The Maze: Most AI business cases die from loneliness. A chatbot improves service. A pricing model improves a trading decision. A demand model improves a forecast. Each can work. None automatically changes the economics of an e-commerce business. McKinsey’s four ranges make the better point: agentic AI creates disproportionate value when growth, productivity, value-chain efficiency, and profitability reinforce one another. The flywheel is the product. The individual tool is just a spoke.
The biggest stated range sits in productivity, not customer theatre. McKinsey places a 30–50% productivity improvement in core commercial and support activities at the top end of the framework. That is a useful corrective to the industry’s obsession with conversational storefronts. The short-term economics may sit in customer care, content operations, account management, merchandising, and back-office workflows. A retailer can enjoy a good demo without changing the cost of work. It creates a durable advantage when the saved capacity turns into faster, better commercial decisions.
Growth requires relevance that can be operationalized. The framework assigns a 10–15% increase to hyperpersonalization and account management. That is not permission to spray more recommendations into an already crowded interface. It requires connected signals: what a customer needs, what is available, what can be delivered, and what the retailer can profitably offer. Personalization detached from inventory is a beautiful way to recommend disappointment. The commercial system has to know the difference between a relevant offer and a fulfilment problem with good creative.
Value-chain efficiency is where the loop becomes physical. McKinsey puts a 10–20% improvement range around more streamlined operations, exact demand–supply matching, and fewer shortages. This lever matters because it links the promise made at discovery to the order actually delivered. Better demand signals should change replenishment, allocation, and routing. If they do not, the company has added intelligence upstream while leaving the cost and failure modes downstream untouched. That is not a flywheel. It is a faster way to generate exceptions.
Margin is the outcome, not a fourth dashboard. The framework assigns a 3–5% improvement in operating margin to automation of key processes. It should not be added to the other ranges. They use different denominators and describe different mechanisms. The economic logic is sequential: productivity creates capacity, capacity supports better customer and trading actions, those actions improve demand quality, and better demand quality reduces waste. Margin appears when the loop stays connected long enough to alter daily operating choices.
Why it matters: The sensible AI roadmap is not a list of pilots sorted by novelty. It is a map of dependencies. Start where data can travel between customer interaction, merchandising, pricing, inventory, fulfilment, and service. Then measure whether one improvement makes the next decision better. The winners will not be the companies with the most AI labels. They will be the ones whose commercial system gets less wasteful and more useful every time it runs.
