What an AI Co-Pilot Actually Does in a Boring Wholesale Business

I run a food wholesale business. Fourteen years now. We move dry goods and frozen stock to West-African and Surinamese shops and restaurants across Europe. It is the opposite of a software demo. The margins are thin, the stock spoils, and the day is mostly paperwork, phone calls, and chasing a pallet that should have left yesterday.

So when people ask what my "AI co-pilot" does, the honest answer is: nothing you would put in a launch video. It does not close deals or write poetry. It does the boring middle of the business, the part that used to eat my mornings. Let me walk through a normal Tuesday.

The morning

I open one screen with my coffee. I call it Mission Control. It pulls from everything the business already runs on, and it tells me what happened while I slept.

Overnight it sorted the inboxes. We run 8 email accounts: sales, two warehouses, purchasing, accounts, and a few tied to specific brands. New mail gets read, tagged, and sent to the right person or the right pile. An order confirmation goes one way. A late-delivery complaint goes another and gets flagged for me. By the time I sit down, the noise is already gone.

It also read the ERP. We built a small custom API into our system so the co-pilot can see live numbers: stock levels, open orders, what shipped, what is sitting too long in the cold store. So instead of me opening four screens, it just tells me the three things that moved overnight and look off.

The boring wins that compound

None of this is clever on its own. The point is that it never skips a day. A human assistant routing 8 inboxes will have an off morning. This does not. After a few months, the small stuff that used to slip stops slipping.

The meeting summaries are the same kind of dull and useful. Calls and supplier meetings get transcribed and turned into a short note: the decisions, and who owes what by when. I used to write those myself. More honestly, I did not write them at all and lost the thread two weeks later.

It pings me before I ask

The thing that actually changed how I work is that it does not wait for me to go looking. It watches roughly 10 departments and the numbers under them, and when something drifts it pings me. A margin on a product line slipping. A customer who orders every week going quiet. A stock item aging toward its date. I find out the same day, not at month-end when it is already a loss.

The honest ceiling

Right now about 40% of the operation runs without me touching it. I want to get past 80%. I think that is reachable for the routing, the reporting, the reconciliation, the first-draft replies.

What I am not handing over: pricing a big new customer, deciding which supplier to trust when a container is late, the call to a buyer who is upset. Those are judgment and relationship, and getting them wrong costs too much to automate for the sake of a number. The co-pilot drafts, I decide.

It still gets things wrong. It miscategorizes a supplier email as a customer one. It over-flags, so some mornings I get a nudge about a number that turned out to be fine. I would rather it over-tell me than stay quiet, but it is not finished. I tune it most weeks.

If you run a real business and want to try this

Do not start with a chatbot. Start with the single workflow you repeat most and hate most. For me it was the inbox. Wire it to your real data, not a clean demo copy, because the mess is the whole point. Then resist the urge to build something impressive. Build the thing that quietly removes one chore, prove it works for a month, and add the next one.