The Pulse
AI Maturity Model

A practical AI adoption framework for mid-market and enterprise teams, from scattered experiments to an organization that runs on AI.

Most AI maturity models are a straight line. Crawl, walk, run. You start at the bottom, you climb, and one day you arrive. It's tidy, and it's wrong.

We've watched too many companies run seriously capable AI, custom-built tools that save real hours, while the entire capability lives in one person's head. That company looks mature from the outside. It's actually one resignation away from losing everything it built.

That's the problem with a ladder. It can only measure one thing. AI adoption moves on two.

AI adoption has two axes

The first axis is who the AI serves. On one end is single-player: one person making themselves faster. On the other is multi-player: the whole organization working off a shared, governed foundation.

The second axis is what the AI is. On one end, AI is a tool someone reaches for. On the other, AI is a system that runs on its own.

Put those two axes together and you get four quadrants, not four rungs. We call them phases because most companies do move through them in rough order, but the point of the grid is that you can be strong on one axis and stuck on the other. Here is what each phase looks like.

Where every workforce starts. People try AI on real work, unevenly, often in the shadows on whatever they signed up for with a personal account. The question here isn't "how do we get people using AI." They already are. It's how you get them experimenting safely, out in the open, on sanctioned tools.

The wins are individual and immediate. Someone drafts a follow-up email or a first-pass proposal in seconds instead of staring at a blank page. Someone turns a 45-minute call or a dense document into the key points and next steps. People build their own quick dashboards and calculators instead of waiting on an IT backlog, so demand on IT goes down, not up. Teams pressure-test an idea with a working prototype before anyone commits budget. Client-facing people walk into meetings prepared without an analyst behind them.

The outcome of Phase 1 is faster individuals. That's real, and it's also the ceiling if you stop here.

Your strongest people stop repeating themselves. Instead of prompting from scratch every time, they build systems for their own work. A repeatable process gets packaged into a reusable skill. Recurring work runs overnight and lands finished each morning. Core tools get wired together so context flows in one place instead of copy-paste. Routine workflows get handed to a standing assistant, and the end-of-day closeout, time, status, and follow-ups, runs itself.

This is where the payoff gets real, and where the risk gets real too. All of it still lives with the individual. When your best power user walks out the door, the systems they built walk with them.

Phase 2's outcome is systems, not one-offs. The job of leadership here is to capture those wins before they leave.

This is the jump most companies underestimate. You go from a handful of superusers to the whole team, and you put everyone on a shared, governed base. Two foundations get built at once.

The first is the people foundation. One person's skill becomes a shared asset the whole department runs. Training, SOPs, and playbooks turn scattered superusers into a team that performs to one standard. New hires reach full AI productivity in days instead of months, by design. This is change management as much as it is technology.

The second is the data and infrastructure foundation: the clean, governed, connected data the whole organization can build on. One source of truth every tool and skill draws from. Role-based access so people and systems see only what they should. One sanctioned tool stack replacing shadow AI. Core systems integrated at the org level so context flows company-wide, not in one person's setup. And the unglamorous part, getting your data structured and AI-ready.

That last piece, data readiness for AI, is the gate to everything that comes next, and almost nobody clears it. Roughly 7% of organizations are actually data-ready, and the ones that are see up to a 1.6x margin uplift. Phase 3's outcome is a governed foundation.

The operating model itself changes. This is agentic AI as a way of running the business: people stop doing the work and start directing fleets of agents, with humans accountable for the outcomes.

A purpose-built platform with your process embedded replaces a manual operation end to end. This is the biggest and most defensible ROI in the model, measured in P&L impact, not hours saved. Agents run the operation while your people oversee the exceptions and own the result. Always-on agents surface issues before anyone has to ask. Leaders query the whole business in plain English and get governed answers across every system, inside each person's access. Workflows monitor their own output and improve over time.

None of that is safe without control. Agent governance, audit trails, approval chains, and continuous monitoring of every agent are the price of admission, and only about 21% of companies have that today. Phase 4's outcome is an agent-first organization.

Why it's a grid, not a ladder

Now the reason the two axes matter. A company can be deep on the "system" axis and still stuck on "single-player": advanced tools, zero shared foundation. On a ladder, that company looks near the top. On the grid, it's sitting in Phase 2 with a Phase 4 toolset and a single point of failure.

The fix isn't more sophisticated AI. It's moving up the other axis, from single-player to multi-player, before you scale anything further. A crawl, walk, run ladder can't show you that gap. The grid can.

The four rails: is your foundation ready?

The matrix tells you where you are. It doesn't tell you whether your foundation can hold the weight. That's what the rails are for, and together they add up to a working AI readiness assessment. These are the four dimensions we watch mature across every phase, the gauges a CIO, CISO, and CFO read before they'll trust any of this.

• Data readiness is the single biggest predictor of whether AI scales. It moves from ad hoc files scattered across inboxes and drives, to personal stores siloed with your power users, to data cleaned and governed at the org level, to one unified, AI-ready layer your agents draw from.
• Governance and responsible AI is the rules, roles, and accountability for how AI gets used. From a basic acceptable-use policy, to standards around which tools are sanctioned, to defined roles and approvals, to full agent governance with audit trails.
• Security and compliance is about who can access what, and whether your use holds up in a regulated environment. From basic guardrails, to a vetted stack that replaces shadow AI, to role-based access, to continuous monitoring built for regulated environments.
• ROI and value measurement is how you measure AI ROI, proof it's moving the business and not just being used. From time saved by individuals, to process-level ROI, to team KPIs, to impact on the P&L.

A company usually can't sit two full phases ahead on the matrix while a rail lags back at Phase 1. That gap is what stalls projects, and it's almost always the rail, not the technology, that's the real blocker.

How to use the model

Find your honest quadrant first, then your weakest rail. The quadrant tells you what to build next. The rail tells you what will stop you if you skip ahead.

Most companies we work with, manufacturers, accounting and wealth-and-tax firms, law practices, and enterprise teams already running Claude, start somewhere in Phase 1 or 2 and assume they need better AI. Nine times out of ten, they need a foundation. Nobody starts at Phase 4, and nobody gets there by climbing a ladder one rung at a time. They get there by moving on both axes on purpose, which is the heart of our AI transformation and custom AI development work.