A full chapter from The 2026 AI Strategy, published open. The complete ebook is free; this chapter stands on its own.
The capability gap becomes actionable the moment you can place yourself on the path. Five levels run from no operating AI to AI as the operating model. Read the descriptions, not the labels, and place your company where the description fits, not where the ambition does.
Exhibit 5: The maturity path.
| Level | Stage | What it looks like inside the company | Where the AI effort sits |
|---|---|---|---|
| 0 | Exploring | Slides reference AI. Demos exist. Nothing runs in production. The story is louder than the system. | Slideware only |
| 1 | Experimenting | Individual users on off-the-shelf tools. Personal productivity wins. No operational pattern, no business case, no governance. | Individual desks |
| 2 | Strategy | A real plan: use cases ranked, ROI projected, data readiness understood. Nothing yet built. | On paper |
| 3 | In production | At least one production workflow runs on AI, integrated, with humans in the loop. Others still on the roadmap. | One function |
| 4 | Operating model | AI is part of how the company runs: continuous monitoring, retraining, governance, measured economics. AI runs the company, not just works inside it. | Across functions |
Two honest calibrations before you place yourself. First, the top is rare: McKinsey found only 1% of leaders describe their generative AI rollouts as mature, and BCG found 74% of companies struggle to scale AI value beyond pilots (McKinsey, 2024; BCG, 2024). Most mid-market companies we meet sit between Level 0 and Level 2, and the copilot deployments that feel like progress are Level 1. There is no shame in the reading; there is only cost in misreading it. Second, notice how this maps to the three responses from earlier in the ebook: bolting on features is Level 1 wearing a budget, isolated pilots stall between Levels 2 and 3, and the rebuild is the move that crosses into Levels 3 and 4, where the 6% live.
Your level is not one judgment. It is a composite of seven dimensions, each scored on the same 0-to-4 scale as the path itself, and the composite obeys one unforgiving rule: your level is your lowest dimension score. A company at 3 on technology and 1 on governance is a Level 1 company, because the operating model is only as strong as its weakest load-bearing wall. This is the same physics as the flywheel's slowest lobe, applied to capability.
Here is how to score. For each of the seven dimensions, ask the dimension's question and place your answer on the same ladder you just read: 0 means nothing real exists (talk and slides); 1 means it exists at individual desks, ad hoc, undocumented; 2 means it exists as a real plan on paper; 3 means it runs in production in at least one place; 4 means it runs systematically across the company. The 0-and-4 anchors in the table calibrate each row.
Exhibit 6: The seven-dimension self-score. One score per row, 0 to 4. Your maturity level is the lowest number in the column.
| Dimension | The question | 0 looks like | 4 looks like | Your score |
|---|---|---|---|---|
| Strategy | Is there a funded AI strategy tied to P&L outcomes, with a named executive owner? | "We know we need to do something" | Ratified, sponsored, funded, tied to specific outcomes | |
| Data | Could an agent trust your data today? | Fragmented, manually reconciled, distrusted | Fast, clean, integrated, trusted for decisions | |
| Technology | Can you walk me through one production workflow where AI does the work, end to end? | "We use AI," with no walkthrough | Multiple monitored production workflows | |
| Talent | Who owns this, and what happens if that person leaves? | The knowledge lives in one or two heads | A named owner with budget, expertise spread across people | |
| Organization | When did this company last ship a change of this size, and how did it go? | Rollbacks, stalls, "people resist new tools" | Transformations shipped, executives aligned, change capacity proven | |
| Governance | What is AI allowed to do here, who reviews it, and what happens when it is wrong? | Nobody has written anything down | Written rules, review roles, escalation paths, all in use | |
| Business value | What number will tell you this worked, and what is its baseline today? | "We would like to figure that out" | Named KPIs, baselines set, P&L linkage reported quarterly |
A worked example, so the arithmetic is concrete. A distributor scores itself: strategy 2 (a real plan on paper), data 1 (spreadsheet reconciliation everywhere), technology 2 (a pilot, nothing in production), talent 1 (one enthusiast owns everything), organization 3 (they shipped an ERP migration well), governance 0 (nothing written), business value 1 (no baselines). The lowest score is governance at 0, with data and talent close behind at 1: this is a Level 0 to 1 company, whatever its strategy deck says, and its first three fixes just named themselves. That is the payoff of scoring: not the number, but the discovery that your lowest rows are your first fixes, and usually your cheapest ones. The research is blunt about where outcomes come from: organizations with successful AI initiatives invest up to four times more in the foundational dimensions (data quality, governance, AI-ready people, change management) than those with poor outcomes (Gartner, 2026). The glamour is in the agents. The variance is in the foundations.
Your level also sets your realistic entry point on the path from the Realistic Path chapter: Levels 0 and 1 start with the diagnostic and the data-readiness work, Level 2 is ready to pick the first workflow and build, and Level 3 companies should be scaling what already works and instrumenting it properly rather than starting anything new. Skipping your level is how pilots join the 95%.