An AI Maturity Model for SMEs

Why enterprise maturity models do not fit

Consultancy maturity models tend to have six dimensions, five levels and a radar chart. They assume a chief data officer, a data platform team and a transformation budget. Applied to a 70-person manufacturer, they produce a depressing score and no useful advice.

SMEs need something simpler. Where are we, honestly? What is stopping us getting further? What is the next move that actually pays back? The model below is the one we use in first conversations at SpiderHunts. It is deliberately blunt, and most businesses place themselves one stage higher than they are.

The five stages at a glance

Stage 1 and 2: from scattered use to managed tools

At stage one, AI is already being used, you just cannot see it. Staff draft emails, summarise documents and write code snippets with whatever tool they prefer. Some of it is excellent. Some of it involves client data pasted into consumer accounts.

The move to stage two is not technical. Choose approved tools for business use, write a short policy, find out what people are already doing well and share it. This is cheap and fast, and it reduces risk more than anything else you will do this year.

Signs you are genuinely at stage two:

• You can name the approved AI tools and which data each may handle
• Business accounts have replaced personal ones for work use
• Someone owns AI tools and policy, even part-time
• There is a way to request a new tool that gets an answer within a couple of weeks

Many businesses stall here for a long time, and that is not always wrong. Stage two delivers real individual productivity. It just does not change how the business runs.

Stage 3: the first integrated project

Stage three means an AI system is wired into a real process with real data: supplier invoices read into the finance system, enquiries classified and routed in the CRM, a demand forecast feeding purchasing. It is the first time AI output affects operations without someone copying and pasting it.

This is where most SMEs hit data problems. Customer records are duplicated, product codes differ between systems, and the history you need to train or evaluate a model lives in spreadsheets. The project becomes half a data project, which is normal and should be budgeted for.

Our advice for a first integrated project: pick a boring, high-volume process where a human can review uncertain cases, measure the baseline before starting, and treat the plumbing (integration, logging, exception handling) as the main deliverable. It is the work we do most often through our AI integration service, and the first project sets patterns for everything after.

Stage 4: repeatable delivery

A business reaches stage four when the second and third AI projects are faster and cheaper than the first, because the lessons were kept. That usually means:

• A standard way to test accuracy on real cases before go-live
• Logging and monitoring that every AI system uses
• A measurement template with baselines and review dates
• A prioritised list of AI ideas, sorted by expected value and readiness
• Data that has been cleaned up once and kept clean
• Clear ownership of each live system after the project team moves on

The blocker at this stage is usually prioritisation. Once people see AI working, ideas multiply faster than capacity. Businesses that manage their ideas as a portfolio of quick wins and bigger bets tend to progress. Those that start everything tend to finish little.

Stage 5: operational AI

At stage five, AI is part of how the business operates rather than a series of projects. Models and agents have owners, budgets, monitoring and review cycles like any other critical system. Some customer-facing products or services depend on AI. Governance is routine rather than reactive, including on regulatory questions such as the EU AI Act where relevant.

Few SMEs are here, and not every SME should aim for it. A professional services firm may get most of its available value at stage four. A software company or an operations-heavy distributor may find stage five is where the competitive advantage sits.

Maturity is not the goal. The goal is getting the value available to your business, and for some businesses that is stage two done well.

How to assess yourself honestly

Answer these without generosity:

1. Could you produce a complete list of AI tools and uses in the business by Friday?
2. Is there at least one AI system connected to your core business data that has run for three months?
3. Do you have a measured before-and-after figure for any AI project?
4. Was your most recent AI project noticeably quicker to deliver than your first?
5. Does every live AI system have a named owner who would notice if it started performing badly?

One 'no' is enough to place you below the stage that question relates to. If the first answer is no, you are at stage one whatever else is happening.

Moving up without skipping stages

The most common and expensive mistake is jumping from stage one or two directly to an ambitious stage-five idea: a customer-facing agent, a pricing engine, an AI-driven product. Without the data, delivery habits and measurement of stages three and four, those projects usually stall, and the business concludes AI does not work for it.

Take the next stage, not the most exciting one. For many SMEs that means a plain twelve-month plan with one integrated project and one foundation piece of data work. Our guide to writing a twelve-month AI roadmap shows how to set that out, and if you are right at the beginning, the small business AI adoption roadmap covers the first year in more detail.

Frequently asked questions