Visa 测试 22 个 AI 模型购物，42.8% 运行失败
Visa Finds AI Models Fail 43% of Shopping Test Runs
Visa 对来自 12 家实验室的 22 个 AI 模型进行真实购物场景测试，57.2% 的运行通过、42.8% 失败，表现最好的模型通过率 82.2%，最差的仅 28.9%。该测试属于 Visa Agentic Commerce Benchmark，首轮任务包括购买电视、笔记本电脑和个人护理电器。Visa 还发现，54% 的失败运行最终仍完成了购买，但未满足或未完全识别消费者需求。
By PYMNTS | November 5, 2026
|
Visa found that when 22 artificial intelligence models from 12 labs were put through realistic shopping scenarios, 57.2% of the test runs passed and 42.8% failed, according to a Thursday (Nov. 5) press release.
Among the best-performing AI models, 82.2% of test runs passed, while among the worst-performing models, 28.9% of test runs passed, the release said.
These are some of the findings of Visa’s new industry benchmark for AI shopping, Visa Agentic Commerce Benchmark, which is designed to create a roadmap for industry players to deliver better agentic shopping experiences. The test can also be used by consumers, the release said.
Visa Agentic Commerce Benchmark tests AI models’ ability to identify the needs of a consumer based on a conversation, uncover enough information to justify its decision to buy, and opt not to buy if nothing meets the customer’s requirements, per the release.
For its first round of tests, Visa put the AI models through shopping challenges that included buying a TV, a laptop or a personal care appliance, according to the release.
Visa found that the five models with the highest average cost per run passed 67.1% of test runs, while the five least expensive models passed 50.2% of runs, the release said.
The testing also found that 54% of failed runs still ended in the AI making a purchase, even though that purchase did not meet the customer’s identified requirements, or the AI did not uncover all the customer’s requirements, per the release.
“AI agents have the potential to power the next era of commerce, but trust is key to making it work,” Rubail Birwadker, global head of growth at Visa, said in the release. “Visa Agentic Commerce Benchmark is yet another way Visa is working to boost standards across the industry.”
The PYMNTS Intelligence report “Will the 2026 Shopping Season Go Agentic?” found that nearly 132 million U.S. adults have bought a product with AI’s help. However, while AI has won a place in product discovery, consumers still control the purchase.
For all PYMNTS AI coverage, subscribe to the daily AI newsletter.
来源：PYMNTS · pymnts.com