Northwind · 82% weekly adoption

The challenge

Northwind’s operations team had been pitched two internal AI assistants in eighteen months. Both shipped. Neither stuck. The third attempt had to be different, or the team would tune out the next AI initiative entirely.

The call

The first two attempts had failed because they were generic — chat boxes layered on top of internal docs, with no opinion about what the team actually did. We took the opposite approach: pick the three workflows the ops team ran most often, and build the assistant around them specifically. Generic AI later, if at all.

What we shipped

A copilot scoped to three workflows: shipment exception triage, carrier dispute drafting, and route-change customer comms. Each workflow had its own dedicated UI surface, its own prompts, and its own integration with Northwind’s TMS. The “ask anything” surface was deliberately small — a fallback, not the front door.

We trained two ops leads to extend the prompts and add new workflows. By week six, they had added a fourth themselves.

The handoff pattern we followed is detailed in why AI pilots stall; the post-build enablement plan in leaving your team AI-capable.

The outcome

• 82% weekly active adoption among the ops team within the first month.
• Average shipment exception triage time dropped from 14 minutes to 3.
• The team has shipped two additional workflows on their own since launch.