SPAR: Experience-Driven Strategy Evolution in Multi-Session B2B Sales
| Source: arXiv AI
Tags: SPAR, 1688, Alibaba, B2B sales, Episodic memory, EMNLP 2026, Sales agents
SPAR, a self-evolving LLM sales agent that treats selling strategies as testable hypotheses, delivered a +15.8% conversion lift in a live A/B test with over 200,000 merchants on Alibaba's 1688 platform.
Details
Most LLM sales agents work within a single session and cannot learn across outreach rounds. SPAR adds a double loop: an inner loop adapts to each merchant via episodic memory, while an outer loop evolves a shared strategy playbook using causal attribution, separating flawed strategies from external factors. No model weights are updated. In a live A/B test on the 1688 platform covering more than 200,000 merchants, SPAR achieved +15.8% conversion lift and +35.8% growth in high-intent merchants. Domain experts rated its sales quality 4.62 out of 5, the highest among compared methods. The work appears in the EMNLP 2026 Industry Track.