Skip to content

vadim-koenen/scoring-model-design-patterns

Repository files navigation

Scoring Model Design Patterns

Working patterns for behavior scoring and fit scoring in B2B marketing automation — for Marketo, Pardot/Marketing Cloud Account Engagement, HubSpot, and platforms that follow the same general model.

Most published guides on lead scoring are written by vendors. Vendor guides optimize for a recommendation that sounds defensible in a sales conversation. This repo optimizes for a model that survives contact with a real sales team for more than one quarter.

What's in here

File What it does
01-what-scoring-is-actually-for.md The misconception that breaks most scoring models. What scoring is genuinely useful for, and what it is asked to do but cannot.
02-behavior-scoring-patterns.md Signal selection, weighting, decay rules, anti-decay exemptions, and the boring math of keeping a behavior model calibrated.
03-fit-scoring-patterns.md Firmographic and demographic fit. How to score it without making the model brittle when the ICP shifts.
04-anti-patterns.md The eight ways scoring models go bad. What they look like, why they fail, how to fix or retire them.
templates/scoring-model-canvas.md A one-page canvas for designing a scoring model with marketing and sales in the room.

How to use this

Scoring models do not fail at the math. They fail at agreement on what the model is supposed to do, and at maintenance.

  1. Before any model design: run the canvas in templates/scoring-model-canvas.md with marketing and sales together. Half a working session, not async.
  2. Read 01 together: agree on what scoring is for before debating weights. Most teams skip this and end up arguing about thresholds for a model that does not match what they actually need.
  3. Design behavior and fit separately (02 and 03). Combine them only as the final step. A two-dimensional model (separate behavior score, separate fit score) is more interpretable and easier to debug than a single combined number.
  4. Quarterly: run the model against 04-anti-patterns.md. The patterns there are the dominant failure modes I have audited in real instances.

Why this exists

Lead scoring is the most rewritten and least maintained piece of marketing automation. Most instances have a scoring model that someone built two years ago, that nobody trusts, that everybody works around, and that nobody has the political mandate to retire.

A working scoring model is not a sophisticated one. It is a model both teams understand, can explain, and are willing to act on. The sophistication comes second — and only after the first version has proven the team can agree on something.

Companion repos:

Field notes on lifecycle and scoring as one system: vadim-koenen.github.io/marketo

License

MIT. Fork it, adapt the weights to your data, PR back patterns that have held up in a real instance for at least two quarters.

About

Maintained by Vadim Koenen — marketing automation, RevOps, and GTM systems consultant.

About

Working patterns for behavior and fit scoring in B2B marketing automation — Marketo, Pardot, HubSpot, and platforms that follow the same model.

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors