AI maturity is not the number of AI tools a company has. It is how deeply AI is built into the way work is done, governed and measured. Research from MIT describes four stages, from experimenting to being AI future-ready, and finds that companies in the later stages perform well above their industry average while those in the first stage perform well below it. Most companies are still early, and the ones moving ahead are not buying more tools: they are redesigning their workflows. This roadmap closes our series by showing what each stage looks like, what to build in each department, and how to move up one stage in the next 90 days.
What does AI maturity mean?
The clearest model comes from the MIT Center for Information Systems Research. Its Enterprise AI Maturity Model, based on a survey of 721 companies, describes four stages:
- Experiment and prepare: 28% of companies. Learning, policies and first experiments.
- Build pilots and capabilities: 34%. Pilots in several areas and the platforms to support them.
- Develop AI ways of working: 31%. AI built into how processes run across the company.
- Become AI future-ready: 7%. AI embedded in decisions, products and new business models.
The financial gap between the stages is large. Companies in the first stage had growth 12.6 percentage points and profit 9.6 percentage points below their industry average. Companies in the fourth stage were 17.1 points above on growth and 10.4 points above on profit.
Where do most companies stand today?
Most are using AI, few are getting real value. In McKinsey's State of AI 2026, 44% of respondents say AI is scaling across their enterprise, but only 37% attribute any EBIT impact to AI, and the high performers that attribute 5% or more of EBIT to AI are just 6% of respondents.
Maturity also decides whether AI lasts. A 2025 Gartner survey found that 45% of high-maturity organisations keep their AI projects in production for at least three years, against 20% of low-maturity ones. In high-maturity organisations, business units trust and are ready to use new AI solutions in 57% of cases, compared with 14% in low-maturity organisations.
What does each stage look like, from startup to enterprise?
Maturity is not the same as size. A ten-person company can be more mature than a large group. But each stage has a typical shape:
| Stage | What it looks like | What to build | Main risk |
|---|---|---|---|
| 1. Experiment and prepare | People use AI on their own, often with personal tools | A one-page AI policy, approved tools and basic training for everyone | Data leaks and confident wrong answers nobody checks |
| 2. Pilots and capabilities | A few teams run AI pilots on real tasks | Clean data in core systems, a baseline for each pilot and named owners | Pilots that never reach production |
| 3. AI ways of working | AI is part of daily processes in several departments | Connected systems, permissions by role, approval rules and shared measurement | Adding AI on top of old processes instead of redesigning them |
| 4. AI future-ready | AI shapes decisions, products and how the company competes | Governance at scale, reusable AI services and new AI-based offers | Complexity, cost and losing human judgement in key decisions |
What should each department build at each stage?
Every article in this series covers one area in detail. Here is the short version, with where to start and what scaling looks like.
| Department | Start (stages 1 and 2) | Scale (stages 3 and 4) |
|---|---|---|
| Whole team | Policy, approved tools and a 30-day training plan | Role-based training refreshed regularly and AI skills in every job description |
| Marketing | Drafts, variations and repurposing in a saved brand voice | Personalisation from CRM data and AI-assisted planning with human approval |
| Sales | Follow-up drafts and call summaries | AI briefs, lead scoring and pipeline questions answered from the CRM |
| Customer service | Suggested replies from approved policies | Assistants that resolve routine cases and hand over to people |
| Finance | Invoice data extraction with checks | Automated reconciliation and e-invoicing with exception review |
| HR | Job descriptions, scheduling and handbook answers | Structured screening support with bias checks and human decisions |
| Operations | Document processing and order questions | AI demand forecasting and purchase suggestions with approval limits |
| Management | Meeting summaries and simple data questions | Managers asking live company data within their permissions |
| Software and IT | Coding assistants with review rules | AI across the delivery pipeline with security scanning and stability tracking |
Read the full guides: training your team, founding a company with AI, marketing, sales, customer service, finance, HR, operations, management and software development.
What separates the companies that get ahead?
They change the work, not just the tools. In McKinsey's March 2025 State of AI, workflow redesign had the biggest effect on whether organisations saw EBIT impact from generative AI out of 25 attributes tested, yet only 21% of organisations had fundamentally redesigned at least some workflows. In the 2026 edition, nearly three-quarters of high performers report fundamentally redesigning workflows because of AI, against about one-quarter of other companies. High performers are also more than three times as likely to be scaling AI agents and twice as likely to report commitment from senior leaders.
In practice, that means four habits: leaders who use AI themselves and set the direction, processes redesigned around what AI does well, clean and connected data, and every AI workflow measured against a baseline. Our guide on how to measure AI ROI covers the measurement part.
What should you avoid on the way?
- Chasing the newest trend. Gartner predicts that over 40% of agentic AI projects will be cancelled by the end of 2027 because of rising costs, unclear business value or weak risk controls.
- Tool sprawl. Ten disconnected tools with no owner are less mature than one well-run workflow.
- Skipping the data work. AI cannot fix records that disagree with each other.
- Pilots with no exit. Every pilot needs a baseline, a target and a date to decide whether to scale or stop, as we explain in the AI pilot graveyard.
How do you move up one stage in the next 90 days?
- If you are in stage 1: publish an AI policy, approve tools, train everyone and pick two pilots with a recorded baseline.
- If you are in stage 2: clean the data in your core systems, decide which pilot to scale and which to stop, and name an owner for each AI workflow.
- If you are in stage 3: redesign one end-to-end process around AI, connect the systems it needs and set approval rules by role. Our AI governance policy helps.
- If you are in stage 4: review which decisions should stay human, reduce duplicate tools and look for new services your AI capabilities make possible.
A note to close the series
We started this series with a simple observation: most people are using AI before anyone has explained it to them, and that is why so much of it is used wrongly. Across ten articles we tried to show the same pattern in every department. The traditional way is slow but understood. AI used wrong is fast and risky. AI used right is fast, checked and owned by people.
At HBS we used AI on our own company before offering Amrak to anyone, and we still run our own website, content and marketing on it every day. The lesson we keep relearning is that the technology is rarely the hard part. Deciding what AI is allowed to do, who checks it and how you measure it is where maturity is built.
Not sure which stage your company is in, or what to do next? Book a free discovery session and we will map your departments against this roadmap with you. Talk to us.
