The claim
A forecast has no business value until someone uses it to change inventory, labor, pricing, capacity, or another decision. That makes the complete system—not the model—the meaningful unit of engineering.
What the model cannot solve
A model cannot repair an incorrect product hierarchy, stale inventory, missing promotion context, or a location calendar that disagrees with the operational system. Better accuracy on a benchmark does not remove those dependencies.
A better design question
Instead of asking only “How accurate is the forecast?” ask: “What evidence produced it, which decision consumes it, what conditions make it unsafe, and how will a user understand a change?”
The question to carry forward
When a prediction changes, can the system explain whether the cause was demand, data, business policy, or platform behavior?