David Abitbol, AI lead at Revolut, told OpenAI that AI lets the bank hold support and fraud-review staffing flat as customer numbers climb. The work covers shopping fraud and support tickets, a high-volume remote problem rather than a local device flaw.
Abitbol framed the issue as humans versus demand, per OpenAI. Revolut added 1 million customers every 17 days for the last year, he said, a pace that once forced one-to-one growth in support and financial-crime review. Now, he said, a flat team plus AI answers the tickets and questions.
He called Codex the first app harness to make building accessible to non-technical staff.
That claim matters because Abitbol tied it to a company-wide standard for AI work, including development checks, business metrics and A/B testing, per OpenAI. Teams using Codex daily cut time from task start to feature merge by 30 to 60%, he said, and spent less time hunting across knowledge bases. The freed time now goes to design discussions with engineers instead of implementation, he said.
Outside that account, TokenPost reported that Revolut used OpenAI models and Codex to handle more suspicious-transaction investigations without a proportional increase in investigative staff. SME Cyber Insights separately reported Revolut-themed phishing texts that copied visual cues and sought identity material through a fake liveness check. One account stresses faster internal response, the other stresses remote impersonation of the bank itself.
Abitbol gave coding time but no fraud-loss number, no precision rate and no human-review threshold. How much shopping fraud does Revolut now close with AI alone, and at what false-positive cost?




