Visa Finds AI Models Fail 43% of Shopping Test Runs

Recent testing by Visa reveals that AI shopping assistants are not yet ready for prime time. In a study of 22 different AI models, nearly 43% of shopping tasks ended in failure. While the best models succeeded 82% of the time, the worst performers failed more than 70% of their attempts. For retailers and financing providers, this means that fully automating the customer journey through AI still carries significant risk. The failures often occurred during complex tasks like comparing financing terms, applying discounts, or navigating checkout flows. For your business, this highlights a 'trust gap.' If you are looking to implement AI to help customers select financing plans or manage their accounts, you cannot leave the system unattended. The study shows that these models struggle with the nuance required for high-stakes financial decisions. Customers who receive incorrect information about interest rates or payment schedules due to an AI glitch will likely blame the merchant, not the software provider. Use AI as a tool to assist your sales team, but keep a human in the loop for final transactions and financing approvals. The technology is evolving, but it is currently too inconsistent to manage the entire conversion funnel alone.

Source: PYMNTS

← Return to the library· Submit a correction