How would you design a scoring model from scratch for a B2B SaaS company?

I would start from the sales team, not the platform. I would ask which behaviours have historically preceded a real conversation, then work backwards.

A reasonable starting structure: low points for passive signals (email open, generic page view), moderate points for content engagement (blog subscription, webinar registration), high points for buying-intent signals (pricing page view, demo request, ROI calculator, security documentation). I would use page actions for the high-intent pages and completion actions on forms and custom redirects for the rest.

Two things I always include: negative scoring or decay so a prospect who went quiet six months ago does not stay 'hot' forever, and a careers-page rule that decrements rather than increments, since job seekers otherwise pollute the model. I would also use scoring categories if the company sells several distinct products, so a prospect can be hot for one line and cold for another.

Then I would validate: pull the prospects who converted to opportunity last quarter and check what their score was at the point of conversion. If the model does not separate them from the rest, it needs retuning.