Session Overview

Three AI Agents Shipped to Production that Move Revenue.

Many companies lose money on AI by treating it like an expensive parlor trick, isolating it to chatbots that never impact the P&L. In this pragmatic session, Rwitapa Mitra, Director of BI at FabHotels, challenges the static dashboard paradigm. She demonstrates how directly plugging AI into business bottlenecks can drive real revenue, proving that the win goes to those who treat AI as an automation pipeline rather than a conversational gimmick.

Through three real-world case studies—a Call-QA Scorer, a Pricing Agent, and an NL-to-SQL Bot—Rwitapa shares the triumphs and the critical failures of deploying AI in a live business environment. Attendees will learn that the model itself is only the "easy 80%"; the real job, and the true mark of product taste, lies in building the guardrails that prevent an LLM from becoming a liability.

Key Takeaways & Concepts

Prompt as Policy: How a Call-QA Scorer achieved 100% audit coverage and a +3% conversion bump, and why treating prompts as strict legal documents with examples is crucial to prevent models from literally misinterpreting rules.
The Freshness Gate: Insights from a Pricing Agent making 250+ updates daily, highlighting the critical need for "real, recent, sane, or stop" data gates to prevent "scraper drift" from quietly feeding bad data into production.
Clarify & Read-Only: Lessons from an NL-to-SQL Slack bot that turned every employee into an analyst, demonstrating why AI with database access must default to read-only and require human editor approval to prevent hallucinated data destruction.