Search for "AI maturity levels" and you will most often find the same answer: a five-level scale, from basic awareness to full enterprise transformation. It is a reasonable starting point for placing your organisation. It is also a grid that describes without helping you decide.

AI maturity is an organisation's capacity to create value with artificial intelligence, up to and including AI agents running in production. This article describes it on a five-level synthesis scale, from awareness to transformation. The Koneetiv framework (2026 edition) measures it in 4 stages and 6 dimensions: strategy, data, adoption, industrialisation, skills and governance.

The 5 AI maturity levels: a synthesis scale

The scale below adapts a structure widely shared online and attributed to Gartner under its former naming (Awareness, Active, Operational, Systemic, Transformational), which the firm has since replaced. The labels and descriptions are our own.

  1. Level 1: Awareness. AI is a topic of research and conversation. A few demos circulate, nothing is structured. Leadership asks questions, nobody owns the subject.
  2. Level 2: Experimentation. PoCs are launched, often by isolated teams. Licences for generative tools multiply without a framework. It is a deceptive level, where activity looks like progress.
  3. Level 3: Operationalisation. One or more use cases run in production, with real users and tracked metrics. Crossing this level is the real wall. Stalling at this step is rarely a technical matter.
  4. Level 4: Systematisation. AI is embedded in core processes, governed and monitored. Deployments follow a repeatable method rather than a series of individual feats.
  5. Level 5: Transformation. AI redefines the operating model, sometimes the business model. New offerings are designed with AI from the start.

To place yourself, look at facts rather than intentions: how many use cases run every day with real users, and which business metric moves when they work. A company that has rolled out assistant licences without instrumenting a single process is still at level 2, whatever the board presentation says.

What the scale gives a leadership team, and where it stops

The scale has real merit: it gives the board, the CIO and business teams a shared vocabulary. Its limit is structural, because it describes states without pointing to what gets you out of them. Knowing you are "level 2" tells you neither why you are stuck there nor where to start.

Gartner, MIT Sloan and BCG, IBM, Microsoft: what the reference frameworks describe

A handful of public frameworks come up regularly in searches on AI maturity, each built for a specific purpose. The descriptions below follow their public sources, as consulted on 16 September 2026.

Gartner AI Maturity Model: maturity stages and assessment pillars

Gartner groups organisations into five stages: Foundational (ad hoc, loosely coordinated experimentation), Emerging (early pilots and growing executive interest), Operational (AI embedded in selected processes, with named owners), Scaled (AI deployed across functions, with a measurable return) and Transformational (AI reshapes decision-making and the operating model). The associated Gartner AI Maturity Assessment covers seven pillars: strategy, data, governance, engineering, operating model, culture, and AI product and value. It leads to a prioritised roadmap, and the detailed report is reserved for Gartner clients.

MIT Sloan Management Review and BCG: a survey segmentation

In September 2017, MIT Sloan Management Review and Boston Consulting Group published "Reshaping Business With Artificial Intelligence", based on a global survey of executives and managers. The report sorts organisations along two criteria, their understanding of AI and their level of adoption, and derives four groups from them: Pioneers, Investigators, Experimenters and Passives. The September 2018 edition kept the same segmentation, which mainly serves to read the survey results and to compare yourself with the respondents.

IBM AI Ladder: a data-first ladder

In 2019, IBM was presenting the AI Ladder, notably at its Think conference, as the path for bringing AI into the enterprise. The ladder has four rungs, Collect, Organize, Analyze and Infuse: collect data, organise it into a governed foundation, analyse it, then infuse AI into business processes. IBM summed up this logic as "There's no AI without IA", IA standing for information architecture.

Microsoft: a maturity model for agent adoption

Microsoft publishes an "Agentic AI adoption maturity model" on its Microsoft Learn documentation site, in a version dated 31 March 2026. Based on the Capability Maturity Model, it describes five levels numbered 100 to 500 (Initial, Repeatable, Defined, Capable, Efficient) and crosses them with five capability pillars: AI strategy and experience, business strategy, AI governance and security, technology and data, and organisation and culture. The guidance targets agents built with Microsoft's own tools, such as Copilot Studio, Agent Builder in Microsoft 365 Copilot or Microsoft Foundry.

Choosing a framework for the decision at hand

These grids serve different purposes. Gartner's and Microsoft's are self-assessment and roadmap tools, the latter for agents. The MIT and BCG segmentation helps read a survey, IBM's ladder sequences the work from data through to infusing AI into business processes. Before adopting one, decide what you want it to settle: benchmarking, allocating a budget, or getting a use case into production. Gartner's and Microsoft's roadmaps also aim at the move into production. The Koneetiv framework concentrates its granularity on that move and on the governance of AI agents.

The wall between level 2 and level 3: what the research shows

Several studies published since 2025 measure or forecast what happens to AI projects between the pilot and production.

These figures cover different things (abandoned PoCs, pilots with no financial effect, weak governance), yet they point to the same step: the moment a project leaves the protected space of experimentation and enters real business processes.

The move from level 2 to level 3 is the step where a maturity model should place its granularity, far more than on the line between "systematisation" and "transformation", which only matters to organisations already in production.

Why projects stall on the way to production

The causes are rarely technical. A PoC runs on a hand-cleaned data extract, with no integration into existing systems, carried by a motivated team that covers the gaps. Production asks for something else: access to the real sources, continuous monitoring, an owner who answers for errors, and a budget that outlives the pilot. Until those points are settled, the project remains a demonstration, however good.

The same mistakes come up across very different companies. The first use case gets picked for its demo effect rather than its business value. Teams track activity (number of prompts, active users) instead of a metric the business already follows. Governance is pushed back to the scaling phase, even though it is often what holds up sign-off from the security committee.

What AI agents add to the production requirement

This applies twice over to AI agents in the enterprise. An agent that acts on your systems, triggers actions and handles customer data demands a level of industrialisation and control that a simple conversational assistant never required. You need to decide what it may do on its own, what it must submit for approval, and how each of its decisions will be traced.

Before connecting a first agent, check your agentic AI readiness: permissions, guardrails, decision auditability and where humans sit in the loop. For companies operating in the EU, the AI Act adds another layer. Depending on the risk level of the use case, some of these controls, such as human oversight, logging and technical documentation, become legal requirements.

The Koneetiv framework (2026 edition): 4 stages, 6 dimensions

Hence the choice made in the Koneetiv framework (2026 edition): 4 stages rather than 5. Granularity moves to where it is useful, since each stage corresponds to a different mode of action, and the score is built on 6 dimensions rated out of 100. The thresholds given for each stage are the ones the online assessment applies.

Stage 1: Explorer (score 0 to 25)

AI is discussed, but no practice is in place yet. The risk: staying stuck at the intention stage while the market moves ahead. The priority: set an ambition at the right level and identify two or three high-impact use cases before choosing any tool. At this point, most of the work is getting leadership to agree on what it expects from AI and on who will be accountable for it.

Stage 2: Experimenter (score 26 to 50)

Real usage exists, but it is barely industrialised and barely governed. The classic trap: staying stuck at the PoC and never generating ROI. The priority: pick one use case, push it all the way to production and instrument value measurement from day one. In front of a leadership team, a single use case tracked by a metric the business already watches carries more weight than a string of unmeasured pilots. A Claude integration partner that puts AI agents into production can carry that first move.

Stage 3: Scaler (score 51 to 75)

Agents run in production and governance is starting to take shape. The challenge: scaling without losing control. The priority: formalise governance, monitoring and compliance before multiplying deployments. The questions change in scale: who approves what when several agents run in parallel, how to classify them by the risk of their actions, and how to show an auditor what each one did.

Stage 4: Pilot (score 76 to 100)

AI is steered and creates measured value. The challenge changes in nature: keep the lead and industrialise the ability to ship new use cases without reinventing the method each time. The assessment's scoring is deliberately strict at this end of the scale, rewarding established practices rather than intentions.

How the stages map onto the five-level scale

Mapping this onto the five-level scale is straightforward. Level 1 corresponds to Explorer, level 2 to Experimenter, level 3 to Scaler, and levels 4 and 5 merge into Pilot. The difference between "systematisation" and "transformation" has little bearing on decisions as long as your agents are not in production. The framework therefore concentrates its precision on levels 1 to 3, where the decisions that unblock production are made.

The 6 dimensions: where the scale becomes actionable

A global score is not enough to decide. Two "level 2" organisations can have opposite bottlenecks: one has impeccable data and zero adoption, the other the reverse. The framework rates every organisation on 6 dimensions, each out of 100:

Reading your profile dimension by dimension

The profile matters more than the average. In a finance department, for instance, a good data score (a well-kept ERP, up-to-date master data) can sit alongside a weak skills dimension, because nobody in-house knows how to supervise a reconciliation agent. The action plan will then start with training and with naming an owner, ahead of any new connector.

The reverse situation shows up in sales teams too: assistants used every day, hence a high adoption score, and no written rule on which customer data may be sent to an external tool. The first workstream is then governance, starting with the list of information that must never leave your systems.

Weighting of the global score: strategy and governance

Two dimensions carry extra weight in the global score: strategy and governance. The framework weights them this way because it treats them as conditions for moving into production. Designed well, enterprise AI governance speeds up deployments, because each new agent lands in a framework where the rules are already written.

Place your organisation, then act in the right order

Knowing your theoretical level only matters if the result leads to an action plan. The Koneetiv AI maturity assessment evaluates your 6 dimensions in at most 8 questions, in about three minutes. It is free and starts without an email address; a work email is requested to reveal the result (overall score, stage, dimension radar and action plan) and receive the PDF report. The plan prioritises your three weakest dimensions, adjusted to your sector.

The plan reads in three stages: an action you can launch on your own, a quick first project to run with Koneetiv, then a structural programme. If you are still unsure whether to start, the article on the signals that show a company is ready for AI offers a shorter self-assessment focused on data, sponsorship and business ownership.

The order of progression from one stage to the next

Whatever your starting stage, the rule is the same: you do not climb the maturity ladder by skipping steps. An Explorer starts with strategy, an Experimenter with getting a first use case into production, a Scaler with governing its fleet of agents. The adopting AI in the enterprise section includes the two articles mentioned above on agentic AI readiness and company readiness for AI.