Most businesses start with AI in the wrong place, either with the most impressive idea in the room or with a licence for everybody and no particular job in mind. The first job you hand over matters because it decides whether your team comes to trust the technology or quietly stops using it. A good first job is small, frequent and easy to check, and this guide shows how to find one in your own operation.

Make a List of Repeated Work

Ask each team to name the tasks they do every day or every week that follow the same steps each time, and write them down in their own words. You are looking for work such as copying figures between systems, assembling a regular report, answering the same customer questions, matching one document against another or sorting what arrives in a shared inbox. Ten to twenty candidates is plenty, and the people who do the work will produce them faster than any manager could.

Put Each Job Through Five Tests

TestThe question to askWhy it matters
FrequencyDoes it happen at least every week?A job done twice a year never repays the effort of handing it over
PatternCould you write the steps on one page?AI is dependable on work that follows a pattern and unpredictable where every case is new
RecordsIs everything it needs already held in your systems?A job that depends on what somebody remembers cannot be handed over yet
CheckingCan a person tell within a minute whether the result is right?Quick checking is what keeps a person in charge without slowing the work down
Cost of a mistakeIf it goes wrong once, is the damage small and easy to undo?The first job should be one where an error is an inconvenience and nothing worse

A job that passes all five is a strong first choice. One that fails the last test should wait, however much time it would save, because an early mistake with money or with a customer sets the whole effort back.

Good First Jobs and Later Jobs

Usually a good first jobUsually better as a later job
Drafting a regular report from figures you already holdDeciding prices or discounts
Matching supplier files against orders and flagging what does not matchSending anything to a customer before a person has read it
Reading incoming documents and filing them against the right jobApproving payments or refunds
Preparing replies to routine queries for a person to sendAnything involving staff records or decisions about people
Answering staff questions from your own records, with the source shownWork where the rules are different every time

The jobs on the right can all be handed over in time. They come later because by then you will know how the AI behaves with your records, and the approvals around it will already be a habit.

Set the Limits Before You Start

Write down three things for the job you have chosen: what the AI may read, what it produces and who approves the result. For a first job the AI should prepare and a person should decide, with every result approved, edited or rejected by someone who would otherwise have done the task themselves. This is also the moment to agree what may be put into an AI tool and what may not, since customer details and commercial figures need a service that keeps them private and does not use them to train anybody else's model. Our article on what "AI-ready" really means covers the groundwork underneath that decision.

Run It Beside the Old Way

To begin with, keep doing the job the old way as well and compare the two. Note every result the approver changed and why, because those corrections are what improve the instructions, and a pattern in them usually points to a record that is missing or wrong. When the approver has stopped changing anything of substance, the old way can stop.

Decide How You Will Know It Worked

  • Time. How long the job took before, and how long approving it takes now.
  • Corrections. How many results needed changing, and whether that number is falling.
  • Use. Whether the team still relies on it once nobody is watching, which is the most telling of the three.

Choosing the Second Job

The second job is easier, because the limits, the approval habit and the record of what the AI did are already in place. Go back to the list, take the next job that passes the five tests, and consider allowing a little more freedom on the first one if its record supports it. Handing work over in this order, one proven job at a time, is how the routine work comes off a team without anybody losing sight of what the AI is doing.

BY BANKS builds AI and automation into the systems a business already runs, with a person approving what matters and a record of everything the AI does, and we run workshops that help a team find its own first job. If you would like help choosing yours, get in touch.