kobzevvv / prospects_scoring_input_output

This repo is for standartised input for prospect scoring

usecase

1. Take data from marketing events and other sources and score prospects by chance of won
2. understend what features are the most important to make portret of Ideal Customer Profile (ICP)

marketing events as features to score

marketing channel, campaign, refferer or whatever

visited page as features

page category, some part of page path and exactly the page (for populer page)

standard input

size

scoring should handle with 20k — 100k uniq features

format

prospect with features array

table structure: — prospect_id — features list (comma separated)

example https://docs.google.com/spreadsheets/d/1QMH8kIB26srrJSox4Gl29Vm0X6ywrhJlsCRz5Jb5Uec/edit#gid=2059325635

Important: Features shouldn't be an outcome of the goal completion event (for example: agreement amount)

Historical funnel data

table structure: — prospect_id — is proving started (boolean) — is funnel end event (boolean)

Question: Why so many columns if it could be just "prospect_id" with goal completion? Answer: BC it's straightforward to make a mistake here. With explicit fields it should be less often problems like these:

• wrong sampling (when we have additional subset of prospects that not a part of funnel, we do this via: "is proving started" (boolean)
• causation missing. when goal completion were before the "is proving started" event

example of dataset: https://docs.google.com/spreadsheets/d/1PY3ee3V6hUlAyhj2kZz8-xiU8V5iyvnLECbtse-zRsQ/edit?usp=drive_link