The AI readiness assessment you can run in an afternoon

Most finance teams are asked about AI before they are asked about their work. A vendor demo arrives, a board member forwards an article, and the question becomes which tool to buy. The better first question is cheaper and quicker: which of our recurring tasks could an agent run, and what stops the rest? An AI readiness assessment answers that, and for one finance team it does not need consultants or a survey. It needs an afternoon, three people and a list of tasks.

What an AI readiness assessment should answer

"Are we ready for AI?" is the wrong question, because the answer is always "partly". Readiness is a property of tasks, not of teams. The bank reconciliation might be close. The year end tax provision almost certainly is not.

So the assessment should end with three short lists, not a score out of ten:

1. Tasks worth trying with an agent drafting and a person approving.
2. Tasks that would be worth trying if they were written down properly.
3. Tasks to leave with people for now, and why.

If your afternoon produces those three lists, it has done its job. Everything else is detail.

Before you start: one area, three people, one list

Pick one area of work, not the whole department. Month end close is a good first choice because it repeats on a fixed calendar and everyone knows it. Accounts payable or client reporting work as well.

Book two hours with three people: the person who runs the area, someone who does the work every month, and someone who reviews it. The doer matters most. They know which steps are rule bound and which ones only look that way from a distance.

Bring a list of 15 to 25 recurring tasks in that area, one line each. If you do not have one, spend the first half hour writing it together. Our post on building the recurring task list has a template, but for this exercise a name, a frequency and an owner per task are enough.

Six questions for every task

Go down the list and score each task 0, 1 or 2 on six questions. Keep moving. If a question takes more than a minute of debate, score it 1 and note why.

1. Does it repeat? 2 if it runs weekly or monthly, 1 if quarterly, 0 if yearly or less.
2. Is it done the same way each time? 2 if two people would follow the same steps, 1 if the steps are mostly fixed with some judgement, 0 if it changes every time.
3. Are the inputs steady? 2 if they arrive in the same format from the same places, 1 if they mostly do, 0 if someone has to chase or rebuild them.
4. Is it written down? 2 if a colleague could run it from the manual alone, 1 if notes exist but rely on memory, 0 if nothing exists.
5. Is a mistake contained? 2 if an error is caught before it leaves the team, 1 if it reaches a manager or client but is easy to fix, 0 if it reaches a filing, a payment or a regulator.
6. Can more than one person do it? 2 if three or more, 1 if two, 0 if one.

The last question is there for a reason. A task only one person can do is a key person risk before it is anything else, and a task nobody else understands is very hard to describe to an agent.

We went deeper on the individual signs in five signs a task is ready for an agent. For an afternoon, the six scores are enough.

Reading the result

Add up each row. The maximum is 12. Then sort, but do not just take the top of the list. Look at where the points were lost.

• High totals, with a 2 on mistakes contained. These go on the first list. They are the candidates for an agent that drafts while a person approves. Pick no more than three.
• Lost points mainly on "written down". These go on the second list. The work is ready and the documentation is not. This is usually the longest list, and it is good news: writing a manual is work you control, and it helps the next new colleague as much as any agent.
• Lost points on "done the same way" or "inputs steady". These go on the third list. The task depends on judgement or on messy inputs, and no tool fixes that by itself. Sometimes the right move is to fix the inputs upstream first.
• A 0 on "mistakes contained". Third list, whatever the total. Start where errors are cheap and visible.

As an illustration, a team scoring 20 close tasks might find three for the first list, nine for the second and eight for the third. Your split will differ, but expect the middle list to be the biggest.

Three questions for the team, not the tasks

Before you finish, spend ten minutes on three questions about the team itself.

1. Who owns the task list? If nobody keeps it current, the assessment is a snapshot that will be wrong by spring. Name an owner.
2. Are exceptions written down anywhere? Most of the hours in back office work sit in the cases that do not match. If they live only in people's heads, an agent will hit them blind.
3. Does the team know what happens next? Say plainly that documenting a task may lead to an agent drafting it, and that the aim is to take the repetitive half of the job away first so people spend more time on review, clients and judgement. If the doers in the room do not believe that, the scores they give you will not be honest, and you will not deserve them to be.

What the assessment is not

It is not a buying decision. Nothing in the afternoon depends on which tool you use, and that is deliberate: the same three lists apply whether the agent is a vendor product, an automation tool you already own, or a colleague with a chatbot and a good manual.

It is also not a headcount plan. The output is a list of tasks and the gaps in how they are described. Treat it that way, in the room and afterwards.

This is the decision Beaks is designed to keep live, with readiness scored against every recurring task as the work runs. But an afternoon with a spreadsheet will tell you most of what you need for a first step.

One thing to do this week

Book the two hours. Pick your month end close, bring the three people, and score 20 tasks on the six questions. Leave the room with no more than three tasks to try, and one manual to write before next month's close.