Everyone has an AI strategy. Nobody has a list of what their team does.

Ask the managing partner of almost any finance firm whether they have an AI strategy and you will get a confident yes. There is a working group, a pilot, a licence for a chat assistant, maybe a slide with three horizons on it. Then ask a simpler question: can you show me a list of every piece of recurring work your team does each month, how often it runs, who does it and how it is done?

That question usually ends the meeting early.

This is the quiet gap in most AI plans. They start from the tools and work backwards, hoping to find work that fits. The firms that will actually move work to AI start from the other end. They know what the work is.

You cannot hand over work you have not named

Every serious attempt to give a task to an AI agent, or to anyone new for that matter, needs the same four things:

1. What the task is. Not "month end", but "reconcile the three operating bank accounts on the third working day after month end".
2. How often it runs. Daily, weekly, monthly, quarterly, once a year at year end.
3. Who can do it. The person who normally does it, and whoever covers when they are off.
4. How it is done. The checklist, the template, last month's output, the exceptions that only come up in December.

Most teams can answer the first two from memory for the big items. Almost nobody can answer the last two across the whole team. The knowledge exists, but it lives in people's heads, in a shared drive folder called "OLD do not use", and in a spreadsheet one person built in 2019 and nobody else opens.

An agent cannot run on that. Neither can a new hire, which is why onboarding in finance teams still means sitting next to someone for six weeks.

The list is the strategy

Here is the uncomfortable part. The task list is not a step you do before the AI strategy. In practice it is most of the strategy.

Once you have the list, the hard questions answer themselves:

• Which work should go first? The tasks that run often, follow clear rules, are well documented and would not cause much damage if a draft were wrong.
• Which work is a risk today, AI or not? The tasks only one person can do. If that person leaves in February, the close does not happen in March.
• What should we document next? The tasks that would be good candidates for automation if only someone had written down how they are done.
• What would it actually be worth? You cannot put a value on handing over "reconciliations" in general. You can put one on a specific task that takes four hours a month, every month, for every client.

Without the list, every one of these becomes a debate between whoever is most excited about AI and whoever is most worried about it. With the list, it becomes a ranking.

Why nobody has one

If the list is that valuable, why is it so rare? A few honest reasons.

The work does not look like a project. Most project tools are built for work that has a start, an end and a Gantt chart. Recurring back office work has none of those. It just comes back. So it never gets entered anywhere, and the tools that teams do use show only a fraction of what they really do.

Writing it down feels like extra work. Nobody wants to spend a Friday afternoon typing up how they do the payroll check. And documentation written once, in a burst of good intentions, goes stale within a quarter.

People worry about what the list is for. This one matters most. If a team suspects the list is a redundancy plan in disguise, the list will be incomplete, vague and out of date. The only list worth having is one people are happy to keep honest, and that only happens when the first thing it does is make their week easier.

How to build it without a project

The firms that get this right do not run a three month process mapping exercise. They capture the work as it happens.

• Every recurring task gets entered once, with its rule ("third working day after month end"), and from then on the system creates it on the right days.
• The person who does the task records it once while they do it, and that recording becomes the manual attached to the task.
• Time and ownership are measured from the work itself, not estimated in a workshop.

After a couple of months, the list is simply there. Nobody sat down to write it. That is the idea behind Beaks: the inventory is a side effect of running the week, not a separate job on top of it.

What the list makes possible

Once a firm knows exactly what it does, AI stops being a vague ambition and becomes a series of small, specific decisions. This task is ready. That one needs a better manual first. This other one should stay with a person for now because a mistake would be expensive.

It also changes the conversation with the team. Instead of "AI is coming", it becomes "we are going to give the most repetitive half of your month to an assistant, starting with the drafts, and you check its work". People can see exactly which work is moving, and why. That is a very different message, and it is the one that keeps the list honest.

One thing to do this week

Pick one team. Ask each person to write down every task they did last month that they will do again next month, with how long it took and whether anyone else could do it if they were off sick. Do not tidy it up. Just count.

Most leaders are surprised twice: once by how long the list is, and once by how many lines have only one name next to them. Both numbers are the start of a real AI strategy.