Put an economic gate before the anchoring experiment
A bounded test sequence
- Confirm that the candidate price covers the selected cost boundary and user-chosen contribution target. (not complete)
- Use truthful candidate and control presentations for the same audience, offer and observation period. (not complete)
- Measure contribution as well as conversion so a higher order count cannot hide weaker economics. (not complete)
- Name the sample, attribution limits and stop condition before reading the result. (not complete)
Start with the margin boundary, control and stop condition
Prerequisites, sequence and stop points
- Confirm the same offer, segment, price, cost basis and indirect-tax presentation. (not complete)
- Define exactly one truthful anchor treatment and one control. (not complete)
- Choose contribution or gross profit as a primary commercial measure and conversion as supporting evidence. (not complete)
- Record refund, cancellation and trust-complaint guardrails before the test. (not complete)
- Randomise allocation when feasible; otherwise label a bounded sequential comparison observational. (not complete)
- Retain, revise, stop or gather evidence without turning a small sample into a causal claim. (not complete)
Compare the control and treatment without claiming causation
gross contribution before other period costs = (price − unit cost) × orders
- price
- Candidate selling price on the declared tax basis (currency per order) — user-entered scenario
- unit cost
- Matched cost boundary (currency per order) — business record or labelled assumption
- orders
- Observed orders in the aligned audience and window (orders) — test observation
The formula measures the labelled scenario; it does not prove that the anchor caused any difference.
| Field | Control | Anchor treatment |
|---|---|---|
| Offer and price | Same offer at 100 CU | Same offer at 100 CU with truthful comparison anchor |
| Cost basis | 60 CU per order | 60 CU per order |
| Audience and window | 100 comparable visitors | 100 comparable visitors |
| Orders | 8 | 9 |
| Gross contribution | 320 CU | 360 CU |
| Refund, cancellation and trust guardrails | Record the same refund, cancellation and trust-complaint definitions and observation window | Stop or investigate if any guardrail worsens; no uplift claim overrides the guardrail |
| Evidence quality | Control observation | Not causal unless valid randomised analysis supports it |
Interpret a labelled test scenario
Labelled scenario
The treatment records one extra order, but the sample is not asserted to be adequate and the difference is not automatically caused by the anchor.
| Case or record | Inputs and arithmetic | Interpretation |
|---|---|---|
| Observed base case | (100 − 60) × 8 = 320 CU; (100 − 60) × 9 = 360 CU; conversion moves from 8% to 9% | That is +1 percentage point, or a 12.5% relative percentage change from 8%; neither expression proves adequacy, causation or uplift |
| Downside guardrail | Extra refunds, service cost or a lower-quality audience remain unresolved | Result is inconclusive; stop or gather better evidence |
Retain, revise, stop or gather more evidence
Use the calculation owner for the next step
- Target Margin & Pricing Planner
Set a price for a target gross margin
- Price–Volume Trade-off Planner
Find the sales volume needed to preserve contribution after changing price
Questions and limitations
- Does an anchor increase margin?
- Not necessarily. The candidate price must first pass the margin boundary, and the test must measure contribution and guardrails.
- Is nine orders versus eight enough proof?
- No universal sample claim is made. Treat the example as arithmetic and evidence-governance, not causal proof.
Sources and scope
- Judgment under uncertainty: heuristics and biases — Science: Original anchoring evidence; not a sales-uplift guarantee.
- Online experimentation at Microsoft — Microsoft Research: Controlled-experiment evidence and interpretation boundaries.