AI for small business: where to start and what to put in place first
AI is already in most small businesses, used unevenly and mostly unmanaged. Here is a practical way to decide where it adds value, how far to take it and what needs to be in place first.
Ben Chappell · 19 October 2026 · 7 min read
AI is already in the business. People use Copilot, ChatGPT and similar tools to draft, summarise and research, often on their own initiative. That use is uneven and mostly unmanaged: it varies by person and team, and few leaders know which tools are in use, what information goes into them or how the output is checked.
Meanwhile, the advice ranges from prompting tips to large technology programmes, and it is hard to tell what fits a business of your size. The useful questions are simpler: where does AI add value, how far should you take it, and what needs to be in place first?
This guide covers AI used inside the business to improve how work gets done. It does not cover AI as a product or service you sell to customers.
Prefer it as a PDF? Download the full 14-page guide, free, no sign-up.
What AI does, in practical terms
Most business AI tools are built on large language models, which work with language, documents and data. In practice, they do three things:
- Generate. Produces drafts, summaries, ideas and options from a brief. For example, drafting a proposal or summarising a meeting.
- Interpret. Reads and makes sense of documents, messages and data. For example, extracting invoice details or classifying enquiries.
- Act. When connected to business systems, carries out defined steps. For example, updating a customer record or routing a support ticket.
Know its limits. AI can be confidently wrong, and it only knows what it is given. Check its output in proportion to the cost of an error.
Four levels of AI involvement
Every use of AI sits somewhere on two scales: how much AI does (depth), and how far one use reaches across the business (breadth).
| Level | What it means | Example | Who is in control |
|---|---|---|---|
| 1. Assist | A person asks AI for help with a task | Draft an email, summarise a document | The person decides what to use |
| 2. Augment | AI contributes to a workflow; people review its output | Analyse enquiries and recommend responses | People review before anything happens |
| 3. Execute | AI carries out agreed steps within systems | Classify enquiries, update records, route exceptions | Rules set the limits; people handle exceptions |
| 4. Coordinate | AI handles connected steps within set boundaries | Investigate an order delay across systems | People approve consequential actions |
Breadth is how far one use reaches: one person's task, one team or system, several functions or the whole business. Reaching further needs joined-up data; going deeper needs more control. Neither more depth nor more breadth is automatically better. The right position is the one each task needs. (This is a CoveOps organising model, not an industry standard.)
A readiness check for every use of AI
Whatever the use, the same five areas decide whether it works. The minimum is small; what you need grows as AI does more and reaches further.
| Area | The question | Minimum for any use | As you go further |
|---|---|---|---|
| Process | Is the work and its purpose clear? | A short brief: the problem, the benefit and a success measure | A mapped process, agreed rules and exception routes |
| Data | Can AI use the right information? | Clear rules on what can and cannot be shared | Quality data, shared definitions and a central data store |
| Systems | Does it have the right tools and access? | An approved tool with business accounts and settings | Integrations between systems and secure AI access |
| People | Who uses it, checks it and owns it? | Practical training, a checking habit and a named owner | Data owners, ownership across teams and support for change |
| Controls | What can it do without asking? | A short AI use policy | Permissions, approval limits, logs, monitoring and recovery |
Why this is an operations question, not a technology one
Those five areas are the same ones that decide whether anything in the business works well. They map directly onto the six areas I look at in every business:
| AI readiness area | Where it sits in the operation |
|---|---|
| Process | Objectives (the problem and the benefit) and Workflows (how the work gets done) |
| Data | Data & Information |
| Systems | Systems & Technology |
| People | Management, People & Accountability |
| Controls | Performance & Control |
That is why AI is best treated as one part of how the business runs, not a separate project. A business with clear workflows, reliable data and named owners will get more from AI, faster and more safely, than one that starts with the tool.
Where the effort goes
Beyond the minimum, the effort depends on where a use sits:
- Everyday assistance (AI assists, one task or team, such as email drafting). The minimum is enough: an approved tool, a use policy, training and a checking habit.
- Workflow automation (AI acts inside one process, such as enquiry routing). The effort is in defining that process and controlling what AI can do: process, controls and systems.
- Connected insight (AI only advises, but across the business, such as business-wide performance insight). It needs trustworthy, joined-up data and systems, and is often more work than it looks.
- Connected operations (AI acts across teams and systems, such as investigating order delays). Every area matters, plus ownership agreed across functions.
Choose where to start: value against effort
Plot each possible use by the value it would add and the effort it would take.
| Less effort | More effort | |
|---|---|---|
| Higher value | Quick wins: start here. Email drafting, meeting summaries | Bigger bets: plan the groundwork first. Enquiry routing, business insight |
| Lower value | Easy extras: fine for individuals. Social post ideas | Not now: revisit later. A chatbot for rare queries |
The minimum for any AI use
- An approved tool. Business accounts, with your data kept out of model training.
- A one-page AI policy. What can be shared, which tools to use, and check before use.
- Practical training. One session on real tasks: writing good briefs and checking the output.
- A named owner. Someone to ask, who keeps the policy and tool list current.
Three next steps
- See what is happening. Ask each team which AI tools they use, for what, and with what information.
- Put the minimum in place. An approved tool, a one-page AI policy, practical training and a named owner.
- Trial one or two quick wins. Write a short brief for each, name an owner and agree how success will be judged.
How I can help
An AI review is a focused version of the Operations Deep Dive: I map how AI is used across the business today, put the minimum controls in place and identify the one or two trials worth running, with a clear brief, owner and success measure for each. Because it looks at the operation as a whole, it also shows where AI is not the answer, and a simpler process or system change would do more.
See the Operations Deep Dive and AI review, or read more about making automation and AI work as part of wider operational change.
Related reading
Systems · Workflows
Why software does not fix a broken process
How to distinguish a technology gap from unclear ownership, poor workflow design or an adoption problem.
27 July 2026 · 5 min read
Workflows
Your business does not need to be broken to work better
How operational friction accumulates in established SMEs and why “that is how we have always done it” hides cost and risk.
27 July 2026 · 5 min read