RESPONSIBLE AI AT CRESTA

Cresta meticulously balances fairness at the model level by leveraging diverse datasets, customizing models with proprietary customer data, and mitigating biases through robust monitoring. At the application level, Cresta fosters inclusivity by reinforcing context-specific best practices and minimizing human bias in performance management, empowering agents with diverse skills and backgrounds to be successful.

Cresta upholds the highest standards of privacy and ethics by focusing on enhancing agent performance and customer experience without compromising personal data. We do not use any protected explicit signals, such as metadata related to medical or financial attributes, in the process of training our models. We use industry-leading personally identifiable information (PII) redaction algorithms, as well as Payment Card Industry Data Security Standard (PCI-DSS) and ISO27701 compliance.

Transparency is at the core of Cresta’s responsible AI approach. We invest heavily in R&D designed to make it easier for humans to interpret and understand the outputs of our AI models, including cutting-edge techniques like Chain-of-Thought Reasoning (CoT) and Model-based Critique. We’ve developed a variety of internal tools designed to observe and benchmark our models, and will continue devoting resources and roadmap to exposing more of these directly to customers in ways that can add value and increase trust.

Our commitment to quality optimization and risk mitigation is evident throughout our development lifecycle. Our AI models are fine-tuned for context and use cases. We do extensive post-processing, leveraging techniques such as filter models, re-ranking, and self-critique to double-check the outputs of our LLMs. By continuously monitoring usage trends, implementing feedback loops, and employing risk mitigation strategies, Cresta ensures high-quality, ethical AI that evolves alongside our customers’ needs.