The facts

• ~4,000 employees in scope (M365 Copilot + Copilot Chat)
• ~998k queries / yr → ~1,004 kg CO₂e · ~24,957 L water
• ~60% of measured footprint from unprocured Copilot Chat
• Teaching comparator: bio-PU ~22.8 vs fossil-PU ~5.7 kg CO₂e/kg

The friction

Sustainability commitments do not get a free pass because AI is invisible. We needed honest Scope 3 math for Copilot — and a way to teach people that “green” labels are not automatic wins.

The fix

Built a GHG Protocol Scope 3 Category 1 baseline with Sustainability partners. Named the governance wound: Chat nobody formally approved and nobody can currently turn off. Flagged Copilot Studio agents as a ~30× energy gap still under-measured. Used bio-based vs fossil PU as the assumption-check that sticks.

The finding

An unusual pair: enable the tools and measure their footprint. Scope stayed honest. We know what not to build (product-level EPD theater). Knowing that is part of the credibility.