Great paper from Alibaba on self-evolving agents, with a live A/B test across 200,000+ merchants.
(bookmark it)
This is another brilliant example of leveraging episodic memory with agents and the importance of attribution.
SPAR runs B2B sales outreach on Alibaba's 1688 platform. Conversion rose 15.8%, and high-intent merchant growth rose 35.8%.
The agent learns at two levels without updating weights. An inner loop keeps episodic memory for each merchant across sessions. An outer loop maintains a shared playbook of selling strategies, each written as a testable hypothesis.
The key step is attribution. Before changing the playbook, SPAR decides whether a lost deal came from a bad strategy or an outside factor, so a deal lost to outside factors doesn't change a strategy that works.
Domain experts rated it 4.62 out of 5 on sales quality, the highest of all methods tested.
If you are building agents that learn from production outcomes, it might be worth adding an attribution step before writing any lesson to memory.
https://arxiv.org/abs/2611.01695
Chat with Paper: https://academy.dair.ai/papers/spar-experience-driven-strategy-evolution-in-multi-session-b2b-sales-2611.01695