Model drift
The gap that opens between a saved model and what the data now says, as the mix of leads and the way they close move on.
Why it matters
A frozen model is what stops the same lead being worth two different amounts on two consecutive days. But frozen forever is a model pricing last year's business.
What goes wrong
Refitting on every upload is the failure in the other direction: a 30-day window one morning and a 90-day window the next produce different multipliers, the same lead is worth two different amounts for no reason anyone can see, and the platform learns from a moving target. The resolution is to fit, freeze, apply, and measure the drift rather than assume it.
What this product does about it
A refit is called for at 20% change, or when any factor enters or leaves the model. Nothing refits itself.
Next to this
How it applies in your trade
The same method, worked through for each kind of business: what a lead is worth, what is known on arrival and what to prepare.
The rest of this step
- Day-0 value
- Close rate
- Average deal size
- Expected value per lead
- Cohort
- Multiplier
- Value cap
- Profit on ad spend (POAS)
- Value spread
- Sample size confidence
- Flat feed
- Early gate
- Lead scoring vs lead valuation
- Propensity modelling
- Predicted lifetime value
- Conversion value rules
- New customer acquisition goal
- Conversion adjustments
- Recalibration
Measure this on your own data
The diagnostic reads a CRM export in your browser and reports your volume, your match rate and the spread between your leads against the thresholds in this glossary. Nothing is uploaded and no account is needed.