AI Customer Interactions Need New Evaluation Metrics, EVALS+ Framework Shows

Generative AI now handles customer conversations, product recommendations, and brand emails, but traditional CX metrics such as CSAT, NPS, and AHT fail to capture response quality, tone, safety, and trust. The article by Shruti Tiwari, an AI product manager at Dell Technologies, argues that auto evaluations are necessary to assess AI outputs on clarity, helpfulness, grounding, tone, and safety. Tiwari introduces the EVALS+ Pyramid, a six-layer framework covering metric selection, real-world scenario validation, automation, localization and personalization, governance, and data and drift monitoring. The article notes rising adoption of LLM-as-a-judge setups and open-source tools like RAGAS and Deepchecks. It concludes that continuous AI evaluation is a prerequisite for safe, helpful, and trustworthy customer experience at scale.