Engineering is 31% ahead of May budget pace
Claude Code usage from Platform and Data Science pushed forecast EOM to $48.7K against a $37K team budget.
One action queue for the signals that matter across cost, compliance, adoption, inventory, and outcomes. 2 critical alerts need attention now, with 2 high-priority items queued behind them.
Sorted by blast radius and time sensitivity. Each alert includes why it fired, who owns it, and the next action.
Claude Code usage from Platform and Data Science pushed forecast EOM to $48.7K against a $37K team budget.
A support analyst pasted a customer transcript containing email, phone, and account metadata into an unmanaged ChatGPT session.
The QBR Builder workflow connected to a local MCP server that is not registered in Inventory and has no retention policy.
Legal Contract Reviewer continues to call claude-3-opus-20240229 despite the approved model policy moving this workflow to Sonnet.
Weekly active AI users in Sales fell from 44% to 29% after the pilot cohort ended, concentrated in Account Management.
The agent completed 212 invoices but downstream edits were 2.8x above its baseline, suggesting poor first-pass quality.
Alerts are deduplicated across modules, then routed by policy owner, blast radius, and whether Proxon can take the first action automatically.
Warn at 80% monthly pace, page at 110% forecast
Block unmanaged sessions and attach evidence packet
Assign owner if usage exceeds 25 calls/week
Notify when active usage falls 10 pts/week