All case studies
Case / 01 · Enterprise AI · Heavy Industry
The CEO Cockpit
A unified enterprise-AI command center for a $6B aluminum producer — collapsing twelve weekly board packs into a single, real-time decision surface.
- Reporting Latency ↓
- 70%
- Asset Base
- $6B
- Time to Production
- 9 wks
Context
The C-suite consumed twelve disconnected reports — production, energy, treasury, ESG — assembled by analysts across three time zones. Decision latency had stretched to a week, and material plant events were being discussed in retrospect rather than acted upon.
Approach
- Mapped the eleven systems of record feeding the existing board pack and ranked them by signal-to-noise.
- Designed a single semantic layer with a domain-specific data model, exposed through a chat-native operator interface.
- Wired retrieval over twenty years of plant telemetry, treasury positions, and counterparty exposures with evals at every join.
- Shipped to four named executives in a closed beta with a weekly cadence review — escalated to the full ExCo at week nine.
Outcome
- Board pack assembly time dropped from 38 hours to under 11.
- First-time correct answers on factual treasury questions reached 96% (eval-set).
- Two material counterparty exposures surfaced in week three that the prior reporting cycle would have missed.