Bad Controls, Colliders and the Adjust-for-Everything Reflex

A colleague estimates the effect of a premium onboarding call (offered to some new B2B accounts by sales reps) on 12-month contract value. Their regression includes: company size, industry, region, number of seats activated in month 1, whether the account opened a support ticket, and the sales rep's identity. The dataset is restricted to accounts that renewed at least once.

1. Identify each control that is problematic and explain the causal structure that makes it so.
2. Explain the sample restriction using a DAG argument.
3. What adjustment set would you use, and what would you add to the analysis to address the confounding you cannot measure?