The 90-day pilot
Enterprise AI initiatives fail in predictable ways: open-ended proofs of concept, demos on toy data, security brought in last. The pilot is engineered backwards from the question executives actually ask: can we defend this in production?
What this replaces
Open-ended POCs never end
A pilot with an open budget is a science project. Procurement cannot approve it and vendors cannot be held to it.
Toy-data demos prove nothing
A demo on sample data says nothing about your environment, your sensitivity rules or your teams.
Slideware is not evidence
The pilot should end with artifacts your security and audit functions keep — not a deck.
What the governed copilot does
Days 1–15 — Deployment
Appliance installed on your infrastructure; roles configured; kill-switch drill on the record.
Days 16–35 — Knowledge ingestion
Your SOPs, tickets and reports classified, redacted and approved into team-scoped knowledge — including a deliberate secret-block demonstration.
Days 36–55 — Copilot activation
Real users asking real questions; citations, confidence scoring and refusal behaviour validated.
Days 56–75 — Assurance evaluation
A customer-specific golden set measures grounding, citation coverage and refusal recall before operational reliance.
Days 76–90 — Evidence & decision
Signed closeout evidence, measured outcomes and a structured go/no-go conversion decision.
Under governance, always
- Fixed price and fixed scope, agreed before the pilot starts
- Runs on your infrastructure with your data — never a hosted demo
- Ends with evidence your security and audit teams keep
Output is advisory and decision-support only. The system does not execute operational actions, and refuses to recommend production-changing steps.
Deep-dive on the product site: locs.maib.io/apply