SmartSalesBench
SmartSalesBench: A Resistance-Aware Multi-Turn Benchmark for Evaluating LLM Sales Agents
A multi-turn conversational benchmark evaluating LLM sales agents against resistant buyer personas across 13,680 sessions involving Samsung smartphone products.
How to run
$ npm i -g lettokenburn-cli$ ltb bench info smartsalesbench
External harness. No Harbor adapter has been generated for this benchmark yet. `ltb bench info` prints the known links.
Runs happen on your machine with your own subscription; only the score summary is pushed. CLI docs
Scores
4 sourced entries| Harness | Source | Confidence | Link | |||
|---|---|---|---|---|---|---|
| Phi3.5-mini | 72.6 | N/A | Independent | medium | Oct 9, 2026 | arxiv.org |
| Gemma2-2B | 43.7 | N/A | Independent | medium | Oct 9, 2026 | arxiv.org |
| Qwen2.5-3B-Instruct | 37.1 | N/A | Independent | medium | Oct 9, 2026 | arxiv.org |
| Llama-3.1-8B-Instruct | 24.2 | N/A | Independent | medium | Oct 9, 2026 | arxiv.org |
Every number links to where it was published. Low-confidence rows are muted and need per-cell verification.
Reproduced by the community
All runsAbout
SmartSalesBench features 5 agent profiles, 12 buyer personas, and 15 products in up to 20-turn dialogues. It uses RAG-grounded generation and includes four controlled experiments totaling 13,680 sessions.
Evaluation employs nine metrics across three tiers, scored via automated LLM-as-judge and human assessment. Baseline models achieved a 44.4% average conversion rate, with Phi3.5-mini leading at 70.7%.