# How to run an AI maturity assessment: a 5-dimension self-check

> An AI maturity assessment you can run yourself: 5 observable dimensions, 3 levels each in verifiable behaviors, and which level each ambition demands.

- Canonical: https://arkatai.com/en/ai-maturity-assessment/
- Site: Arkatai (https://arkatai.com) — agentic operations as a service
- Language: en
- Published: 2026-07-19

---


AI maturity in a company isn't measured by how many licenses you hold or how many people have tried a chatbot. It's measured by whether your operation is in shape for an agent to run real work with guarantees. Here's a self-assessment across five observable dimensions, each with three levels described in behaviors you can verify by looking at your own house, not in vague brochure labels.

I skip the consultancy maturity models with five grandly named phases, because they don't help you decide anything. What helps is asking yourself, dimension by dimension, what level you're really at and what level your ambition demands. That last part is the key: you don't need the top level in everything. You need the level your specific goal requires.

## The five dimensions

When I assess whether an operation is ready for agents, I look at these five pieces. They're not optional or interchangeable: a chain is only as strong as its weakest link, and I've watched projects with excellent data fall over for lack of traces.

| Dimension | Level 1 (initial) | Level 2 (underway) | Level 3 (operable) |
|---|---|---|---|
| Mapped processes | The process lives in two people's heads and in email | Documentation exists, but doesn't fully match what's executed | Rules, exceptions and thresholds are written and match the real operation |
| Accessible data | Data sits in silos and is requested by email | Reachable with manual effort; uneven quality | A system can read and write it via integration, with defined permissions |
| Controls and traces | No record of who decided what or why | Partial logs, reviewed only when something breaks | Every action leaves a queryable trace and permission limits are explicit |
| Evaluation capability | "It works" is decided by impression | Odd cases get tested, with no threshold or repetition | Test cases with known answers exist and quality thresholds are measured |
| Maintenance structure | No one answers for the process when the business changes | An informal owner reviews it when they can | An assigned owner and a defined cycle to review rules and evaluations exist |

Read it as a diagnosis, not an exam. Mark your real level in each row and keep the lowest one: that's what governs what you can do today.

## What each dimension looks at

**Mapped processes.** An agent only runs what's defined. If your process lives in unwritten habits, the first job isn't technical: it's making it explicit. The gap between level 2 and level 3 is subtle and decisive: having documentation isn't enough if the documentation describes the approved process while the operation does something else. How that definition becomes something executable is what I develop in [the operating model as code](/en/operating-model-as-code/).

**Accessible data.** The agent works inside your ERP, your email and your platforms. Level 3 doesn't mean owning a perfect data lake. It means a system can read and write what the process needs, with concrete permissions. Which data you actually need, and which is an excuse not to start, I lay out in [what data an AI agent needs](/en/data-requirements-ai-agents/).

**Controls and traces.** Without a record of what the agent decides, there's no audit, and without an audit a serious board shouldn't approve a deployment. The jump to level 3 is having explicit permission limits (what it can read, what it can write, what's off-limits) and a trace of every action. The concrete implementation is in [AI agent permissions and controls](/en/ai-agent-permissions-controls/).

**Evaluation capability.** This is the dimension most companies underrate. Before giving an agent real work, you have to measure how it does it against cases with known answers and quality thresholds. Without this, "it works" is an opinion. The detail of how it's measured is in [AI agent evaluation](/en/ai-agent-evaluation/).

**Maintenance structure.** An agent system without maintenance degrades like a machine without it: models change, the process changes, and evaluations have to be re-run. Level 3 is having an assigned owner and a review cycle, not a project with an end date. This dimension decides whether the rest holds up over time.

## Which level each ambition demands

The expensive mistake is chasing level 3 across all five dimensions before you start. You don't need to. What you need depends on your goal.

- **Automate a bounded, low-risk process** (say, extracting data from documents to feed a system): level 2 across most dimensions and level 3 in accessible data and traces is enough. Human supervision covers the rest while you level up.
- **Run a process tied to revenue or margin** (reconciliation, order tracking, support cases): here you want level 3 in mapped processes, controls and evaluation. This is where a mistake is costly and where the trace stops being optional.
- **Become an organization that redesigns its operating model around agents**: this demands sustained level 3 across all five, and above all in maintenance structure, because the change advances process by process over years. What that redesign means I cover in [the AI-first organization](/en/ai-first-organization/), and how the split of responsibilities changes in [the AI operating model](/en/ai-operating-model/).

In practice, pin down your realistic ambition for the next twelve months, see which dimension falls below the level that ambition demands, and start by raising that one. This self-assessment diagnoses the starting point. What to do before accepting a deployment I review in [prepare your company for agents](/en/prepare-your-company-for-agents/), and the full picture of adoption is in [the pillar article on AI agents for business](/en/ai-agents-for-business/).

## Frequently Asked Questions

### How do I assess my company's AI maturity without a consultant?

With this five-dimension self-check: mapped processes, accessible data, controls and traces, evaluation capability and maintenance structure. Mark your real level in each using the verifiable behaviors in the table and keep the lowest: that governs what you can do today.

### Do I need the top level in every dimension to start?

No, and chasing it before you begin is the most expensive mistake. The level you need depends on your goal: a bounded, low-risk process asks for less than running a margin-linked process. Pin down your ambition and raise only the dimensions it demands.

### Which dimension do most companies underrate?

Evaluation capability. Many companies decide an agent "works" by impression, with no test cases with known answers and no quality thresholds. Without that measurement there's no serious way to claim the system does its job well, and the risk surfaces in production.

### How is this different from preparing the company for agents?

This self-assessment tells you what level you're at. Preparing the company is the work of raising the level where needed before a specific deployment. One is the diagnosis and the other is the action that follows from it. It's worth doing them in that order.