A lead score is a promise: work this one first and you'll do better than working the others. Most teams inherit a score that was fit on a static dataset, shipped in a spreadsheet, and never touched again. Six months later the market has moved, the ICP has shifted, and the score is quietly lying.
The failure isn't the algorithm. It's the assumption that a ranking is a property of the lead rather than a property of the relationship between a lead and a specific business at a specific moment.
We build the Bluestar Score the other way around: a transparent weighted model for the first cold start, then an online learner that nudges those weights every time someone marks a lead as booked, won or lost. The score you see on day one is explainable; the score you see on day ninety is yours.
That has a practical consequence. You can always ask 'why is this lead a 74?' and get a sentence, not a shrug — and you can watch the weights drift as your own outcomes accumulate.