Where I would start with AI in an ecommerce team

I'd start an AI project with a piece of work the team already understands. That gives us something concrete to compare against, and the people doing the work can usually tell quite quickly whether the result is useful.

A regular report or the first pass through a set of meeting notes might be a reasonable place to look. The right choice depends on the team, the information involved and how much checking the output needs.

Understand the existing work

Before building anything, I'd ask the person doing the job to walk through it. What information do they collect? What do they check? Where do they use judgement, and what happens after they finish?

That conversation often reveals steps that aren't written down. Someone knows that a particular number needs checking against another report, or that a customer name is recorded differently in two systems.

Those details belong in the workflow. If we leave them out, we may produce something that looks finished while giving the team more work to correct.

Keep the first job small enough to inspect

For an initial test, I'd choose an output a person can review without too much effort. A short summary with links back to the source is easier to inspect than a long report assembled from information nobody has time to check.

I'd also decide which information the system is allowed to use. Client information and access permissions need to be understood before anything is connected or uploaded.

The aim at this stage is to learn where it helps and where it makes mistakes. I'd keep a record of both, because an impressive first result tells us very little about how it will behave across the ordinary work of a week.

Put the review into the process

Human review needs an owner and a clear purpose. Asking someone to glance at the output is unlikely to tell us much. I'd want them to know which numbers, statements or decisions need checking and where to find the source.

If the output would change an account, contact a client or create a commitment, I'd keep a person's approval before that action. A draft and an action are different stages of the work.

Measure the work after checking

The time saved is the time left after reviewing and correcting the result. I'd also ask whether the team can understand what happened when something looks wrong.

If the process takes longer to supervise than to do, that is useful to know early. We can narrow the task or choose another one.

For me, a worthwhile first project gives the team a repeatable piece of help and a clear way to check it. Once that works, there is a better basis for deciding what to build next.

Work through it

If this is the question in front of you, my free worksheet gives you a structure to work through it. It comes with my newsletter on Amazon, ecommerce and practical AI.