What it is

Price sensitivity is the pricing engine's tier for one variant: how much its sales moved, on average, when its list price moved. The tier is one of High, Moderate, Low, Inconsistent or Still learning, and a variant the engine holds no tier for reads Not scored. Beside the tier the engine prints a Confidence and a Basis. The tier and the basis come from two different reads, so they can disagree: the tier is measured from price moves, the basis from the prices customers actually paid. It is part of the pricing engine.

How the tier is measured

The tier is computed inside every sync, after the step that records each variant's price changes. It reads the last two years of list-price moves. A move counts when the price has drifted at least 3% from where it last settled, then held for at least 14 days, and landed at least 14 days after the previous move that counted.

For each move it compares average daily sales in the 14 days before with the 14 days after. A move is thrown out when either window had more than two days of unusual spikes, a window where 30% or more of the units sold at least 10% under their own original price (a promotion: a code, an automatic discount or a sitewide sale), more than a fifth of its days out of stock or not selling, or data on fewer than four days in five. A list-price move is not itself read as a promotion. The effects are shrunk toward zero, so a few moves say less than many.

• Fewer than three clean moves: Still learning, with no confidence.
• Six or more moves whose average size of effect is at least 5% but agree on direction less than 60% of the time: Inconsistent.
• Otherwise the tier reads the price response scaled by the size of the move: each move's units change divided by its price change, so a move is judged by how much it moved sales for its size, not by the raw percentage. Below 0.5 is Low, from 0.5 up to 1.5 is Moderate, 1.5 or more is High. A cut followed by fewer units is not sensitivity and never lifts the tier.

Confidence is Low below five moves and Medium below eight, and it drops further when the moves agree less than 70% (Low) or 90% (Medium) of the time. Inconsistent reads Medium below ten moves and High from ten. A tier whose moves have aged out of the two years is removed on the next run. A failed pass never fails the sync.

How the basis is decided

The basis is measured on each view of the variant, from the price actually paid per selling week, weighted by quantity, in one currency. It asks whether there is enough price variation, for this variant's own swing in weekly units, to measure a price response to within a set precision.

• Measured on this item: the variant's own weeks carry it.
• Borrowed from its cohort: the variant is too thin, and every variant of the same product pooled together carries it instead, so the figure describes the product.
• Not enough price history: neither can.

An item that sold in fewer than two weeks, sold at one price throughout, or sold the same quantity every week cannot be measured, and the section says which in a sentence of its own, for example:

Sold in 1 week. A price response needs at least 2 selling weeks before there is anything to compare.

The words in the Price section

On a variant's Price section the rows read Tier, Confidence, Basis, Price changes in 12 months and Sales through. Confidence reads High confidence, Medium confidence, Low confidence, or Not stated. Where the basis is not measured, a Category reference row gives a published starting point, worded like 2.0% fewer units for 1% on the price, with a hover saying whether it is published for this category, extrapolated from the nearest category, or a general default, not measured for this category. Apparel has no published figure and takes the default.

Where the tier is measured, the section's one sentence is the engine's own summary, for example:

Only price decreases have been tested. The effect of a price increase is unknown.

Where the tier is missing or borrowed, the evidence sentence takes its place. If the read fails the section says The price read did not answer. with Try again.

Where you see it, and what you can do

Every plan sees the Price section and the Price sensitivity column and filter on Sales performance, where a line with no tier reads Not scored. Nothing is pressed to measure a tier: the price has to move, and the next sync measures it. Read the Basis before acting on a tier, and use Ask Tightly beside the section's title to ask what to do about it.

Glossary line

Price sensitivity: the pricing engine's tier for how far a variant's sales moved when its price moved.