If you're interested in using AI but haven't found time to investigate it, start by giving someone a small, specific task to explore. They need time in their week, access to an approved tool and a colleague who can judge the results.
That person could be you. If your diary makes that unlikely, choose someone who knows the work and wants to improve it. Agree what they will try and when you'll look at the results together.
Pick someone who knows the task
Your first AI champion doesn't need to be a developer. They need to understand the job well enough to notice when an answer looks plausible but is wrong, and to explain what happened to the rest of the team.
Ask who would like to try it. Someone who regularly prepares quotes or sorts enquiries may already have a good idea of where help would be useful. Give them a colleague to work with so the experiment doesn't depend entirely on one person.
An outside adviser can help with choosing tools, setup or training. Having someone inside the business who can explain the process and its exceptions will make that help more useful.
Make room for the work
Agree a short trial, perhaps two weeks with a few protected hours each week. Decide which existing work can wait so this doesn't become another job to fit into the evening.
Set a question you can answer. For example: “Can we prepare a draft response to a routine enquiry faster, while keeping the same standard of accuracy?” That's enough scope for a first experiment.
Before anyone starts, agree:
- Which tool and account they can use, and who approves access.
- What information they can put into it.
- Who will check the output.
- A spending limit and review date.
Make it clear that reporting an unsuccessful trial is welcome. You want to know what happened, including the awkward bits.
Use a tool you can already access
Ask whoever manages your software whether an existing subscription includes an AI feature suitable for the task. Check that your account is approved for the information you intend to use. A familiar brand or a paid plan alone doesn't answer that question.
For a first experiment, you could try a general assistant for drafting, a document question-and-answer tool for finding information, or an AI feature in the application where you already do the work. Choose one so you can spend the trial learning about the task rather than comparing interfaces.
Use public material or invented examples while you're checking the data terms. A free account can be enough to explore an idea, but check its limits before planning work around it. If you need a paid trial, confirm the renewal date and cancellation terms.
Try something small enough to check
Here are three possible starting tasks:
| Task | Material to use | What to check |
|---|---|---|
| Draft an enquiry reply | A fictional enquiry and a public service description | Does the reply answer the question without inventing prices or promises? |
| Summarise a document | A public report you have already read | Are the main points accurate? What did it leave out? |
| Turn notes into actions | Invented meeting notes with clear decisions | Are owners and dates correct? Has it added anything? |
Once an example works, try a less tidy one. Leave out a date or include conflicting information and see whether the tool asks for help or fills the gap itself. Record the time spent checking as well as the time spent generating the answer.
Share what you learned
At the review, ask your champion to show one useful result and one that needed correction. Discuss whether the time saved was worth the effort, what access a real trial would require, and who could repeat it.
Save the instructions, approved source material and checking steps in a shared place. Have the second person try the same task. That will tell you whether the process is ready for someone else to use.
If it works, agree a slightly larger trial with the people who do that work. If it doesn't, narrow the task or leave it there. You can make the next decision from what you've seen, without committing to a company-wide rollout.