Define it before you compute it

Tuesday, 14:30, a whiteboard. Three people, three conversion rates.

• Dana: won ÷ (won + lost). "Open deals haven't converted or failed yet."
• Lena: won ÷ created. "That's what the CRM report has always shown."
• Tomás: won amount ÷ closed amount. "I don't care how many deals. I care how many dollars."

All three are legitimate. All three will produce a different chart. The Data PM's job is not to pick the winner in the meeting; it is to put every definition on the Metric Dependency Tree with a name, a formula, and an owner, so the fight happens once instead of every month.

Raj shows you something else: deal stages are overwritten in the CRM, but there is an append-only deal_stage_history log. That changes what you can measure. Stage-to-stage conversion, time in stage, funnel drop-off: none of it exists on the deals table alone.

This week: draft the logical model (entities, dimensions, measures, relationships, SLOs) and compute the funnel from the history log. The SQL gets real: CTEs, self-joins on the history, and your first window function.