Price Elasticity Testing for E-Commerce: What Customers Will Actually Pay

"Raise prices 10%" confidently still loses money. The reason isn't the move — it's the assumption. Pricing decisions made on instinct, on what the last buyer said, or on a single bad week last quarter produce either too small a raise (and no margin recovery) or too large a raise (and volume walks out the door). Price elasticity testing replaces the instinct with a curve: here is what happened at $X, here is what happened at $Y, here is the price the customer will actually pay without abandoning the basket.

Why "raise prices 10%" still loses money

A blanket price move on a full catalog assumes two things, and both of them are usually wrong:

• That elasticity is uniform across the catalog. It isn't. Headline SKUs with strong brand affinity sit near vertical demand curves — they absorb price without volume loss. Long-tail, commoditized SKUs sit on elastic curves — they hemorrhage volume at the first move.
• That a single past move was representative. It usually wasn't. The seasonal promotion, the MAP reset, the vendor's co-op campaign — every prior move confounded price with something else. Reading a single data point as the elasticity is how teams anchor on the wrong number.

The fix is to isolate a single variable and read the curve it produces. That is the entire job of price elasticity testing, and it is the missing layer between monitoring competitor prices and shipping a recommendation.

The elasticity curve, in one sentence

For a given SKU (or a tightly-defined segment), the curve maps every candidate price to the volume it produces over a comparable window. Plot it with price on the x-axis and resulting units on the y-axis; the shape tells you almost everything you need to know about that SKU's pricing posture.

Three regions show up on a healthy curve, and each one drives a different decision:

The knee

The first price point where a small move produces a disproportionate volume response. To the left of the knee the curve is uneventful — small price moves, small volume moves — and to the right of the knee the curve bends sharply. The healthy question to ask is "where on the curve is our current price?", and the answer decides whether the SKU is being underpriced (left of the knee) or squeezed (right of it).

The dead zone

A region where price moves produce almost no volume response at all. A SKU in its dead zone is under-earning — it could move 8% upward without losing meaningful volume, and the margin is being left on the table. Dead zones are where pricing teams recover the most revenue fastest.

The cliff

The price point past which demand collapses. Every category has a cliff; the question is where. A safe pricing posture sits left of the cliff by enough margin to absorb a competitor undercut without crossing it. That safety gap is not a multiple of cents — it is a configured percentage of the cliff price, and it is what makes the competitive pricing band actually defensive.

Designing a clean A/B (or holdout) test for an e-commerce SKU

A meaningful elasticity curve requires a clean experiment. The five rules below are the ones we ship in production, and violating any one of them invalidates the resulting curve:

1. One variable, one move. Isolate price. Promo, free shipping, bundle composition, and merchandising rank all confound the result; hold them constant for the duration of the test. If a competitor undercut lands mid-test, restart the test window.
2. One SKU or one tightly-defined segment per test. Don't bundle three SKUs into a single "category elasticity" test — the curve you get back is an average of three different curves and it tells you almost nothing actionable.
3. Comparable windows. Compare the test window against a control window of identical length, season, day-of-week mix, and competitor state. Two weeks of "Black Friday first half" against two weeks of "Black Friday second half" is a defensible comparison; "the week before the sale" is not.
4. A configured margin floor on every test SKU. The test must not be able to generate a result that breaches the floor. A test that returns a $19.99 optimal price for a SKU with a $22 margin floor is a misconfigured test, not a useful data point.
5. A pre-registered stopping rule. Decide the test duration, the numbers of impressions per arm, and the success criterion before the test starts. Walking the test, extending it because the early signal is interesting, or declaring the winner because it "looks right" all invalidate the curve.

An elasticity test that isn't pre-registered is a survey, not an experiment. The difference matters: a survey tells you what the team hoped would happen, an experiment tells you what the customer did.

Reading the curve — knee points, dead zones, and the floor

Once the test window closes, the curve is the artifact. Three readings turn it into a recommendation:

1. Find the floor. The lowest price on the curve at which the SKU still produces meaningful volume. That's the margin floor for the SKU — distinct from a cost-plus number, distinct from a competitor-anchored number. It is the price at which demand collapses.

2. Find the dead-zone ceiling. The highest price on the curve at which the SKU still produces comparable volume to the floor. That's the upper anchor for the competitive pricing band — distinct from the highest the merchant would ever charge, which is rarely the right ceiling.

3. Mark the knee. The bend in the curve. The knee is the price where further moves stop being free. Pricing posture is the choice between operating left of the knee (defensive, conservative) or right of the knee (aggressive, recoverable).

From the curve to a margin-safe recommendation

A test answers one question — what would happen at price X? — and the answer gets used four ways in the layering above it:

FAQ-style add-on

• How often should the curve be re-tested? Quarterly for stable categories, monthly for elastic ones (apparel, beauty, consumer electronics with frequent SKU churn), and any time a competitor's posture on the SKU shifts sharply.
• What if the curve is too flat to read? Then the SKU's elasticity is too low to be answered with one test — pool three to five consecutive tests against the same SKU before drawing a posture conclusion, or accept a wider confidence band on the floor.
• What if the curve is too steep? Then the SKU is hyper-elastic — every move is loud. Operate left of the knee, widen the safety gap to the cliff, and treat any competitor undercut as a category-wide signal, not a per-SKU one.
• Does the curve replace competitive monitoring? No — the two layers compose. The curve gives the floor and the dead-zone ceiling for the SKU; competitive monitoring tells you where the market sits between them. Without one or the other the recommendation engine is working with half a picture.

Closing the loop with the rest of the cluster

Elasticity testing is the measurement layer that sits underneath the rest of the pricing cluster. Without it, "raise prices 10%" confidently still loses money. With it, every downstream recommendation has a curve behind it.

For the monitoring half that feeds the in-band decisions, see our competitor monitoring guide. For the loop that turns the curve into a daily recommendation, see dynamic pricing strategy. For a worked example of the floor + ceiling output, try the pricing report tool, or compare the tier with daily recompute on the pricing page.