Share with your CMO
SAP CX president and CPO Balaji Balasubramanian is drawing a hard line on what he calls deployment-first AI: the practice of shipping AI agents into customer-facing channels before the underlying data and systems can actually support what those agents promise. The argument, made to CX Today, is that fragmented customer data and disconnected backend systems don’t just limit AI effectiveness, they get amplified by it. Brands racing to deploy agentic CX experiences without that foundation risk unfulfillable promotions, consent violations, and inconsistent billing at machine speed.
What this means for your business
The organizations most exposed here aren’t the ones that haven’t adopted AI agents yet. They’re the ones that moved fast, shipped something that looked impressive in a demo, and assumed the data quality and system integration problems underneath would sort themselves out later. If your AI agent can hold a conversation but can’t confirm inventory, honor a consent preference, or route a service case correctly, you haven’t automated customer experience. You’ve automated the appearance of it, and the gap between those two things closes badly and publicly.
Balasubramanian’s framing carries a specific tilt worth naming: SAP CX sells deeply integrated enterprise data and commerce infrastructure, so an argument that AI only delivers value when grounded in connected enterprise systems is also an argument for what SAP is already selling. That doesn’t make the underlying claim wrong. The failure mode he describes, where an AI agent recommends an out-of-stock product or offers a discount the fulfillment system can’t honor, is real and well-documented. But the prescription conveniently assumes that integration happens through a unified platform rather than through disciplined data governance across existing systems, which is a choice, not a law of physics.
The governance angle in the piece is the part that will age into a budget conversation faster than most CMOs expect. As AI agents gain the authority to take actions, not just generate responses, the question of what they’re permitted to do, whose data they can touch, and when a human has to sign off stops being an engineering concern and becomes a liability one. CMOs who treat agent governance as the CIO’s problem to solve are one regulatory inquiry or customer complaint away from discovering it was always theirs.
Concept deep-dive: Agentic AI
Agentic AI refers to AI systems that don’t just answer questions but take actions autonomously on a user’s or business’s behalf, booking an appointment, issuing a refund, updating a record. Think of it as the difference between a GPS giving directions and a self-driving car actually steering. The business implication is significant: when AI moves from advising to acting, errors don’t stay in a chat window. They propagate into order systems, billing records, and customer accounts before anyone notices.
Based on reporting from Deployment-First AI Is a Dangerous Bet According to SAP CX, originally published 2026-07-08 07:45:00.
