Visa (V) Benchmark Finds AI Models Fail Nearly Half of Shopping Tests

What happened

Visa launched an AI shopping benchmark that tested 22 models from 12 labs across 15 challenges. The highest-scoring models passed 82.2% of attempts. Visa said it plans to add scenarios and test checkout, payment and order placement.

Confirmed

Global impact / market context

Testing models on checkout and payments could help Visa assess whether AI shopping tools can complete transactions, not just recommend products. If such tools become more reliable and trusted, they could influence how purchases are made and processed, though the article reports no financial impact.

Analyst inference

The benchmark is an early strategic effort in agentic commerce, meaning AI systems that can act for shoppers. The article gives no revenue or cost figures, so it does not establish a near-term financial effect for Visa; broader use and successful transaction testing remain uncertain.

Analyst inference

What to watch

1. Visa plans to add scenarios and test checkout, payment and order placement; those results would show whether the benchmark extends beyond shopping tasks to completed transactions. Confirmed
2. Whether Visa actually expands the benchmark and publishes results on checkout and payment tests; the article describes these as plans, not completed work. Proposed
3. Evidence that shoppers trust AI to make payments could indicate whether automated purchasing gains adoption, which may affect transaction processing opportunities for Visa. Analyst inference

Evidence

• Original source link unavailable