SPAR: Experience-Driven Strategy Evolution in Multi-Session B2B Sales
Organizations: South China University of Technology · Alibaba Group
Abstract
B2B outbound sales involves high transaction values and relationship-based trust, so closing a deal typically requires multiple rounds of outreach building on prior context. Existing LLM-based sales agents operate within single sessions and cannot carry forward context or refine strategies from experience. Methods that do learn from experience store lessons as unstructured text and treat all outcomes as equally informative, with no mechanism to distinguish a flawed strategy from an external factor. We propose SPAR, a self-evolving framework that models selling strategies as testable hypotheses and refines them without parameter updates. A double-loop architecture lets an inner loop adapt to individual merchants through episodic memory while an outer loop evolves a shared strategy playbook through causal attribution and hypothesis-driven evolution. In a live A/B test on the 1688 platform with over 200,000 merchants, SPAR achieves +15.8% conversion lift and +35.8% high-intent merchant growth. Human evaluation by domain experts rates SPAR at 4.62 out of 5 on overall sales quality, the highest among all methods.
Figures & tables
| Metric | Score |
|---|---|
| Outcome agreement | 98.7% |
| Key-behavior recall | 91% |
| Turn-count deviation | 0.9 1.3 |
| Paired comparison accuracy | 87% |
| Expert approval | 89% |
| Objection Handling | Dialogue Quality | Memory | Sales Quality | |||||
| Method | Recognize | Respond | Evidence | Monologue | Repetition | Recall | Score | |
| GPT Eval | Direct Prompting | 0.741 | 0.540 | 0.428 | 0.186 | 0.005 | – | – |
| ExpeL | 0.705 | 0.623 | 0.607 | 0.262 | 0.016 | – | – | |
| Reflexion | 0.759 | 0.638 | 0.672 | 0.221 | 0.017 | 0.053 | – | |
| SPAR w/o Inner Loop | 0.684 | 0.579 | 0.596 | 0.281 | 0.018 | – | – | |
| SPAR w/o Outer Loop | 0.800 | 0.672 | 0.672 | 0.272 | 0.009 | 0.065 | – | |
| Method | Conv. | Sessions | M1 Rate |
|---|---|---|---|
| Direct Prompting | – | – | – |
| ExpeL | +1.1% | +0.3% | +9.2% |
| Reflexion | +9.6% | 4.6% | +27.9% |
| SPAR w/o Inner Loop | +10.1% | 6.3% | +34.5% |
| SPAR w/o Outer Loop | +12.6% | 11.2% | +22.3% |
| SPAR w/o Evolution | +9.6% | 3.0% | +18.7% |
LLM Agent Self-Improvement
Large language model agent self-improvement studies how agents use experience and feedback to improve their behavior, tools, and workflows.
See all 296 papers