Proof

The task is done, the box ticked, the secondary placement is in place, the checkout is staffed, replenishment runs differently than before. After that, no one follows up. Revenue, service and productivity move in the weeks that follow, only no one knows why. Cortex records every action with timestamp, store, zone and item in the same grid as the metrics, so that the question of whether it worked can have an answer at all.

The obvious calculation, before versus after, does not hold, because season, weather, holidays and the promotion in the next aisle move the figure in the same period. A falling conversion rate can simply mean more footfall, because conversion decreases as footfall increases, as has been measured in retail. Without context, it is read as a deterioration, and the wrong action follows.

Cross-check

Cortex forms the comparison group from the locations that did not have the intervention, stores with similar footfall, similar assortment and a similar trend in the previous weeks. What is measured is the difference between the two trends, not the change alone. What weather and season move, they move in both groups, and what remains is the action.

How clear the difference can be is shown by a field experiment in 214 stores of a German perfumery chain. Eight items moved from their shelf to a display at the checkout and sold 197 percent more per week, all other items in the same stores 0.54 percent more in the same period, and per item the effect ranged between 80 and 478 percent. An action that is rolled out everywhere at the same time no longer has this cross-check, and Cortex says so before the rollout, not after.

Difference

The same action works differently at each location. At a retailer with over 500 stores, an additional dollar of staff costs brought between four and 28 dollars of revenue depending on the store, seven times as much at one location as at another. The comparison across the network is therefore not a ranking, but the question of where an action yields more.

This turns the network comparison into action. Actions are rolled out where the impact is measured and left where it fails to materialize, and the deviation between stores is justified instead of judged. Reallocating the existing staff budget to where it brings more revenue alone lets this retailer expect an increase of 2.6 percent at no additional cost.

Learning

Every measured action feeds into the next decision as experience. Cortex tracks per type of action the success rate, the time to insight and the uplift it delivered on average, and next time suggests the action that worked in a comparable situation.

Also from what went wrong. Complaints and recurring errors are held against the action that was supposed to fix them, and the repeat rate shows whether the error has been eliminated or merely shifted. The circle only closes when the action is not just implemented but proven, and the sum of the proofs yields an operation’s knowledge of what works for it.