Post by Abir Mukherjee

Business Analysis automation: A POC proves the model can answer. Production demands that we can trust the answer. Analytics accuracy is a context and verification problem, not a code generation problem. (source in comments) The distinction matters more than it sounds. A coding agent works in an open-ended space where many solutions are valid, and tests act as guardrails. Analytics is the opposite. There is usually one correct answer, drawn from one correct source, with no compiler to tell you when the number is wrong. So the model confidently returns a figure. It looks precise. And no one in the room can say whether it queried the right table, applied the right filter, or used the company's actual definition of "delinquency rate" or "attrition." That is the real work of scaling GenAI in analytics. Not cleverer prompts. Governed data foundations. Owned definitions. Curated context. Continuous verification. So the honest question for every team running these pilots: what are we actually focusing on, and is it the thing that will let us scale?