One button, one model call

The AI Rating button on the lead form triggers a single call to the language model behind your own AI access and writes the score, the reasoning and the timestamp onto the record — no batch job, no schedule.

Grounded in the lead and its chatter

Designation, company, expected revenue, source, sales team, country, stage and creation date feed the rating, together with the most recent chatter messages — up to 15, capped at 4000 characters.

• Includes internal notes, not only customer correspondence — keep that in mind before a confidential remark is added to a lead.

This customer's own closing rates

Computed on demand from the customer's finished leads — overall, and by source, country, sales team and revenue class, each with its case count, so the model weighs a rate from three cases differently from one built on three hundred. No prior history? The model is told so explicitly instead of assuming a rate that does not exist.

• Archived (lost) leads count too — a query that only saw the active ones would report a misleadingly perfect closing rate.

Enriched company data, when available

If eq_ai_enrichment_crm happens to be installed and the lead has already been enriched, its website, address and phone number join the context. Nothing about this module requires that bridge to be present.

The team keeps control

The rating never touches Odoo's own probability field by itself. It lands in its own field, and only the Apply Rating button transfers it — a run that goes wrong changes nothing about the sales forecast until someone deliberately agrees with it.

Traceable in the chatter

Every rating is posted to the lead's chatter as one entry with score, reasoning and timestamp — built as formatted text, not escaped tags, so a later reviewer can read it as intended.

Odoo's own predictive scoring steps aside

The AI icon, the tooltip and the automated-probability percentage disappear from the views, while Odoo's own computation keeps running untouched underneath. Uninstalling this module brings the original display back unchanged.

No silent failures

A missing AI profile, a failed model call, or an answer that does not parse as the expected {"score": ..., "reason": ...} JSON each raise a clear error naming the cause — instead of leaving a stale rating on the record.