Amazon's Weekly Business Review

Amazon’s Weekly Business Review turns operational drift into a weekly decision cycle. Its crucial distinction is between outputs that reveal performance and inputs that teams can actually change.

Amazon’s leaders reportedly inspect 400–500 metrics in 60–90 minutes. Yet the more unsettling lesson comes from a single metric: when teams were pushed to increase product listings, they added products—but sales did not rise, while inventory costs did.

Separate the scoreboard from the controls

The review works by dividing measures into two roles. Output metrics—such as sales—show whether the business is producing the desired result. Input metrics—such as product listings—represent actions teams can influence. A weekly cadence then shortens the gap between action, observed consequence, and correction.

But an input is only a hypothesis about what drives the output. Repeated reviews test that hypothesis against both intended and unintended results. Listings rose exactly as requested; flat sales and higher inventory cost revealed that the chosen input could be moved without creating value.

Where it shows up

More listings, no more sales

Teams responded to the visible input by adding products. The target successfully changed behaviour, but the output exposed the missing causal link: assortment growth alone did not produce additional demand.

The cost metric tells the rest

Inventory cost rose while sales stayed flat. Watching several connected metrics made the side effect visible; judging the initiative only by listings would have made it look successful.

A dashboard can still reward the wrong work

Weekly scrutiny does not defeat Goodhart’s Law by itself. Once an input becomes a target, teams can optimize it literally while the business outcome stagnates. Reviewing hundreds of metrics also creates a selection problem: attention must remain focused on causal relationships, not exhaustive recital.

Put one claimed driver on trial

At your next weekly review, pair one team-controlled input with its intended output and one likely cost or side effect. Ask for all three trends together. If the input moves but the output does not—or the cost worsens—replace the assumed driver instead of demanding more of it.