G2 tested 10 AI agents on the same 50 support tasks, and Yuma led both accuracy and policy compliance at 97% and 95% while ranking #2 overall.
Yuma AI, the AI customer service platform for ecommerce brands, today announced that it ranks #1 for accuracy and #1 for policy compliance in G2’s AI CX Agent Evaluations. G2 tested 10 AI customer service agents on the same 50 support tasks, and Yuma also ranked #2 overall.
An AI agent that answers wrong costs you more than one that doesn’t answer at all. Accuracy and policy compliance are the two scores we’d pick, and an independent test put us first on both.”
— Guillaume Luccisano, CEO & Founder
Yuma scored 97% on accuracy, which G2 defines as whether an agent’s claims are supported by evidence. It scored 95% on policy compliance, whether the agent followed the written support policy. The next-best scores among the 10 ranked agents were 96% and 92%. On task success, Yuma earned 240 of 250 points, passing 48 of 50 tasks on average across 4 runs, 3 points behind the top-ranked agent.
Read More: SalesTechStar Interview with Sam Branson, Head of Partnerships at Backstory (formerly People.ai)
Most AI support tools are judged on demos and vendor claims. G2’s evaluation takes a different approach. It built a simulated support company with realistic customers, order and account data, a written support policy and 38 working tools for refunds, subscription changes and account updates. Each product was set up the way a customer would set it up, then given the same 50 tasks drawn from buyer research, design partners and edge cases.
Every task produced a full transcript, a log of every action the agent took and the final state of the systems. LLM judges scored accuracy and policy compliance. Deterministic checks confirmed the required outcome actually happened. Vendors could flag setup mistakes before results went live, but couldn’t change their scores.
Read More: Reimagining subscription revenue with agentic AI
“Support leads don’t lose sleep over the tickets AI can’t answer. They lose sleep over the ones it answers wrong: the refund outside the return window, the delivery date nobody promised,” said Guillaume Luccisano, CEO and founder of Yuma. “Yuma doesn’t improvise where your Handbook is silent. It hands the ticket to a person on your team. G2’s results show what that looks like when someone else does the grading.”
Key results for Yuma in G2’s AI CX Agent Evaluations
Accuracy. 97%, ranked #1 of 10. G2 measures whether the agent’s claims are supported by evidence.
Policy compliance. 95%, ranked #1 of 10. G2 measures whether the agent followed the written support policy.
Task success. 240 of 250 points, ranked #2 of 10. Yuma passed 48 of 50 tasks on average across 4 runs.
Against raw models. G2 also ran two general-purpose AI models as baselines with no support product around them. They scored 83% and 76% on accuracy.
Yuma’s guardrails are built for exactly this. The Handbook sets how Yuma sounds, what to do in each situation and the exact steps to follow. Hard Limits put strict caps on refund amounts, coupon values and loyalty points the AI can never exceed. Gradual Rollout puts a new playbook live on 10% of matching conversations, with the rest handled as drafts.
