Business-to-business sales, where companies sell to other companies, usually take several conversations, and a good salesperson remembers each customer and learns what works. Software agents built on large language models, the technology behind chat assistants, handle one conversation at a time and treat every outcome as equally instructive, unable to tell a bad pitch from a customer who would never buy anyway.
SPAR keeps a memory of each merchant plus a shared playbook of tactics written as guesses to be tested, kept or dropped, and checks whether an outcome came from the tactic before revising it.
In a live test on the 1688 wholesale platform with over 200,000 merchants, conversion rose 15.8% and merchants showing strong buying intent rose 35.8%.