G2 report · November 2026
Yuma AI is #1 for accuracy and policy compliance on G2
G2 put 10 AI customer support agents through the same 50 tasks, then tested Yuma again on a retail task set. The report has every score and how G2 ran the tests.
PDF, 9 pages. Enter your work email to download.
Trusted by high-volume ecommerce teams
General CX evaluation
Yuma's scores on G2's leaderboard
This is G2's ranked test: 50 tasks, 4 runs, the same setup for every agent. Ten agents are on the leaderboard.
- #1 of 1097%AccuracyAre the agent's claims supported by evidence?
- #1 of 1095%Policy complianceDid the agent follow the written policy?
- #2Overall of 10 AI agents240 of 250 task points, with 48 of 50 tasks passed on average across 4 runs.
- 98%CompletenessDid the required outcome happen in the underlying systems?
How Yuma compares
The top accuracy and compliance scores of every agent G2 tested
G2 also ran two general-purpose AI models with no support product around them. They scored 83% and 76% on accuracy. The gap is the product.
Accuracy
- Yuma97%
- Next-best agent96%
- Average of the others86%
Policy compliance
- Yuma95%
- Next-best agent92%
- Average of the others77%
Averages cover the other 11 agents on G2's leaderboard, including the two model baselines. Leaderboard viewed November 16, 2026.
Retail evaluation
48 of 50 retail tasks passed
Most Yuma customers run ecommerce stores, so G2 also tested Yuma on a retail task set: orders, returns, exchanges and order changes in a simulated online store.
- 48/50Tasks passed
- 240/250Task points
- 98%Relevance, completeness, accuracy and policy compliance, each
Tasks passed by category
- Product inquiry33 of 34
- Refund request19 of 21
- General information13 of 13
- Complaint handling11 of 12
- Billing dispute4 of 4
The retail evaluation ran once and is reported separately from the leaderboard. A task can carry more than one category.
Methodology
How G2 ran the evaluation
- 1
A simulated company
G2 built a support company with policies, customer data and working tools, then set up each agent the way a customer would.
- 2
50 buyer-informed tasks
Tasks came from research with CX leaders and vendors, plus synthetic edge cases. Every task is multi-turn: a simulated customer asks follow-ups and pushes back.
- 3
A strict pass rule
A task passes only if every answer check is true and the required action happened in the system. A failed task earns none of its 5 points.
SourcesYuma's results on G2 ↗G2 CX agent leaderboard ↗Our write-up of the evaluation
Take the full report with you
Every score and G2's methodology in one PDF. Share it with whoever signs off on your AI agent.
PDF, 9 pages
Questions about the G2 evaluation
See it on your tickets
Test Yuma on your own policies
G2 tested Yuma in a simulated store. Yours is the test that counts. We'll set Yuma up on your tickets within a week.
- 97% accuracy on G2
- 140+ ecommerce brands
- SOC 2 Type II compliant
