Agents Are the Demo. The Data Layer Is the Product.

Every enterprise AI pitch deck in 2026 sells the same number. Thirty percent faster fulfilment, thirty percent lower transport cost, thirty percent more accurate delivery: the figure has become shorthand for the moment a supply chain became "agentic." It is also the most misread number in the category.

The thirty percent is not the work. It is the receipt. Behind every reported gain sits a year of unsexy infrastructure: cleaned master data, harmonised supplier schemas, transaction platforms retooled for machine consumption. Lenovo's iChain spans 180 markets, thirty factories, a hundred logistics centres, two thousand suppliers and fourteen million order lines a year. The HBR reconstruction is explicit: the data layer came first, the agents second. The Simor Consulting automotive deployment, Kohler on Databricks, and Belden on its multi-tier supplier graph followed the same sequence.

The vendor pitch inverts the order. Agents up top, data work implied, sometimes deferred. The buyers who accept that sequencing are about to fund the missing year twice: once as consulting, once as the gap between promised and realised gains. The selection story is loud: the firms that publish a thirty-percent result are the firms that already had clean data. The causal story is quieter: the agents monetised the foundation, they did not build it.

Ask any agent vendor what their data layer looked like twelve months before deployment. The answer is the real product.