Get the free AI Starter Map
A practical guide for small businesses that are exploring AI, already using it in day-to-day work or preparing a more defined adoption. Understand the landscape, recognise your current situation and choose the right kind of next step.
Inside, you will:
- understand the difference between AI tools, assisted work, connected workflows, automations and bounded agents
- see where AI may support thinking, customer journeys, operations, knowledge and monitoring
- identify whether your next decision is to explore, choose a priority, prepare a workflow or design a bounded test
- recognise where human judgement, approval and accountability must remain
Find the right next decision for AI in your business.
You may be exploring AI for the first time, using it every day without a coherent direction or preparing one defined piece of work. The AI Starter Map helps you understand the landscape, locate your current situation and choose a practical next step without forcing you onto a maturity ladder.
You do not need to be a beginner to need better direction.
AI experience and adoption readiness are not the same thing. Locate yourself by the decision you need to make, not by how advanced your tools appear.
I am still exploring
You know AI may matter, but you cannot yet name a worthwhile business problem or opportunity. Your next task is to observe where time, quality, access or decision-making repeatedly breaks down.
I already use AI, but the use is scattered
You may use AI for thinking, research, drafting or analysis, but the activity is not yet connected to a clear business priority, shared method or review rhythm. Your next task is to decide where greater coherence would create real value.
I have several possible opportunities
You can see multiple areas where AI might help, but you do not yet know which should come first. Your next task is to compare them by value, frequency, boundedness, risk and ability to test.
I have one opportunity in mind
You can name the work you want to improve. Your next task is to establish whether the workflow, inputs, exceptions, human decisions and success measures are clear enough to prepare or test.
Understand the parts without starting with the technology.
The same AI model can support very different kinds of work. The useful question is not which label sounds most advanced. It is what job needs to improve, how the work happens now and where human judgement must remain.
The underlying system that interprets and generates language, images or other outputs.
The interface you use directly, such as ChatGPT or Claude.
The business information, examples, rules and history that help AI respond usefully rather than generically.
A person uses AI to support thinking, research, drafting or analysis while remaining in control of each step.
A repeatable sequence of work with clear inputs, decisions, hand-offs and outputs.
A defined movement of information or action between systems, usually triggered by an event or schedule.
An AI-enabled role that performs a narrow set of actions using approved tools, information, permissions and human escalation rules.
Different ways AI can support a business.
These are patterns of use, not levels to climb. A business may use several at once, stay with one that works or decide that AI is not appropriate for a particular task.
Thinking and creation
Use AI with a person for research, sense-making, planning, drafting, critique and creative development. The person supplies direction, context and final judgement.
Repeatable assisted work
Turn useful individual practices into consistent prompts, templates, examples and review rules. This can improve quality and continuity without automating the work.
Customer information and enquiry support
Help people find approved information, understand routes and provide useful context before a human conversation. Keep exceptions, advice and relationship-sensitive decisions with people.
Internal workflow and administration
Support repetitive steps such as extracting, organising, summarising, routing or preparing drafts inside a defined process. Human review remains explicit where consequence or judgement is involved.
Knowledge, monitoring and bounded agentic work
Search approved internal knowledge, monitor defined sources or carry out a narrow sequence of actions within clear permissions. Use stronger controls as access, autonomy and consequence increase.
Choose what matters. Then make it work.
Some AI use helps you think about direction: which customers to focus on, which offer to prioritise or where your time creates the most value. Other AI use helps you carry out recurring work: preparing follow-ups, handling routine enquiries, drafting approved content or supporting a defined workflow. Both matter. The mistake is to automate execution before the direction and the work are clear.
Direction
Which customer group should we prioritise?
Execution
How should we qualify and follow up those enquiries consistently?
One meaningful opportunity. One bounded test.
Follow a structured five-step path from observation to learning before expanding autonomy.
Choose
Select a problem or opportunity that is frequent, meaningful and worth changing. If several candidates compete, compare them before designing a solution.
Understand
Map how the work happens now: the people, steps, hand-offs, information, systems, exceptions and delays.
Define human boundary
Decide which judgements, approvals and relationships remain human-owned, and what AI may support within those limits.
Prepare bounded test
Start with one owner, a limited sample, a clear review period and an agreed way to stop or escalate.
Learn before you extend
Review value, quality, failures, user experience and human workload before expanding the practice or adding autonomy.
Which decision are you ready to make?
Select the situation that best matches where you are today to see your honest next action.
Route A: Explore the landscape
Choose this when you cannot yet name a worthwhile problem or opportunity. Observe recurring burdens and valuable decisions before forcing an AI project.
Honest No-Purchase Action
For five working days, record one recurring burden, delay or important decision each time it appears. Note who owns it, how often it occurs and what better would look like. Do not choose a tool yet.
Human review required
If the proposed use touches sensitive data or consequential legal, financial, clinical, employment, welfare or safeguarding decisions, explicit human ownership and specialist review are required before implementation. This guardrail can apply to any route.
What a meaningful opportunity can look like.
These examples show possible forms of adoption. They are not a maturity sequence and the Scorecard will not prescribe one from a broad pain point alone.
Strategic thinking
Compare customer groups, offers or growth options using clear criteria and retained business context.
Drafting and knowledge work
Create a repeatable, human-reviewed practice for proposals, reports, content or research.
Customer information
Answer routine questions from approved information and hand exceptions to a person.
Administration
Extract, organise or route information inside one defined workflow, with review before consequential action.
Knowledge access
Help staff locate and interpret approved internal material while preserving source visibility and escalation.
Monitoring
Check a defined set of sources and prepare a reviewable summary against agreed criteria.
Keep consequential decisions human-owned.
AI can support administration, information retrieval, drafting, pattern recognition and preparation. It should not make final decisions where errors can materially affect a person's rights, safety, health, employment, finances or access to support. Define who reviews the output, who approves action, what must be escalated and how the organisation can stop the process.
Turn orientation into an honest next decision.
Use the free AI Opportunity & Readiness Scorecard when you want to compare several possible opportunities or assess one candidate more closely. It will return an evidence-based next-decision route rather than a maturity label or automatic technology recommendation.
If you are still exploring, you do not need to buy anything. Observe one recurring burden or important decision this week and note what makes it difficult.