Warehouse automation is quietly becoming an AI capex line

Vision-driven picking and autonomous material handling have moved past pilot in several large distribution operations. The technology is unremarkable at this point. The accounting is where finance functions are getting caught out.

A conventional automation asset — conveyor, racking, steel — carries a long useful life and a predictable depreciation profile. A system whose capability is defined by control software on a rapid release cycle does not. Treating the whole installation as one asset with a fifteen-year life overstates its remaining value the moment the software generation changes. Several operators are now splitting the asset, with a short-life software and controls component sitting alongside the long-life physical infrastructure.

Lock-in is the second issue and it is not where people look for it. The steel is commodity. The control software, the trained models and the accumulated operational data are not, and a vendor who holds them holds the operation. Data portability clauses are cheap to negotiate at purchase and effectively impossible afterwards.

On the business case, the common error is to model payback on labour cost alone. In the operations we have reviewed, the larger benefit is reduced throughput variance — the ability to hold service levels through peak without contingent labour. That is harder to quantify and it is usually worth more. A payback model that omits it will understate the case and, ironically, get the investment rejected for the wrong reason.