In this article, you’ll find out:
- What each of the seven levels of AI maturity looks like in a real business
- A quick self-check to see where your business is today
- Your next step to move up a level
Artificial intelligence is no longer reserved for large enterprises with specialist data teams and million-dollar technology budgets.
A small business can use AI today to write emails, analyse documents, prepare reports, respond to customers, research markets, organise information and automate repetitive work.
But there is a big difference between using AI and building an AI-enabled business.
One employee asking ChatGPT to write an email is AI adoption. An AI workflow that reads incoming enquiries, classifies them, prepares a response, updates the CRM and sends the right cases to a human is something very different.
So where does your business stand? This 7 Levels of AI Maturity model is designed to help business owners and leaders answer that question. The key question is not simply, “Can AI do this?” It is:
“Should AI do this, how should it do it, and where must a person remain accountable?”
That distinction matters more and more as AI moves from experimentation into business-critical workflows.
A useful lens for responsible AI adoption is the 4D AI Fluency Framework, developed by Rick Dakan, Joseph Feller and Anthropic (framework summary, released under CC BY-NC-SA 4.0). It was written to describe how individuals work well with AI. In this article I’ve adapted it to business workflows:
- Delegation — What should we hand over to AI, and what should stay with people?
- Description — How exactly should the workflow work?
- Discernment — How do we test, evaluate and improve the output?
- Diligence — How do we manage risk, review, accountability and transparency?
The seven-level model below is not a scientific or industry-standard maturity scale. It is a practical business framework for understanding how deeply AI has become part of the way an organisation works. In practice, the levels blend into each other — most businesses sit across two of them.
Level 1: No AI Use at All
At Level 1, work is still largely manual. Employees write emails from scratch, search spreadsheets for answers, compile reports by hand, copy information between systems and rely heavily on personal knowledge, inboxes and informal conversations.
There is nothing inherently wrong with being at Level 1. A business may have good reasons for moving slowly: outdated systems, sensitive information, poor data quality, limited budgets, or simply no worthwhile AI use case yet.
The problem is not the absence of AI. The problem is not knowing where repetitive work is consuming valuable human time.
A hospitality business may have staff answering the same questions about opening hours, dietary options and booking policies every day. A property consultancy might manually collect comparable listings and prepare the same style of client update every week. Each task looks small. Repeated hundreds of times, it becomes an operational cost.
What Level 1 looks like
- No approved AI tools or workflows
- Processes depend heavily on individual staff knowledge
- Repetitive work is completed manually
- Little measurement of time spent on administration
- Information scattered across email, spreadsheets and individual computers
The next move: list five recurring tasks from the past week. For each one, ask: How often does it happen? How long does it take? How standardised is it? What information does it need? What happens if it’s done wrong?
This simple exercise is often the beginning of responsible AI adoption.
Level 2: Individual Experimentation
At Level 2, employees start experimenting with AI on their own. One person drafts emails with an AI assistant. Another writes social-media captions. Someone summarises meetings. A manager brainstorms campaign ideas or a first-draft proposal.
This stage can produce quick wins. AI removes the blank page, summarises large amounts of information and speeds up routine tasks.
But adoption is still personal rather than organisational. There are no shared prompts, no agreed tools, no review standards and no clear rules about what information may be entered into AI systems.
A sales consultant may use AI to research prospects and draft outreach. The results may be good — but if the whole process lives inside that consultant’s personal chat history, the company can’t reproduce it, improve it or govern it.
Common risks at Level 2
- Sensitive customer or business information entered into unsuitable tools
- Output quality varies between employees
- AI-generated claims used without fact-checking
- Employees unsure when human review is required
- Valuable AI know-how locked inside individual employees
The opportunity here is to turn experimentation into organisational learning. Ask employees: What task did AI help with? What did you give it? What did it get wrong? What did you have to correct? Would you use the same process again?
The answers show where AI is genuinely useful.
Level 3: Ad Hoc Tool Use Across Teams
At Level 3, AI use spreads across the business. Marketing uses it for content, sales for prospect research, finance to organise invoices, operations to summarise customer feedback, HR to draft internal documents.
The business is clearly using AI — but in a fragmented way. Teams use different tools, build their own processes and enter data in inconsistent formats. Nobody has a full view of which AI systems are in use or what information is flowing through them.
This is AI tool sprawl: plenty of AI activity, but no AI strategy.
A practical response is a simple AI register that records, for each use:
- The AI tool and the team or owner
- The business task it supports and the data involved
- Expected benefit and how success will be measured
- Required human review and known risks
The purpose isn’t bureaucracy. It’s to make useful AI adoption visible, repeatable and manageable.
This is also where the Delegation question becomes real: not every task should be automated. Some can be handled by AI. Some should be AI-assisted with a human decision. Others should stay human-led because they involve sensitive information, significant financial consequences, legal responsibility or complex judgement.
Level 4: Defined Use Cases with Real Owners
Level 4 is where AI adoption becomes intentional. The business stops asking, “Where can we use AI?” and starts asking, “Which specific business problem are we trying to solve?”
A defined use case has a clear objective, owner, workflow, data source, expected outcome and review process.
Instead of:
“We use AI for customer service.”
A defined use case reads:
“Every weekday morning, AI prepares a summary of new customer enquiries, identifies common topics, drafts responses and routes urgent issues to a human service manager.”
That is a process that can be tested and measured, and the owner is responsible for maintaining it, checking performance and escalating problems.
This is where Description matters. An effective AI workflow needs three things described clearly:
- Product: What output should be produced?
- Process: What data and steps should be used?
- Performance: What quality, tone, limits and review rules apply?
For example, a property research business could have AI prepare a first-draft weekly market briefing from approved transaction data, listing feeds and analyst notes. The research lead then checks the figures, removes unsupported claims and approves the final client report.
The lesson at this level: AI implementation is not mainly a tool-selection exercise. It is a workflow-design exercise.
Level 5: Automated Workflows in Daily Use
Level 5 marks the move from occasional AI help to AI-supported workflows built into daily operations. Systems connect to regular sources of information and produce repeatable outputs on a schedule, trigger or business event.
An AI workflow might:
- Categorise inbound leads and prepare follow-up drafts
- Turn meeting notes into tasks and project updates
- Extract invoice details and flag exceptions
- Summarise customer feedback every week
- Produce first drafts of recurring reports
- Pull information from documents into structured systems
The value compounds because people no longer start every task from zero. But risk rises too. A mistake in an AI chat wastes five minutes. A mistake inside an automated workflow can affect hundreds of customers or transactions.
That’s why mature Level 5 workflows include testing, monitoring, clear ownership and human approval wherever the consequences justify it.
Case Study: Prospect Butcher Co.
This example comes from AI Fluency for Small Businesses, a free course created by Anthropic and PayPal with Prospect Butcher Co. and MAKS TIPM Rebuilders.
Prospect Butcher Co. is a two-location whole-animal butcher shop in Brooklyn. Paying suppliers used to be reactive: invoices sat across several billing systems with different payment terms, alongside rent, credit cards and payroll. The co-founder would review invoices in the evening, check cash and decide what could be paid without putting payroll at risk.
The business built an AI-supported weekly payment recommendation. Each Monday, it brought together outstanding invoices, cash position, payment urgency, supplier follow-ups, sales data and upcoming obligations, and produced a supplier-by-supplier recommendation and cash-flow forecast.
Crucially, the AI did not make the payments. The owner made the final decision, submitted payments manually and kept handling supplier relationships personally.
The workflow was also tested against past periods before it was trusted — and those tests caught real problems, such as already-paid invoices reappearing and large invoices being prioritised over smaller suppliers who were actively chasing payment. The rules were corrected before go-live. The result was one focused weekly review instead of hours of scattered evening work.
This case sits on the boundary between Level 5 and Level 6: a scheduled workflow does the gathering, but its output is a recommendation for a decision. The real lesson is not that AI saved time. It’s that the workflow was designed around AI’s strengths while financial accountability stayed with a person.
Level 6: AI Informing Business Decisions
At Level 6, AI becomes an important input into management decisions. Leaders use AI-generated analysis, forecasts, pattern detection and scenario comparisons to understand what is happening and decide what to do next — for example:
- Forecasting demand for staffing or stock
- Spotting customer-service themes across reviews and support messages
- Analysing sales patterns to guide promotions
- Comparing project risks and deadlines
- Running cash-flow scenarios before buying or hiring
- Flagging unusual patterns that deserve management attention
The difference is simple: Level 5: AI helps execute the work. Level 6: AI helps management understand the business.
But AI recommendations should never automatically become decisions. A model can work from incomplete data, miss context, or give a confident answer built on assumptions management would reject. This is where Discernment becomes critical. Ask:
- What data was used, and what was missing?
- What assumptions were made?
- What would change the recommendation?
- How confident should we be?
- Who makes the final decision?
The goal isn’t to make managers follow AI. It’s to give them better information, combined with human judgement.
Level 7: AI Embedded in the Operating Model
At Level 7, AI is no longer a side project. It’s part of how the business runs — across workflows, knowledge management, customer service, research, analysis, planning and possibly new products or services.
Level 7 does not mean everything is automated. Mature organisations often keep some activities deliberately human-led. The difference is that they have a disciplined way to decide where AI belongs and where it doesn’t.
A mature AI-enabled operating model includes:
- Clear AI ownership and governance
- Approved tools and data-handling rules
- Documented AI workflows
- Human approval points for higher-risk work
- Staff training in prompting, verification and escalation
- Measurement of business value and quality
- Monitoring for errors and unexpected behaviour
- Regular testing against real business scenarios
- Clear accountability when AI-supported work goes wrong
This is where Diligence lives, and it matches the wider direction of responsible AI management. The US National Institute of Standards and Technology’s AI Risk Management Framework, for example, organises AI risk around four functions — Govern, Map, Measure and Manage — with ongoing monitoring and accountability throughout an AI system’s life.
At this level, the question is no longer “Which AI tool should we buy?” It’s “How should our business operate differently because AI exists?”
The Real Difference Between AI Adoption and AI Maturity
One of the most common mistakes is confusing AI usage with AI maturity.
A company with 50 employees using ChatGPT every day can still have low AI maturity. Another with just three carefully designed AI workflows may be far more mature. Maturity isn’t about how many tools you use — it’s about how well AI is built into how the business works.
The research points the same way. McKinsey’s State of AI 2026 survey found that only about 6% of organisations qualify as AI “high performers” — and nearly three-quarters of those have fundamentally redesigned their workflows around AI, compared with roughly a quarter of everyone else. McKinsey’s earlier State of AI surveys also linked workflow redesign closely to the value companies got from AI.
Giving employees access to AI is easy. Changing how work gets done is much harder — and that’s where the value is.
Quick Self-Check: Where Are You Today?
Pick the statement that sounds most like your business right now:
- “We don’t really use AI.” — Level 1
- “A few people use it on their own.” — Level 2
- “Most teams use it, but everyone does it differently.” — Level 3
- “We’ve chosen specific problems, each with an owner.” — Level 4
- “AI runs parts of our daily processes.” — Level 5
- “AI analysis feeds our management decisions.” — Level 6
- “AI is part of how we plan, operate and govern.” — Level 7
Don’t try to jump from Level 2 to Level 7. Aim for the next level that makes business sense.
Find Your Next Step
If your business is still experimenting, don’t start with a company-wide transformation. Start with one workflow — a task that:
- Happens regularly
- Has reasonably clear inputs and outputs
- Takes meaningful staff time
- Has a manageable cost of error
- Can be measured
- Has someone who can own it
Then test it. Use real examples. Compare the AI’s output with what an experienced employee considers correct. Record the corrections. Add approval where the consequences justify it. Measure the time saved and the quality of the result.
That’s Discernment and Diligence in practice. AI systems shouldn’t be deployed and forgotten; they need ongoing evaluation, monitoring and clear oversight.
The businesses that get the most from AI won’t necessarily be the ones using the most tools. They’ll be the ones that combine AI capability with human judgement, well-designed workflows, good data, clear accountability and continuous improvement.
So the most useful question isn’t “How much AI are we using?” It’s:
“What is the next business process we can redesign so that people and AI work better together?”
Not sure which workflow to start with? Browse the free AI resources, or book a 20-minute call and we’ll work out your next level together — no pitch, no pressure.
Sources
- Rick Dakan, Joseph Feller and Anthropic, The AI Fluency Framework (CC BY-NC-SA 4.0)
- Anthropic, PayPal, Prospect Butcher Co. and MAKS TIPM Rebuilders, AI Fluency for Small Businesses
- McKinsey & Company, The State of AI
- NIST, AI Risk Management Framework