A B2B e-commerce business bringing international food into the US market. They could already build their own tools. What they were missing was the layer that lets a tool run, and keeps it running.
Before
GFM already had someone in-house who could build tools — uncommon in a company this size, and the reason everything that followed was possible at all. The problem was not building them. It was what happened afterwards.
A tool ran on the laptop of the person who built it, so everyone else waited on that person to use it. Operations data sat in separately maintained Excel files, each tool pulling its own, and the same number did not agree across two tools. Every new tool meant explaining the business rules from scratch and reorganising the data again — the tenth tool cost as much to start as the first.
What we did
We did not write those tools. We laid a layer underneath them, so the tools they build themselves can be deployed, shared, and fed clean data.
Tools now deploy in one click, and the whole team has a new one the day it is finished. Operations data moved into a database of their own, paired with API documentation that stays in sync and agent skills — so a new tool arrives already knowing what the business looks like, with nothing to explain.
The number of people who can build tools went from one to three. Two of them do not write code.
What the layer is made of
"AI infrastructure" is not one product. It is a set of things that all have to be in place at once. What was actually built:
Before and after
What runs on it now
GFM built all of these themselves. Our part was making them run, be shared, and reach the right data.
We didn't build their tools. We built what makes every tool easy.
More client work is being written up.
