SalesBench: Testing Adaptive Social Intelligence in Extended Goal-Oriented Interactions
· 4 min read

How do agents perform in high-pressure sales scenarios?
We answer this by letting AI agents navigate complex sales conversations. The agents need to handle objections, read buyer signals, and close deals to maximize revenue.
Can AI actually sell me this pen?
AI Agents have become popular with developers because they generate and revise code quickly. This has led to greater engineering efficiency and enabled smaller & smarter teams to move faster than incumbents. But engineering is typically only half of a business. You still have to sell.
Introducing SalesBench, benchmarking AI Agents in adaptive social intelligence in extended goal-oriented interactions.
How do agents navigate complex human dynamics over extended conversations? We answer this by having agents conduct cold-call life insurance sales to diverse simulated buyers. The agents must build rapport, handle objections, and adapt their strategy over long conversations to close deals.
We've evaluated leading models to see how persuasive they can be.
Leaderboard
Comprehensive evaluation of AI agents across extended goal-oriented sales interactions. Ranked by overall performance across all difficulty levels and metrics

The Eval
SalesBench is a simulated environment that tests how well AI models can navigate complex social interactions in a goal-oriented context: conducting life insurance sales calls. The AI agent must build trust, identify needs, handle objections, and adapt their approach based on buyer personality. We break this into individually manageable tasks that, over extended conversations, reveal an AI's ability to maintain coherent social intelligence and manage context over extended periods.
We simulate 1 day for the Agents to try and close as many life insurance sales as possible. We have 100 custom buyer personas generated by combining a randomized assortment of circumstances.
How it works:
A Sales-Operator agent coordinates the sales process.

We've added memory tools to assist with context quality. Previous methods of context maintenance relied on the system deciding what to retain through summarization. We've implemented a sliding context window with memory tools, similar to Vending Bench.
The agent keeps the most recent 30,000 tokens in context before inference and decides what to save to memory using tools. Memories are saved as embeddings and memory reads are done with cosine similarity.
The call_lead tool initiates a subagent that handles the sales call with the prospective buyer agent. We decided to delegate this to a subagent task to make context more efficient. The subagent returns a call summary. The Sales-Operator can save it to memory or use it to update the CRM.
We simulate time by assigning set minute times for each tool call (with the exception of Sales Calls, which are calculated by multiplying the number of agent response loops by 30 seconds), and giving the Agent the ability to wait until the next working day or sleep until a specified time.
Conclusion
SalesBench reveals both the promise and limitations of AI in sales. While Claude’s 86% close rate shows AI can build relationships and close deals, every model’s failure with difficult customers highlights the gap between current AI and human social intelligence.
The differences between Claude’s patient relationship-building and O3’s aggressive tactics suggest that AI models develop distinct “sales personalities” that affect their success. As businesses increasingly explore AI for customer interactions, understanding these differences becomes crucial.
Can AI actually sell? Yes—but only to customers who want to buy. The art of converting skeptics, handling complex objections, and building long-term relationships remains uniquely human, at least for now.
Reference: By Kyle Jeong, Sameel Arif, & Hamza Mostafa
Heavily Inspired by Vending Bench https://andonlabs.com/evals/vending-bench
-> Kyle