Enterprise AI Maturity: the Four Tiers Explained, SHIFTLABS

Guide

An organisation’s AI maturity reads on four tiers, from ad hoc use to an AI-native operating model; your tier is the one your practices demonstrate, not the one your intentions announce. Moving up takes a few precise decisions, not more tools.

6 min read
Updated 15 October 2026

Contents

What AI maturity measures, and what it does not

AI maturity is an organisation’s ability to get a lasting, verifiable effect from AI in its everyday work. It is not the number of tools bought or pilots launched: equipping everyone without changing a single process leaves you far from the third tier.

Use is not maturity. In 2025, 18% of French enterprises with 10 or more employees (excluding agriculture, finance and insurance) said they used at least one AI technology, and 58% of those with 250 or more (Insee). That figure measures use, not where AI sits in processes or how it is governed.

A 2022 MIT CISR survey of 721 companies distinguishes four stages of maturity. Companies in the first two stages had financial performance below their industry’s average, and those in the last two performed above it. That is an observed correlation, not proof that one more tier is enough.

The four tiers, and what you observe at each

This is the scale we use, with the signal that lets you recognise each tier. Look for the proof, not the label.

The four tiers, and what you observe at each
Tier	What the scale says	What you observe
1. Ad hoc use	AI remains occasional, individual and ungoverned.	A few people use consumer tools on their own initiative. No one keeps a list of the uses or of the data involved.
2. Fragmented initiatives	Tools help, but remain local and unshared.	Pilot teams have their own tools. Each department decides alone. A local success is not repeated elsewhere.
3. Integrated AI	AI supports core processes, with human review.	Key processes are described and measured. A shared platform is approved. Every use case has an owner, and human review is traced.
4. AI-native operating model	AI executes at scale; humans validate and arbitrate.	Agents carry out whole tasks. Processes are redesigned around them. Gains are tracked continuously and budgets reallocated.

Scroll the table sideways.

How to place your organisation: six questions, with proof

Answer with what you could show on a Monday morning, not with what is planned: a “yes” without proof counts as a “no”.

The SHIFTLABS express diagnostic covers these six dimensions with twelve questions and four ordered answers, one per tier. It averages them, rounds to the nearest tier (the lower one on a tie) and flags the two lowest dimensions.

Strategy: does a member of the executive team sponsor an AI roadmap, with selection criteria and a budget?
Data: do you know where the data for a first use case sits, who is accountable for it and what state it is in?
Talent: in each business line, do trained people use AI on their actual work?
Process: is a key process described, with the time really spent on it, and does AI have a defined place in it, with human review?
Technology: do teams have a shared tool, approved by IT and connected to your systems (ERP, CRM, messaging)?
Ethics and risk: is a usage charter known to everyone, listing prohibited data and the decisions that stay human?

What it takes to move up each tier

Each move comes down to a few decisions. Buying a tool is not enough: the decisions have to be made, and the evidence kept.

The target tier depends on the expected value, process by process: a solid tier 3 on two or three high-volume processes can be worth more than a tier 4 displayed everywhere.

What it takes to move up each tier
Move	What to decide	Proof it is done
1 → 2	Appoint a sponsor on the executive team. Take stock of the real uses and the data involved. Write a short usage charter.	The list of uses exists, teams know the charter, and two or three priority processes have a measurable success criterion.
2 → 3	Choose a shared tool with IT. Describe and measure a core process before putting AI into it. Set the human review and name an owner for each use case.	A use case is in production, and its gain (time, quality, adoption) is measured against the starting point.
3 → 4	Redesign a core process around an agent, with human validation at every decision point. Tie each use case to an indicator tracked by the executive committee. Keep a register of uses.	Gains are measured continuously. Use cases that do not move the indicator are stopped and their budgets reallocated.

Scroll the table sideways.

Positioning mistakes to avoid

These six mistakes lead organisations to place themselves too high, sometimes too low, and to invest in the wrong place.

Scoring yourself on intentions. An announced roadmap is not a practice: count what runs today.
Mistaking the most advanced team for the organisation. A pioneering department is an island: the organisation sits where most of its processes sit.
Confusing tools with integration. Giving everyone an assistant does not move you to tier 3 if no process has changed.
Forgetting undeclared uses. Individual uses can exist outside any framework; only interviews with the teams bring them to light.
Smoothing over gaps with an average. A 4 in technology and a 1 in governance average 2.5, a figure that describes neither situation.
Scoring yourself alone. A self-assessment reflects one person’s view, often the sponsor’s: have several departments fill it in, then test it against the field.

From a reference point to a plan: five steps

A reference point becomes a plan in five steps, which a SHIFTLABS diagnostic also follows: AI agents interview the teams and measure the actual work, business experts validate every conclusion, and within two weeks you receive the diagnostic, quantified use cases and a roadmap (Pilot offer, from €12k excl. VAT, fixed fee).

Take the express diagnostic (twelve questions, three minutes): an indicative reference point.
Have two or three people from different departments fill it in, and read the gaps as much as the average.
Test that reference point against the field: interviews with the teams, time really spent on one or two processes.
Set the target tier process by process and rank the use cases by impact, feasibility and cost.
Launch one measured use case before opening others.

Frequently asked questions

What is an enterprise’s AI maturity?

It is the ability to get a lasting, measurable effect from AI in everyday work. It is judged across six dimensions (strategy, data, skills, processes, tools, risks), not on the number of pilots launched.

How many levels does an AI maturity model have?

It depends on the framework: MIT CISR and Cegos propose four, other guides five. The number matters little: check that each level can be recognised by observable facts and points to the next decision.

Can you assess your AI maturity without a consultant?

Yes, for a first reference point. The express diagnostic takes three minutes and gives an indicative result. It does, however, reflect the answers of a single person, unverified: before deciding on a budget, test it against interviews with the teams.

Does moving up a tier guarantee gains?

No. The tier describes how AI is set up, not what it earns. Gains come from specific use cases, measured before and after. The MIT study shows an association between maturity and financial performance, not a causal link.

How long does it take to reach the next tier?

There is no reliable standard duration: it depends on the number of processes involved, the state of the data and the availability of the teams. What you can set is the next decision and the proof expected.

Should you aim for tier 4?

Not everywhere. A solid tier 3 on high-volume processes can be a better goal than a general tier 4. The gain simulator helps estimate what a first agent could earn.