Think about the small jobs that interrupt your team’s day: checking an enquiry for missing details, finding the right documents or preparing a follow-up. Each has a recognisable starting point and a result someone can check.
Choose one piece of work
When you explore managed AI, it is tempting to list an entire job description: answering enquiries, quoting complex jobs, handling complaints and updating folders. A whole role mixes routine preparation with decisions that need experience and judgement. Choosing one task makes it easier to agree on what the AI should do and where it should stop.
A practical starting point sits inside the tools your team already uses. An AI employee might prepare replies, summarise information, organise documents or flag missing details. Your people stay in charge of approving customer-facing and sensitive work.
Look for four things
Use these four checks to compare the tasks on your shortlist.
- It happens often. Choose work that repeats several times a week and follows a recognisable pattern. A rare task can take more effort to set up than it saves.
- The information is available. The emails, forms, documents and approved guidance needed to do the work are accessible. If essential context lives only in someone’s head, write it down first.
- The result is clear. Your team agrees on what a finished draft, summary or organised folder should contain. Keep a few good examples to show the standard.
- Someone can check it. A named team member can compare the prepared work with its source, spot mistakes and decide what happens next.
Leave decisions such as waiving a fee or resolving a sensitive complaint with your team. If staff disagree about how a process should work, settle that first. An unclear process is a poor starting task for AI.
An example: following up on missing details
Consider a hypothetical property maintenance business that receives online booking requests. Customers sometimes leave out photos of the repair area or their preferred access times.
The inputs are the submitted form, any attached files and the business’s intake checklist. The AI employee prepares a short job summary, a list of missing items and a draft follow-up email.
If access times or required photos are missing, it names those gaps and adds specific questions to the draft. It does not invent an answer. A complete submission goes to the coordinator for a scheduling decision; it does not become a confirmed booking automatically.
The coordinator checks the summary and questions against the original form, edits the draft if needed and sends it. The task has a clear limit: prepare the follow-up, then wait for approval.
Try completed examples before live work
Start with a small set of recent examples your team has already completed. Five to ten can give you a first look at how the instructions handle ordinary cases and missing information. Include an awkward example, not just the easy ones. A small trial helps you find problems; it does not prove every future case will work.
Count the time spent reviewing and correcting the output as well as preparing it. Record missed details, unnecessary questions and whether the result was useful. Compare that with doing the same task manually. If checking takes just as long, narrow the task or improve the guidance before using it on live work.
Agree on the first task
Before setup, make sure you can answer these questions:
- What starts the task, and exactly what should it produce?
- Where will the AI find the information and rules it needs?
- What should it flag or pass to a person?
- Who will review the work, and where will they find it?
- How will you tell whether it has helped?
Bring a short description and a few completed examples to the first conversation. That gives everyone something concrete to assess and helps keep the starting task manageable.
Related How we set up your AI employee →