Before choosing an AI tool, map the work

There's always another AI tool to try. The demo looks good, a colleague sends you a link, and you can already think of three ways it might help. Choosing one feels like progress. If your team is already using several tools, the same question comes back: which of them is actually making the work better?

We make room for one other thing first: sit down with someone doing the work and follow a recent job from beginning to end. You'll get a much better idea of what the tool needs to do.

You might find a task AI could take off their hands. You might discover that everyone is waiting for an answer only one person can give. Either way, you've found somewhere useful to start.

Follow one job all the way through

Pick something familiar: preparing a quote, handing over a new customer or getting a piece of work approved. Choose a job that happens often enough to examine and is small enough to follow.

Let's use a made-up customer handover. The agreement is signed, and the delivery team needs to know what was promised, who's involved and when the work should begin. The information is spread across the agreement, meeting notes and a few emails. Someone pulls it together into a brief.

Ask that person to show you the last handover they prepared. Where did they look first? What did they have to chase? Which part needed a conversation? Then ask the person receiving the brief what they still had to find out.

Draw the steps on a page as you talk. Include the waiting, checking and trips back to an earlier step. A sketch with a few corrections is fine. It needs to help the people in the room recognise their work.

The interesting bits are often between the boxes

A box labelled “prepare brief” hides quite a lot. Someone collects the facts, makes sense of the conversation, notices a contradiction and decides whether the brief is ready. Those activities need different kinds of help.

Collecting the same fields from the same systems may suit ordinary automation. Turning a long conversation into a first draft may be a useful AI experiment. Resolving a promise that doesn't match the agreement needs someone with enough context and authority to make the call.

Pay attention when someone says, “Usually we do this, unless...” That's a part of the process worth understanding. Where does a disagreement go? Who can change a date? What happens when the person who normally sorts it out is away?

This is what we mean by thinking in systems: looking at how one person's work affects the next person's work. If we speed up the brief but leave delivery with more questions, we need to count those questions too.

Give the small fixes a chance

Once the work is on the page, ask what makes it harder than it needs to be. Perhaps a question asked earlier would prevent three follow-up emails. Perhaps everyone needs an agreed place to find the current customer requirements.

Make those changes when they help. They may be enough to solve the immediate problem. They may also make an AI tool more useful, because it will have clearer information to work with.

We judge the first step by how much it helps the team. There is plenty of room to be ambitious about AI while choosing a modest piece of work to begin with.

Try something people can actually use

Back to the handover. Suppose gathering and organising the information takes too long, while the final decisions need an experienced account manager.

A useful first version could prepare a draft from approved sources, link important details back to their source and list anything missing. Show that draft to the people who will use it. They can tell you whether its order makes sense, which details matter and what would help them check it.

A simple draft might contain three labelled items: a confirmed commitment with its agreement reference; a proposed date still awaiting a decision; and a missing detail with someone responsible for finding it. This is an illustrative format, not a client example. The reviewer can see what needs checking without searching through polished prose.

Keep its job clear. It prepares the brief; the account manager resolves contradictions and confirms commitments before passing it on. Write down what should happen if a source is unavailable or the draft gets something wrong. Give the team an easy way to flag a problem and carry on with the existing process.

Before the trial, look at a few recent handovers. Record preparation time, checking time and the missing details delivery had to chase. Use the same measures during the trial, alongside what the people involved tell you. The question is whether the handover becomes easier to prepare and more useful to receive. There's more on that in how to tell whether AI helped.

Leave with a next step

Keep the map close to the work and change it when you learn something. It should help someone understand what happens, who decides and where to ask for help.

That gives you a useful foundation for AI agents too. As you consider handing over more steps, you'll already understand the information they need, the decisions they can make and how to check the result.

You can begin with the next job on your team's desk. Ask the person doing it to walk you through it, and invite the person who receives it to join you. Leave with one change worth trying and someone responsible for checking whether it helped.

If you'd like a hand finding that first change, tell us what you're working through.