Feature Selection

Features are the input variables the model uses to predict the outcome. The features available in a typical CRM fall into four categories: deal attributes (size, industry, source), engagement signals (meetings held, emails exchanged, days since last activity), buyer characteristics (title of primary contact, number of stakeholders, decision-maker engaged), and timing patterns (time in current stage, velocity relative to average, close date distance).

1. Deal Attributes Deal size, industry, company size, lead source, product interest, and competitor presence. These are static features — they do not change as the deal progresses. They establish the baseline probability. A $200K enterprise deal in financial services sourced from a referral has a different baseline than a $20K SMB deal from a cold outbound email.
2. Engagement Signals Number of meetings in the last 14 days, email response rate, number of unique contacts engaged, content downloaded, and demo completed. These are dynamic features — they change as the deal progresses. Engagement signals are the strongest predictors of near-term conversion because they reflect current buyer interest, not historical characteristics.
3. Timing Patterns Days in current stage versus average, days since last activity, number of close date pushes, and elapsed days since deal creation. Timing patterns detect momentum and stalling. A deal that has been in Discovery for 3x the average duration has a materially lower propensity than one tracking to average, even if all other features are identical.