Insights / Guide / AI Prototyping and MVP Development
GuideAI for small and mid-sized businesses: a practical starting guide
You do not need a data science team to benefit from AI. Start with one painful, repeated task, set simple rules and measure the result.
Many smaller businesses assume AI is for large companies with large budgets and specialist teams. That was true a few years ago. Today the tools are accessible enough that a ten-person team can benefit, provided it starts in the right place and keeps the first project small.
Adoption is already widespread. McKinsey’s 2025 survey found that 88% of organisations now use AI in at least one business function. For a smaller business, the real question is no longer whether to use AI but how to start well, without wasting money or creating new risks.
This guide sets out a simple path: choose one task, set a few rules, use what you already have, measure honestly and know when to ask for help.
Start with a task, not a tool
The most common mistake is to begin with a product. Someone sees a demonstration, buys a licence and then looks for something to use it on. The result is usually a tool that a few people try once and nobody uses after a month.
Start instead with the work. List the tasks that take the most hours each week and follow a recognisable pattern. Good candidates share three features: they are repeated often, the information involved is already digital, and you can tell quickly whether the result is right.
- Answering the same customer questions about prices, delivery times or returns.
- Typing details from emails, PDFs or forms into a spreadsheet or accounting system.
- Building the same weekly or monthly report from several spreadsheets.
- Writing first drafts of quotes, proposals or product descriptions.
Choose one. A clear task with a clear measure beats a vague ambition to use AI across the business. If you cannot decide, pick the task your team complains about most. Its improvement will be noticed.
Be wary of tasks that look attractive but fail one of the three tests. Writing marketing copy is repeated and digital, but judging whether it is good takes time and taste, so the benefit is hard to measure. Forecasting next season’s demand is valuable, but you will not know whether the answer was right for months. These can come later, once you have learned how the tools behave on simpler work.
Three places smaller businesses usually win first
Customer enquiries
AI assistants can draft replies to routine questions using your own price lists, policies and past answers. A person reviews and sends the reply, at least at first. For a small team handling dozens of similar emails a day, this can return several hours a week and make responses more consistent.
Documents
Invoices, quotes, delivery notes and contracts contain the same fields in different layouts. Current tools can read them and extract supplier names, dates, amounts and terms into a structured form. A typical example is a wholesaler that receives supplier invoices as PDFs and re-keys each one into its accounts. Extraction with human checking can cut that work considerably and reduce typing errors.
Reporting
Many owners spend Friday afternoons assembling numbers from sales, stock and finance spreadsheets. AI can summarise those files, highlight changes from last week and write a first draft of the commentary. The owner still checks the figures, but starts from a draft rather than a blank page.
The first project should be small enough to finish in weeks and visible enough that people notice.
Write a one-page rule set
Before anyone uses AI with business information, write down a few rules. One page is enough. The aim is not bureaucracy but clarity, so that staff know what is allowed and customers are protected.
- Which tools may be used, and for which tasks?
- What data must never be entered into a public tool, such as customer personal data, payroll details or confidential contracts?
- Who reviews AI output before it reaches a customer, supplier or regulator?
- How will use be recorded, so you can see what is working and spot problems?
- Who should staff ask when they are unsure?
Be specific about review. AI tools can produce confident answers that are wrong, such as a delivery date that does not exist or a discount you never offered. For anything that reaches a customer, a named person should check the output until you have enough evidence that a particular use is reliable. Even then, spot checks remain sensible.
A short, clear rule set prevents most early problems. Review it after the first project, when you will know far more about how the tools behave in your business.
Budget and skills
Start with tools you already pay for. Many office suites, accounting packages and customer service platforms now include AI features within existing subscriptions or for a modest add-on. Testing these first costs little and avoids adding another system to maintain.
Name one champion: someone curious, organised and respected by colleagues, not necessarily the most technical person. Give them a few hours a week for the first project, and give the wider team two or three hours of practical training on the chosen task.
Training works best when it uses your own material. Rather than a general introduction to AI, show staff how to handle a real customer email, a real supplier invoice or last month’s report with the chosen tool. Ask them to note where it helps and where it gets things wrong. Their notes become the basis for improving the approach, and people trust a tool more when they have helped shape how it is used.
When custom builds make sense
Expensive custom development can wait until a simple version has proved its value. Once you know the task, the volumes and the benefit, a custom solution is far easier to specify and justify.
How to measure
Measurement is what separates a useful experiment from a pleasant distraction. Before you start, record three numbers for the chosen task: the time it takes, the error rate and the response or turnaround time. Even rough estimates from a week of tracking are better than none.
- Record a baseline for one or two weeks before introducing the tool.
- Run the new approach for four to six weeks with the same people and volumes.
- Compare time, errors and turnaround, and ask staff and customers what changed.
- Decide: expand, adjust or stop.
If the numbers have not moved, change the approach or stop. Stopping early is a good result. It costs far less than continuing with something that does not work, and it teaches you what to look for next time.
When to ask for help
Most first projects can be run in-house. Get outside help when the data sits in many disconnected systems, when customers’ personal data is involved, or when AI output will influence decisions about people or money, such as credit, hiring or pricing. These situations carry legal and reputational risks that are hard to judge without experience.
A short readiness assessment can save months. It should review your data, systems, rules and candidate tasks, and give you a prioritised list of what to do first.
A simple next step
This week, list your five most time-consuming repeated tasks and note roughly how many hours each takes. Choose the one that is repeated, digital and easy to check. Write your one-page rules, name a champion and start a six-week trial. By the end of the quarter you will know more about AI in your business than any vendor presentation can tell you.
Source: McKinsey, The state of AI in 2025.
