If Odoo’s lead scoring feels underwhelming, the first thing I check is not the model setting.
I check the CRM discipline around it.
Predictive scoring learns from whatever the pipeline teaches it
That is the uncomfortable truth. If the pipeline is inconsistent, incomplete, or barely maintained, the scoring model is not learning from signal. It is learning from noise.
Common issues look like this:
- stages are used inconsistently
- win-loss reasons are missing
- duplicates pile up
- activity history is thin
- qualification rules shift from rep to rep
At that point, the score is not the root problem. It is a reflection of the root problem.
Clean process improves the model
The fix is usually boring in the best possible way:
- tighten stage definitions
- make next-step discipline visible
- require meaningful win-loss data
- clean up duplicates
- standardize qualification expectations
That is how the AI layer gets better without turning into a science project.
Outside lead sources shape the score too
This matters even more when leads are enriched from outside Odoo through forms, inboxes, ad platforms, call notes, or connected systems. If weak data enters at the edges, the model sees weak patterns before anyone notices it in the dashboard.
Good AI still depends on good process. Good process still depends on clean data. Odoo is no exception.