Felt important vs actually predictive
Weight ten ICP criteria — company size, trial-usage depth, job title, "fancy logo" — by how predictive you believe each is of a deal closing. Then backtest your exact weighting against 600 synthetic closed-won/closed-lost deals you never saw. The reveal: the criteria that felt important often carry almost no signal, while unglamorous ones do the real work. Runs entirely in your browser (0 uploads, works offline).
How to use it
- Set a weight (0–5) on each of the ten criteria — how much you think each predicts a close.
- Hit Run the backtest — your weighting is scored against a hidden dataset generated from the real drivers.
- Read the bars and the X-ray — each criterion's actual predictive power, and which ones you over- or under-weighted on instinct.
What this clears up (the fundamentals)
- An ICP is an empirical question — "who is our ideal customer" has a measurable answer in your closed-won data, not just a founder's hunch.
- Feeling ≠ signal — impressive-sounding criteria (a recognisable logo, high email opens) frequently don't correlate with closing, while boring behavioural ones (trial depth, time-to-first-value) do.
- Backtesting beats brainstorming — the same weighting you'd defend in a meeting can score barely above a coin flip; measuring it against outcomes is the only way to know.
- Decoys hide in plain sight — some criteria are pure noise. Weighting them costs you accuracy and pulls attention from the ones that matter.
Where it's used
A sales/GTM and data-literacy drill — the fastest way to feel why lead scoring should be fit to data, not assembled from opinions. It's a miatz build-lab concept playable; learn to build it yourself.
FAQ
What does the backtest measure?
It applies your weighting as a scoring model to a labelled dataset and checks how often it would have predicted the right outcome (closed-won vs closed-lost), compared to a naive baseline and to the best possible weighting.
What is "actual predictive power"?
The absolute correlation between each criterion and the closing outcome across the dataset. High-signal criteria correlate strongly with winning; decoys correlate near zero no matter how important they feel.
Where does the data come from?
It's generated on your device from a hidden "true" set of weights plus noise — a controlled world where the right answer is known, so your instinct can be scored against it. Real datasets are messier, but the lesson transfers directly.
Is anything uploaded?
No. The dataset, your weighting and the backtest all run in your browser — nothing is transmitted or stored. Turn off your Wi-Fi and it still works.
Limits
A teaching model with independent, linearly-predictive criteria and a known ground truth — real ICPs have interactions, non-linearities, and selection bias. Use it to build the "measure, don't assume" instinct, then backtest against your own CRM data.
Related
Part of the Demystify Playgrounds. Explore the rest from the Playgrounds home.
Bookmark this page (Ctrl+D, or ⌘D on Mac) — it works offline the next time you need it.