The AI-Native Scale: six maturity levels for enterprise AI
The AI-Native Scale: a six-level model that measures AI maturity by how autonomously business processes run without human intervention.
The AI-Native Scale: six maturity levels for enterprise AI
Working out how deeply artificial intelligence is embedded in an organisation's operational fabric takes more than a technical audit. The AI-Native Scale offers a reference framework that places companies in six evolutionary stages — from 0 Manual to 5 AI-Native — measuring how far intelligent systems carry the execution of work. What counts here is not the technology installed, but the ability to complete processes end to end without human intervention.
In short
- Six levels, from 0 to 5. The decisive jump is from 2 to 3: corporate memory appears.
- The level is proved with the RAO (Operational Autonomy Ratio), not with a catalogue of licences.
- Most companies that "already use AI" are at level 1: loose generative tools with no context.
- Each level has a recognisable symptom and one single next step. Skipping a level is the most common cause of pilot fatigue.
- The equivalent numeric score is calculated with the IMAN, from 0 to 100 points.
What is the AI-Native Scale and what is it for?
The NEXTAI definition, closed and citable:
The AI-Native Scale is a six-level maturity model (0 to 5) that classifies a company by the degree to which its business processes are run by artificial intelligence agents over a governed corporate memory. It runs from level 0, where AI touches no process, to level 5, where the operating model is designed around the agents and people govern and decide.
It serves three concrete purposes:
- Diagnosis without self-deception. It forces you to describe the real state, not the intention.
- Ordering the investment. Each level has one single next step; the budget goes there and nowhere else.
- Comparison over time. Measured every six months, it shows whether the company is advancing or only accumulating tools.
What are the 6 levels of the AI-Native Scale?
| Level | Name | What happens in the company | Recognisable symptom | Typical RAO | Next step |
|---|---|---|---|---|---|
| 0 | Manual | AI touches no process. Human work and office software | "We don't use that here yet" | 0% | Inventory processes and data; pick a candidate process |
| 1 | Assisted | People use loose generative AI, with no company context | Personal ChatGPT accounts, uneven results, zero traceability | 0–5% | Define a usage policy and connect AI to real data |
| 2 | Automated | Automated flows with fixed rules; occasional AI | Automations that break on every exception | 5–25% | Build the Enterprise Brain: memory and semantic layer |
| 3 | Contextual | Corporate memory exists: the AI knows the company | Correct, citable answers, but somebody still executes | 25–50% | Deploy the first superagent with an owner and a KPI |
| 4 | Agentic | Superagents run complete processes with supervision | The team reviews exceptions, not runs | 50–80% | Orchestrate several agents and formalise AI governance |
| 5 | AI-Native | The operating model is designed around the agents | New processes are born with an agent assigned | > 80% on the core processes | Continuous improvement, audit and sector expansion |
The RAO ranges in the table are the references NEXTAI uses to validate a declared level, from our experience on implementation projects. They are not an industry standard: they are the criteria we use to audit consistency between the story and the operation.
Why is the jump from level 2 to level 3 the hardest?
Because it is the only one you cannot solve by buying software. From 0 to 2 you advance by contracting tools. From 2 to 3 you have to build something that does not come with a licence: the company's memory.
At level 2 the automation follows fixed rules and breaks as soon as an exception appears: a client on a special price list, a contract with a different clause, a supplier with a lead time agreed by phone. At level 3 that information lives in a queryable layer — documents, contracts, decision history, relationship graph — and the AI can reason with it.
The three tasks that make up the jump:
- Connect sources: ERP, CRM, email, documents, tickets, accounting.
- Structure the meaning: semantic layer, product and client taxonomy, knowledge graph.
- Govern access: permissions by role so each query sees only what it should.
Skipping this level is the cause of Context Debt: the AI answers without knowing the company and every answer has to be checked by hand.
What level is the average company at today?
There is no public statistic scoring companies on this scale; that is precisely the gap NEXTAI Lab aims to fill with its annual study. What does exist are adoption indicators that let you place the distribution.
According to Deloitte data carried in Google's AI Overview for "generative AI for business in Spain" (captured 30/08/2026), 44% of Spanish companies report fast adoption of generative AI and close to 49% give a good part of their workforce access to AI tools, with IT and cybersecurity (22%) and marketing and sales (19%) leading.
Translated onto the Scale: mass access to tools is the literal definition of level 1. Handing licences to half the workforce does not move a company from 1 to 3; it only increases the number of people using AI without context.
How do you move up a level? An example
A representative case, based on NEXTAI projects (company anonymised, figures rounded):
An insurance broker in southern Europe. 62 employees, €9M in annual commissions, 14,000 policies on the books, around 700 claims and 1,200 client queries a month.
- Starting point (level 1): seven people using ChatGPT on their own to draft emails. Zero integration with the policy system.
- Weeks 1–3 (mapping): starting IMAN 22. Two processes are chosen: cover queries and opening a claim.
- Weeks 4–9 (level 3): an Enterprise Brain over policy conditions, covers, claims history and insurer correspondence. The AI now answers citing the specific policy.
- Weeks 10–17 (level 4): a cover-query superagent in production, escalating to a human in doubtful cases.
Results at five months, measured by the broker itself:
- RAO on cover queries: 68%.
- Average client response time: from 4 hours to 11 minutes.
- Queries resolved outside working hours with no person involved: 31% of the monthly total.
- IMAN: from 22 to 54.
The relevant detail: between week 4 and week 9 no agent went into production. That was the stretch that made everything else possible.
What mistakes leave a company stuck at one level?
- Declaring a level without measuring RAO. The most frequent and the most expensive mistake.
- Jumping from 1 to 4. Deploying agents without corporate memory produces agents that are confidently wrong.
- Splitting the budget across five processes. None of them reaches production.
- Measuring active users. An adoption indicator, not an autonomy one.
- Leaving governance until last. At level 4 without governance, the first incident stops the whole programme.
- Not assigning an owner to each agent. An agent with no human owner is neither maintained nor retired.
How NEXTAI does it
NEXTAI uses the AI-Native Scale as the first piece of every project: before proposing any architecture, it places the company at a level and calculates its IMAN. From there, the NEXT-5 Roadmap covers exactly the ground that is missing: phase 1 to map, phase 2 to build the Enterprise Brain and cross into level 3, phase 3 to deploy the first superagents and enter level 4, and phases 4 and 5 to orchestrate and govern. Each level reached is validated with RAO measured per process, not with a declaration. The documentation is handed over so the company can run and audit its own ecosystem.
Frequently asked questions
What level of the AI-Native Scale is my company at if we use ChatGPT daily? Level 1, unless those tools are connected to the company's data. Level 1 is defined exactly that way: people using loose generative AI, with uneven results, no corporate context and no traceability. It is a useful starting point, but it generates neither operational autonomy nor measurable RAO.
Can you jump straight to level 4? Technically you can deploy an agent without corporate memory; what you cannot do is sustain it. Without level 3, the agent lacks the context to decide well and ends up needing human review on almost every run, which means a low RAO. The profitable jump is 2 → 3 → 4, in that order.
How long does it take to move up a level? From our experience on NEXTAI projects, going from level 2 to 3 takes 4 to 8 weeks, and from 3 to 4 between 6 and 12 weeks per process. Reaching level 5 is not measured in weeks: it means redesigning the operating model and usually takes 18 to 36 months.
Is the AI-Native Scale the same as the IMAN? No, they are complementary. The Scale describes the stage qualitatively, in six levels. The IMAN gives a score from 0 to 100 across five axes — Data, Processes, Agency, Governance and Culture — which allows comparison between companies and quarters. Each IMAN band corresponds to a level of the Scale.
Does it work for a company with fewer than 50 employees? Yes. The scale describes processes, not size. A company of 20 people can reach level 4 on a specific process — invoicing, customer service, quotes — before a company of 500, because it has fewer legacy systems and fewer approval layers.
Knowing where your organisation sits is the first step towards real change. Find your exact position with the IMAN or request a free Digital Audit to map a precise route.