Can generative AI give us prescriptive analytics?

When I was a partner at one of the Big Four accounting firms, clients asked me to provide recommendations and guidance on what a fraud examiner should do based on what the analytics were telling them about their data. For example, if the analytics showed high risks for bribery and corruption in vendor payments, the software could recommend key steps, relevant company policies or guidance, sometimes even before making a payment in question. Theoretically, this approach could work, but the combinations were just too vast to anticipate every potential outcome. We needed more data. We got close with the “digital twin” concept with GE back in 2018, but we still couldn’t acquire enough data to accurately prescribe each outcome. (See “‘Profit & Loss-of-One’: Preventing fraud, enhancing compliance using digital twins,” by EY Fraud Investigation & Dispute Services and GE executives; Ed. Vincent M. Walden, CFE, CPA, Fraud Magazine, January/February 2018.) What we were reaching for was prescriptive analytics. And unfortunately, it remained at the time a conceptual — rather than realistic — goal for compliance, fraud prevention and detection.