decision throughput, 90 days
Read case studyNorthwood wanted proof before trust. They ran two full weeks of dry-run across 2,800 SKUs, watched the rows the agents would have written, and flipped to live with no surprises.
// the deltas
SKUs under management
Dead SKUs vaulted
Dry-run before live
Surprises at flip
A 2,800-SKU catalog nobody had time to clean
Northwood's catalog had accreted for years: discontinued lines still listed, seasonal SKUs never wound down, margin rules living in a Notion doc that drifted from reality. Cleaning it by hand was a quarter of work no one could spare.
They didn't want an agent guessing at it blind, either. They wanted to see what it would do first.
Dry-run as a proof, not a formality
For fourteen days the Catalog Specialist proposed and logged every lifecycle move — archive, throttle, reprice, draft — and wrote nothing. The team read the rows daily, tuned ALLOWED_TRANSITIONS where the agent surfaced a rule that was too strict, and built confidence one review at a time.
When they flipped to live, the rows looked exactly like the dry-run rows had. The first week archived 1,200 dead SKUs that had been quietly dragging the catalog down.
Policy that enforces itself
The lasting change is that margin policy is now enforced in the type system at decision time, stamped on each row as a cost-confidence tier. The Notion doc is gone. What the agent is allowed to do is encoded where it can't drift.
Two weeks of dry-run meant the day we went live was boring. We'd already read every kind of row the agent would write. Margin policy lives in the type system now, not a Notion doc nobody opened.
// your story next
See your own case study in 90 days.
The fastest path to a numbers-driven story is to connect a read-only token. We'll show you exactly what the agents would have done over your last 30 days — before anything writes.
Dry-run by default · Append-only logs · One-click rollback
