Visa: AI shopping agents bought the wrong product or failed to buy at all in 42.8% of attempts
In a US study Visa ran with 22 models from 12 providers, results varied widely from model to model, and the same model performed inconsistently across repeated attempts.
What's happening
According to Mobile Time, AI agents in Visa's US study completed the purchase in 57.2% of shopping attempts. In the remaining 42.8%, the purchase was left unfinished or the wrong product was bought. The best models succeeded in 82.2% of attempts; the worst, in only 28.9%.
Whose desk it lands on
The head of digital channels. If an agent shopping on a customer's behalf does not get the same job right every time, how much authority do you give it?
Our view
The CX Brief's interpretation, not a claim made by the source.
Getting a purchase right once is not enough; an agent has to deliver the same result every time. The same model won approval in all three of its attempts in only 34.2% of cases. For an agent placing orders on a customer's behalf, that consistency matters as much as the average success rate.
