Where does AI fit in a warehouse? A way to find out without guessing

Most warehouse and distribution companies we talk to want to use AI and are not sure where to begin. The question they feel they ought to answer first is "what do we want it to do?", and it is the wrong first question. Nobody can answer it cold, and guessing at it is how a company ends up buying a demo that never reaches the floor.

The better starting point is something every operation already has: a clear picture of how the work is done today. You do not need an AI strategy to begin. You need an honest description of a shift.

Start from how things work today

Every warehouse runs on a mix of a warehouse system, spreadsheets, emails, paperwork, a dock calendar and a lot of knowledge inside people's heads. Most of it is never written down because nobody needed it to be. The people who do the work know which orders are always trouble, which carrier calls at the wrong time and which stock count cannot be trusted on a Monday.

That knowledge is the raw material. AI does not need a new way of working to be useful; it needs the existing way described well enough that someone can spot where a machine could help. So the first step is to describe a normal day, step by step, in the words of the people who live it.

Walk the floor with the people who do the work

This is not a workshop in a meeting room. It is an hour or two on the floor, during a normal shift, with a supervisor, a receiving clerk, a dock coordinator and whoever answers the client emails. Ask what happens from the moment a truck is booked to the moment the last pallet is put away, and listen for the small detours.

The detours are the point. A clerk prints an email, reads the order off it and types it into the warehouse system. A supervisor checks the pick list against the stock count by hand because the system is sometimes wrong about one aisle. A dock coordinator waits on a carrier to confirm before anyone can plan the afternoon. None of these is a failure; they are how people cope with systems that do not talk to each other.

Look for re-keying, re-checking and waiting

Three patterns show up on almost every floor, and they are the clearest signs of a place where AI fits.

• Re-keying: information already exists somewhere, in an email, a PDF, a scale ticket or a photo, and a person types it into another system. Reading paperwork and emails and turning them into entries is work AI handles well, with a person checking the result.
• Re-checking: a person looks at two sources to confirm they agree, such as a pick list against stock, or a booked appointment against what actually arrived. Comparing what the system says with what the floor sees, and flagging the mismatch, is a natural fit.
• Waiting: someone cannot act until a confirmation, a forecast or an answer comes back. Suggestions that arrive before the question is asked, such as a late order risk or a dock slot around a known peak, remove the wait.

Write each one down with who does it, how often, and what goes wrong when it is done late or wrongly. You will end up with a short list, and that list is worth more than any AI strategy document.

Check the data behind each candidate

For each item on the list, ask where the information comes from and whether a system already holds it. Re-keying from emails needs the emails and the warehouse system's way of accepting new orders, usually through its API (the doorway other software uses to talk to it). Re-checking stock needs the stock table and whatever the floor uses to count. Waiting on a carrier needs the dock calendar and the order book.

Some candidates will turn out to need data that only lives on paper, or in a system nobody can reach. That is useful to know early. It does not rule them out, but it moves them down the list, because the first integration should sit on data that is already there.

Pick one integration, not a programme

The temptation at this point is to bundle the whole list into a programme with a budget and a steering group. Resist it. A programme spreads attention across many things and delivers none of them well, and the floor loses interest long before anything lands.

Choose one item instead: the one where the re-keying, re-checking or waiting is most frequent, the data is reachable, and a supervisor would notice the difference within a week. Build that one, test it with the people who do the work, and roll it out while someone watches. When it holds up through a busy week, the second one picks itself, because the people on the floor will tell you what to do next.

The question is not what AI could do; it is where a person on your floor is re-keying, re-checking or waiting today.

This is the way redfly works with warehouse and distribution companies: you bring what you know about how the operation runs, we walk through it with you and find where AI fits, and we build that first integration into the systems you already run, with the results landing in the redfly Portal for your people to act on.