A useful AI project can start with a task your team already repeats every week, especially if you can describe what a good result looks like and check the output before you use it. A shop might start with delivery questions, a consultant might turn meeting notes into follow-ups, and a farm might develop a checklist for receiving stock. Choose one task that would save time or improve the work, and make sure someone is responsible for reviewing the result.
1. Draft answers to common customer questions
Give the tool your approved opening hours, delivery terms and returns policy, then ask it to draft short replies for WhatsApp or email using only those details. Once someone has checked the answers, save them for repeat use and send any question outside the policy to a person who can make a decision.
2. Plan a week of useful posts
Start with what is actually in stock, who it helps and how customers can order, then ask for a product explanation, an answer to a common question and a reminder you can post later in the week. Replace generic phrases with details from your own business, such as which sizes are available or when customers need to order for collection.
3. Turn rough notes into a quotation
Provide the work, quantities and terms you have agreed, and ask the tool to organise them into a clear quotation that separates what is included from optional extras. Before sending it, compare each price, date and calculation against your own records, because a well-presented quote can still contain an incorrect total or an agreement you never made.
4. Write checklists for recurring work
Describe how your team opens a shop, receives a delivery or prepares an order, then ask for a short checklist that follows the order in which the work actually happens. Have the person who does the job test it during a normal shift, so they can remove invented steps and add anything the draft missed.
5. Find patterns in customer feedback
Remove names, contact details, order numbers and other identifying details before asking an approved tool to group comments by themes such as delivery time, product size or packaging. Read a sample of the original comments before acting, since a few complaints can point you towards a problem worth investigating without telling you how common it is.
6. Make meetings end with clear actions
Use meeting notes that are appropriate to share with the tool and ask it to draft a list with three columns: the action, the person responsible and the due date. Tell it to flag missing information, then confirm the list with the team so that no one is assigned a task they did not agree to.
7. Create practice exercises for new staff
Use your approved procedures to draft short role plays, questions and examples, such as a conversation in which a new salesperson responds to a customer whose delivery arrived damaged. Walk through the exercise with the employee, show them the approved response and explain when they should ask a manager to step in.
8. Explore stock and sales questions
Use a small table without customer identifiers to ask which stock or sales patterns are worth investigating, then check each suggestion against the original spreadsheet. If a product appears to sell poorly, look at when it was actually available before drawing conclusions about demand or deciding how much to reorder.
9. Make instructions easier to read
Ask the tool to simplify a long instruction or draft a translation for customers or staff, then have someone familiar with the language and the task review it for meaning. Check safety instructions, product claims and commitments especially carefully, since a shorter or smoother sentence can leave out a condition the reader needs.
10. Improve the way you hand over work
Ask for a one-page template that tells the next person what was completed, what is still open, what needs a decision and where to find the relevant records. Try it for a week, then ask the people receiving the handover what they still have to chase and use their answers to improve the template.
Try one workflow this week
Choose a recurring task and save an example of how you do it today, so you can compare the time taken, the corrections needed and the usefulness of the finished work when you try AI. Keep the new approach if it helps and someone can reliably check the result, and leave customer records, passwords and confidential documents out of tools that your business has not approved for that data.
A starter prompt: “Using only the facts below, write a short reply in plain language and put any questions about missing information at the end.”
