The model is given context (a thread, a transcript, a CRM record) and asked to produce an artefact from it. Output quality tracks context quality almost linearly, which is why the same underlying model produces generic email in one tool and specific email in another. The teams getting value are generally those who solved the retrieval problem, not those who bought the best model.
Given a discovery transcript, the CRM opportunity and the last six emails, the model drafts a proposal that quotes the buyer's own phrasing of the requirement — rather than one built from a template with the company name substituted in.
Generative AI for sales is the use of large language models to produce sales artefacts — emails, call summaries, proposals, research and CRM notes — from existing customer context.
Almost always a context problem rather than a model problem. A tool with access to the CRM record but not the conversation can only write from firmographics, which is what generic output sounds like.