at Stage 2, and what it actually takes to move past it.
“We are at Stage 3, maybe 4.”
The executive said it without hesitation, and the room nodded.
Three questions later, I established that no workflow in the business had a measured baseline, no AI initiative had a named owner, and the strategy's centrepiece was an enterprise licence agreement.
That is not Stage 3. That is Stage 1 with good procurement.
The gap between where leaders believe they are and where their organisation operates is the most consistent thing I have seen across four continents, South Africa included. This article walks the road properly: the four stages, the honest markers and the stall that consumes most journeys.
The four stages, with markers
Personal AI
Individuals augment their own work. It is valuable but largely invisible to the organisation.
Marker: Productivity anecdotes with no shared assets.
Workflow AI
A team redesigns one workflow around what the machine can hold, with fewer steps than before.
Marker: One baseline, one named owner and measurable before-and-after performance.
Process AI
An entire process is rebuilt across team boundaries and becomes visible in the P&L.
Marker: Cross-functional ownership, kill dates and measurable financial value.
AI-Native
The operating model itself assumes AI. Budgets, roles, controls and planning cycles are designed AI-first.
Marker: AI is structural and visible in the organisation chart and budget architecture.
Stage 1: Personal AI
Personal AI is individuals augmenting their own work. The marker is productivity anecdotes and no shared assets. Valuable, invisible, and where roughly 90% of workers already are with personal tools, according to MIT's 2025 research, whatever the official adoption numbers say. The real value starts with moving beyond personal AI productivity into something the organisation can actually see.
Stage 2: Workflow AI
Workflow AI is a team redesigning a single workflow around what the machine can hold. The marker is one process with a before-and-after baseline, a named owner and fewer steps than it had before. Not the same steps plus a copilot. Fewer. This is also the stage where most firms first try to adopt AI without disrupting operations, and where most of those attempts quietly stall.
Stage 3: Process AI
Process AI is end-to-end: a whole process rebuilt, crossing team boundaries and visible in the P&L. The markers are cross-functional ownership, kill dates as standard practice, and finance being able to point at the line where the value lands.
Stage 4: AI-Native
AI-Native is the stage everyone claims, and almost nobody occupies. The operating model itself assumes AI. Budgets, roles, controls and planning cycles are designed AI-first. The marker is structural; you see it in the organisation chart and the budget architecture, not the tooling.
Why the money is where it is
Value concentrates at depth on the AI maturity model. BCG's 2025 research puts the deep group at roughly 1.7 times the revenue growth of peers. McKinsey's enterprise AI adoption survey finds only 6% of firms attributing more than 5% of EBIT, or earnings before interest and taxes, to AI. Stage 1 produces enthusiasm. Stages 3 and 4 produce financial statements.
The stall at Stage 2
Most organisations I meet, locally and abroad, live between Stages 1 and 2, and the climb they keep failing is from 2 to 3. The reasons are structural, not technical.
Crossing to Process AI means crossing team boundaries, and boundaries are where incentives collide. The workflow owner is a manager; the process owner is a politician. Data architecture starts to matter brutally at this stage. Processes run on data that workflows could work around unnoticed. Governance debt, ignorable at Stage 2, compounds. More automation creates more decisions, often without any added accountability structure.
There is also a quieter reason. Stage 2 feels like success. There are wins to present and dashboards to show. The organisation settles into a comfortable plateau of pilots and copilots, and calls it “transformation”.
What carries the climb
Four supports help organisations move forward, and none of them is another tool.
Secure platform
Give shadow AI usage somewhere sanctioned, governed and measurable to live.
Leadership coaching
Build better instincts for leaders making capital decisions about unfamiliar technology.
Practical training
Build fluency where the work happens, rather than awareness in a town hall.
Build capability
Stage 3 always requires some assembly. Buying technology alone only gets you to Stage 2.
The honest caveats
You cannot skip rungs; every skipped stage shows up later as rework. Speed on the rungs varies enormously, and small firms climb faster than their maturity assessments suggest.
Not every function needs Stage 4. AI-Native finance and Stage 2 legal can be the right answer in the same company. Uniform maturity is a consultant's tidy diagram, not an operating goal.
Frequently asked questions
What is an AI maturity model?
It is a four-stage framework, Personal AI, Workflow AI, Process AI and AI-Native, that places an organisation using concrete markers instead of tooling or enthusiasm.
What are the four stages of AI maturity?
Stage 1 is Personal AI, where individuals augment their own work without shared assets. Stage 2 is Workflow AI, where one workflow has a named owner and measured baseline. Stage 3 is Process AI, where an entire process is rebuilt across team boundaries and is visible in the P&L. Stage 4 is AI-Native, where the operating model is designed AI-first from the budget upwards.
Why do most organisations stall at Workflow AI?
Crossing into Process AI means crossing team boundaries, where incentives collide and data architecture starts to matter. Workflow AI also feels like success, with dashboards to show, so organisations settle into a plateau of pilots and copilots and call it transformation.
How do you know if you are really at Process AI?
Ask three questions. Does any workflow have a measured baseline? Does any AI initiative have a named owner and a kill date? Would your processes survive a tool swap? Honest answers place an organisation more accurately than any AI maturity assessment.
Does every business function need to reach AI-Native?
No. AI-Native finance and Stage 2 legal can be the right answer in the same company. Uniform maturity across an organisation is not the goal.
What helps organisations climb past Workflow AI?
Four things: a secure platform so shadow AI usage has somewhere sanctioned to live, coaching for leaders making capital decisions, hands-on training where the work happens, and deliberately built internal capability. None of them is another tool.
One challenge for your morning
Run these three questions on your organisation. Does one workflow have a measured baseline? Is there a named owner and kill date anywhere? Would your processes survive a tool swap?
Your answers place you on the road more accurately than any AI maturity assessment, and it is free. Then pick one workflow and start the climb where you actually stand.
Related insights
