One client, in production.
We’d rather show you one engagement in real detail than a wall of logos. This is the work, the numbers, and what it actually changed.
Demand planning that ran on one analyst and a spreadsheet.
The problem
A snack manufacturer supplying Costa Rica’s supermarket chains doesn’t just ship product — it runs replenishment planning on behalf of each chain’s buyer. Every week that meant exporting 20,904 rows from a portal, filtering them by hand in Excel, reconciling product codes that differ per chain, and deciding what each of 122 stores should reorder. One person held the entire process, and the logic lived in her workbook.
What we built
An operations brain: it ingests the raw export, applies the rules the business actually uses, and produces the order file the chain already recognises — with an audit line recording why every item was included or excluded.
We proved it before we replaced anything. The first milestone wasn’t a new feature, it was exact parity — reproducing a real week line for line, so the team could trust it against something they already knew was right.
What it found
Then the interesting part. The legacy rule flagged a product as running low at fewer than five units — a threshold nobody had revisited. Tested against the minimums the business actually sets per product, the same week produced 784 shortage lines instead of 87.
- Rows reviewed weekly
- 20,904
- Stores covered
- 122
- Parity with the manual process
- 696 / 696
- Shortages it couldn't see
- ~700
Roughly 700 shortages a week that the manual method structurally could not see — on shelves that were quietly empty.
Now running in production, chain by chain. The same architecture is what a reorder brain looks like for a retailer, or a recall brain for a clinic — a repeating decision, moved out of one person’s head and into a system.
What's next
The next one lands here.
We publish a case study when the work is live and the client is happy for us to name it — not before.
Let's start by understanding the problem.
Six questions, about two minutes. You'll see what we'd build before you ever speak to us.