No Jitter Brief:
NiCE launched Cognigy Simulator, which allows organizations to rehearse, evaluate and harden AI agents before letting them loose to interact with real-world customers, the company announced Tuesday. Contact center vendor NiCE purchased Cognigy, a conversational AI vendor, last July.
Simulator allows a business to create thousands of synthetic conversations based on customer demographics, languages and intents, such as why they reached out for service. The tool will then test how the AI agents perform in realistic, adversarial and edge-case interactions.
AI agents are scored against quantitative metrics such as task completion, guardrail adherence, integration reliability and experience quality. Simulator also allows users to pinpoint where AI agents failed so they can refine prompts, flows or policies. Organizations can evaluate prompt strategies, guardrails, fulfillment logic and the underlying large language model by using A/B and variant testing.
No Jitter Insight:
Agentic AI systems, based on LLMs, can potentially generate misleading outputs, which can pose problems for customer interactions. In customer service, mistakes made by an AI agent – or human agent – can damage a brand’s reputation, expose customer or company data, or create legal or other ethical risks.
Ensuring that an agentic AI agent always interacts with a customer the same way every time is a challenge for organizations. The Cognigy Simulator represents a way to mitigate those risks through exhaustive testing.
“Simulator provides data-informed testing and reporting to help organizations understand AI Agent performance and compliance alignment, so organizations can make deployment decisions with confidence,” said Philipp Heltewig, general manager, NiCE Cognigy and Chief AI Officer, in the release.
The news is an example of how organizations can use synthetic data based on real customers to improve AI agents. It also shows traction for the use digital twins as a way for organizations to understand their customers on a deeper level.