Where should a business start with AI?

A useful first AI project starts with a clear task. Before choosing software, ask your team where information has to be read, rewritten, searched, or summarised repeatedly. Then choose one piece of work that is manageable enough to review properly.

For a business, that could mean preparing an internal meeting summary, drafting a reply from approved information, or turning an enquiry into a checklist of missing details. These are possible starting points, rather than promises that AI will suit every process.

Look at the work before the tool

Choose a recurring task and walk through a recent example with the person who does it. Write down what starts the work, what information is needed, and what a useful result looks like.

Pay attention to the parts that already cause difficulty. If the source information is incomplete or nobody agrees who approves the result, those issues need attention alongside any AI trial.

A short task review should answer:

• Who does the work, and who checks it?
• Where does the information come from?
• Which steps require judgement or approval?
• What mistakes would matter to the business?
• What happens after the task is completed?

This gives you a better brief than a request to “use AI somewhere”.

Choose a trial that people can check

For an initial trial, favour work where your team can recognise a useful answer and spot an error. A draft based on an approved service description is easier to review than an unfamiliar answer on a specialist subject.

An illustrative first trial might be an enquiry summary. Someone provides a sample enquiry and asks AI to identify the requested service, missing information, and questions for a follow-up. A staff member compares the draft against the original before using it.

Keep approval with the person responsible for the work. AI output can contain mistakes or leave out an important detail, so review needs to be part of the process.

Agree how to judge the result

Before the trial, record how the task currently works. You might track the time spent preparing a usable draft, how much correction is needed, or whether required details are consistently included.

Run a small set of representative examples and compare the results with that starting point. Include awkward or incomplete requests, not only the easiest examples. The aim is to find out whether the proposed approach is useful in your actual work.

If checking and correcting the output takes more effort than the original task, revisit the approach. A clearer template, better source information, or a simpler workflow change may help more.

Use a simple go-or-stop checklist

At the end of the trial, agree whether to continue, change the approach, or stop. Check that:

• A named person can verify the result against the source.
• The team knows which information is permitted in the tool.
• Difficult examples have a clear route to a colleague.
• The reviewed result is useful enough to justify the effort.

A trial that exposes a weak process can still be useful. Record what needs to change before committing to a wider rollout.

Prepare the people and information

Decide which information can be used in the selected tool and check the tool's data-handling arrangements before providing business or customer records. Give the team a clear route for asking questions and reporting unexpected results.

Training should use tasks your people recognise. Useful practice includes giving clear instructions, checking outputs against source material, and knowing when to stop and ask a colleague.

Take a practical next step

AIO can help review your current work, identify suitable opportunities, and agree a focused first project. Where training is needed, the AI Opportunity Review helps shape exercises around the business; AIO's AI Essentials for Work training requires that review first.

Explore AIO's consultancy and training services, or tell us about a task you would like to improve.

For a concrete process to review, see how a business can connect enquiries, quotations, and follow-ups.