SDR-Bench Benchmarks LLM Personalization for Sales

A new benchmark, SDR-Bench, has been released to rigorously evaluate the personalization capabilities of large language models (LLMs) within a sales context. Traditional personalization, a two-party problem with independent sender and receiver objectives, is challenging for LLMs which generate messages conditioned on inferred receiver states. Existing benchmarks primarily measure sender-side adaptation, not the crucial third-party action. SDR-Bench, built on a corpus of 6,279 customer success stories across 22 industries, uses a temporally constrained simulation to prevent data leakage. The study found a consistent "personalization plateau" across frontier LLMs and deep-research agents; no model could statistically distinguish between successful and unsuccessful sales outreach in a Fortune 100 tech cohort. A field deployment with professional sales representatives validated the framework, with 48% of model-generated content rated useful, but the core finding highlights that current LLMs lack the ability to truly induce intended actions in a third party, suggesting a gap in genuine two-party personalization.