87% of large enterprises have adopted some AI solution. Only 9% have reached real maturity. And Microsoft admitted in July 2026 that less than 4.5% of its 450 million Microsoft 365 users decided to pay for Copilot, the assistant that has been installed on every corporate screen for three years.
The gap between those numbers has a name: adoption levels. Five of them, to be precise. And most organizations don't know which one they're at.
The Ghost, the Tourist, and the ones who actually build
Level 0: The Ghost. The license exists. Access is enabled. The admin dashboard shows zero activity. The tool sits in the tenant like furniture covered with a sheet: purchased, never used. This happens when the buying decision was made by IT without involving the people who were supposed to use it, and nobody took ownership of adoption after deployment.
Level 1: The Tourist. The user tries it out when they remember it exists. Summarizes an email, asks what a term means, generates a draft they then rewrite almost entirely. They use it like an advanced search engine with which they have a sporadic relationship. It hasn't been integrated into any workflow. This level produces the statistic that 88% of companies "report AI usage," while hiding the fact that that usage is shallow and isolated.
Level 2: The Resident. AI is part of the daily routine for specific tasks. Meetings get summarized automatically. Long emails get drafted with assistance. Documents get analyzed before being read in full. No process has changed, but time is being recovered. Reaching this level requires someone to have done the work of showing how it's used with real cases from that specific organization, not generic demos.
Level 3: The Builder. There are people in the organization who don't just use AI but configure it so others use it better. They create agents with specific instructions. They design shared prompts. They build automated workflows connected to Copilot. AI has stopped being individual and become an organizational resource. McKinsey estimates that only 30% of organizations reach this level or higher. This is the boundary where value stops being marginal and starts being measurable.
Level 4: The Integrator. AI is not a separate tool: it's sewn into the organization's systems and business processes. MCP servers connecting models to internal data. Autonomous agents executing tasks without human intervention at every step. Organizational knowledge has stopped living only in people and documents and started living in the architecture as well. Only 9% of companies get here, and not because it's technologically out of reach, but because the path requires decisions that few organizations make in the right order.
Where each type of organization falls
| Organization type | Typical level | Main barrier | What unlocks the next level |
|---|---|---|---|
| Regulated enterprise (banking, insurance, healthcare, government) | 0-1 | Governance and compliance | Well-scoped "safe" use cases |
| Professional services (legal, consulting, accounting) | 1-2 | No technical team to scale | An internal champion with explicit mandate |
| Tech company / startup | 2-4 | Shadow IT and governance debt | Usage policy + audit process |
| Industrial (manufacturing, energy, mining) | 1-3* | Gap between corporate and operations | Area-by-area differentiated adoption |
| SMB without a tech team | 0-1 | No internal champion | One concrete use case + guided support |
*The asterisk on industrial is intentional: IT may be at Level 3 while the shop floor stays at Level 0. The gap is internal, and it creates its own friction.
Financial services has one of the highest adoption rates on paper (87% of large enterprises with at least one AI workload in production), but one of the largest gaps between what's enabled and what's actually used. Healthcare and government lag even further behind: 61% and 38% respectively in initial adoption, with real maturity even scarcer. Professional services firms have the most common profile: users with high individual willingness, but nobody to carry the process to the next level. Spontaneous champions emerge who discover AI on their own, but that knowledge doesn't get systematized. The organization stays at Level 1-2 with Level 3 islands that depend on specific individuals.
What doesn't separate the levels
It would be convenient if the problem were budget. But it isn't.
Organizations that reached Level 3 and 4 didn't necessarily invest more. They invested differently. Before buying more tools, they defined what they wanted to transform and why. Before enabling access for everyone, they trained someone in how to use well what they already had.
79% of companies investing in AI report serious challenges scaling it. The most cited reason isn't technology: it's that they automated yesterday's work instead of redesigning tomorrow's. They enabled Copilot in Word and called that adoption. They didn't touch the processes that make the work in Word exist.
Buying Level 4 without going through Level 2 doesn't accelerate maturity. It skips steps, and what gets skipped always comes back as a problem.
Three things consistently differentiate organizations that advance:
One person with explicit ownership of adoption. Not the vendor, not IT generically: someone internal, with business context, who took charge of the "how" after deployment.
Measurement of real usage. Not whether the tool is enabled, but how many people open it, how often, for what. Without that data it's impossible to know what level the organization is at or what it takes to move up.
A concrete use case before scaling. Organizations that start with "Copilot for everyone" without specifying what for get stuck at Level 1. Those that start with "reduce support response time by 30%" have an objective they can measure, adjust, and communicate.
The point
The problem isn't that AI doesn't work. It's that buying and adopting are different things, and the industry has spent two years selling licenses as if they were the same.
Most organizations have the adoption level they have because nobody explicitly defined what the next level is or what it takes to get there. The tool exists. The gap is in the process.
If you want to understand what level your organization is at and what it takes to advance, let's talk. At macareno.net we guide that journey from diagnosis to implementation.
Sources
- Enterprise AI Maturity Index 2026 — ServiceNow
- 70 Enterprise AI Statistics for 2026 — Azumo
- Enterprise AI Adoption Statistics 2026 — Presenc AI
- AI Agent Adoption Statistics 2026 — GoGloby
- State of Enterprise AI 2026 — Arjun Jaggi