What an AI adoption maturity model is for
An AI adoption maturity model is a way to name where your company stands with AI, and what the next step is. It stops two common mistakes. The first is buying tools before people use them. The second is counting AI users and calling that progress.
The model below has 5 stages and 6 capabilities. Use the stages to place your company. Use the capabilities to find the gaps that hold you at your current stage.
The 5 stages of AI adoption
Each stage builds on the one before. Skipping a stage rarely works, because the later stages need the shared setup the earlier ones create.
| Stage | What it looks like | What holds it back |
|---|---|---|
| 1. Scattered | People use AI on personal accounts, each with their own prompts and tricks. | No shared setup, no policy, no view of what anyone is doing. |
| 2. Sanctioned | The company buys a team plan, such as Claude Enterprise, writes a policy and runs training. | Everyone still builds their own setup, so gains stay with individuals. |
| 3. Shared | Teams share rules, company knowledge and a library of reusable skills. Key systems are connected once. | Nothing runs on its own yet, and value is still a guess. |
| 4. Automated | Repeatable workflows run on a schedule or trigger. Engineers ship with supervised coding agents. | Automation grows faster than anyone can tell what it's worth. |
| 5. Compounding | Usage, spend and outcomes are measured, and the roadmap follows the evidence. Tools improve from how people use them. | Keeping the measures honest as tools and teams change. |
What moves a company to the next stage
Each move has one main piece of work. Doing it well matters more than doing several things at once.
- From scattered to sanctioned: pick one company plan, write a short policy on what data can go into AI, and train people on the basics. Keep the policy to a page, or nobody reads it.
- From sanctioned to shared: write down how your company works in the rules files and skills AI reads first. Connect the 2 or 3 systems people use most. Start a skills library with a handful of skills people already rely on.
- From shared to automated: choose workflows that happen the same way every week, and let them run on a schedule or trigger. Keep a person on review for anything that touches customers, money or production code.
- From automated to compounding: tie usage and spend to outcomes, and review them monthly. Retire what doesn't pay for itself. Feed what people fix by hand back into the skills, so the tools get better with use.
- Stage 5 is not a finish line. New tools arrive, teams change and measures drift. Companies at stage 5 keep the loop running rather than declaring victory.
The 6 capabilities to assess
Stages describe the whole company. Capabilities show where it is weak. Score each one from 1 to 5, using the stages above as the scale.
- Connected: the systems people need, such as the CRM, docs and ticketing, are wired to the AI tools once, not by hand for each person.
- Company knowledge: how your company works is built into the rules and skills AI reads first, instead of generic advice.
- Reuse: people can find, install and publish shared skills without friction.
- Automation: repeatable workflows run on their own, with a person reviewing what matters.
- Building: people can build new software with AI safely, with review before anything ships.
- Measurement: usage and spend are tied to outcomes, so you know what is worth keeping.
Why engineering usually leads
In most software companies, engineering reaches stage 3 or 4 first. Engineers already work in shared repos, with review, tests and written specs. Rules files like AGENTS.md and supervised agents like our Code Factory fit that habit.
Business teams usually sit a stage or two behind. The gap has three common causes. Key systems aren't connected. Company knowledge hasn't been built into their skills. And there is no path for a non-technical person to add or update a skill.
Closing that gap is often the highest-value work at stage 3. See AI for business teams for how we approach it.
A quick AI maturity assessment
Answer yes or no to each question. Each yes is roughly a point on the capability it names.
- Does a new hire get the company's AI setup on day one, without building their own?
- Are your main business systems connected to your AI tools for everyone who needs them?
- Do your AI tools know your company's terms, processes and standards without being told each time?
- Can a non-technical person publish a skill their team can use?
- Does at least one workflow run on a schedule without someone starting it?
- Does every AI-written change get reviewed before it ships?
- Can you name, with numbers, the three AI uses most worth their cost?
- Did you retire an AI tool or skill last quarter because the numbers said so?
How to move up one stage
Pick the weakest capability that blocks the next stage, not the most exciting one. Then define what good enough looks like for that stage, and stop there.
For example, a company moving from stage 2 to 3 might connect its CRM and publish 5 shared skills for one team. It would not try to automate that team's whole week at once.
Measure before and after each step. Without a baseline, you can't tell a real gain from a good week. Our page on AI ROI measurement covers the measures we use.
Who should own AI enablement
Moving past stage 3 needs an owner who can hold one roadmap across engineering and business teams. That person makes the sequencing calls and writes specs other teams can build from. They also report on maturity in language leadership understands.
Some companies hire for this. Others start with a fractional lead who owns the roadmap until they do. Either way, the owner needs hands-on experience with AI tools, or the roadmap will be built on secondhand reports.