1. Compare the same part of the pricing cycle
A system can calculate a new price quickly while the marketplace displays it later. Another system may use a faster submission path only for selected listings. Comparing those claims without their boundaries is not a useful performance test.
- Observation received → first decisionHow long did the repricer take after it received a usable marketplace signal? This does not measure how long it took to discover the change.
- Decision → submissionWas the request queued, blocked, preview-only or actually sent? Report queue delays separately.
- Submission → acknowledgementDid the marketplace accept the request? Missing acknowledgements are not zero-latency successes.
- Submission → matching later readbackWhen was the requested price actually observed? Show failed and unresolved requests as well as the successful sample.
Ask for median and 95th-percentile latency, sample size, dates, marketplace, listing coverage and exclusions. Do not mix retries of an old event with the first response to a newly received event. A fast small subset is not automatically a fast whole catalogue.
For example, Repricer’s Priority Lane help page distinguishes calculation, upload and marketplace processing, while describing preferential handling for key products. That illustrates why the boundary matters; it is not an independently verified benchmark or a speed guarantee from Lumi. Read the provider’s explanation.
2. Verify Amazon and Walmart separately
A feature on one marketplace does not establish parity on another. Check your exact marketplace country, fulfillment type, signal source, price-write method and verification path.
Aura’s strategy guide dated July 27, 2026 states that Maven Level 1 is available for Amazon only and recommends Buy Box Matching as its Walmart alternative. This is a concrete reason to ask about marketplace-specific behavior rather than assume one headline feature works everywhere. Read Aura’s current strategy guide.
Amazon’s ANY_OFFER_CHANGED documentation describes notifications for active offers and changes involving its top-20 offer context. Notification support does not mean complete visibility into every offer. Walmart’s own documentation also describes inventory, delivery and content as Buy Box considerations beyond price. Amazon notification definitions · Walmart repricer overview.
For Lumi, consult current connection guidance and distinguish sandbox compatibility from a real seller’s confirmed price cycle. Do not treat this educational page as certification that every marketplace/country is generally available.
3. Ask for contribution, not just revenue growth
Suppose a product earns $5.25 contribution per unit before a price cut and $4.40 after it. Under unchanged costs, it needs about 19.3% more units to produce the same total contribution. Selling more is not enough when each unit contributes less.
Break-even volume increase = old contribution per unit ÷ new contribution per unit − 1
A historical performance comparison should use the economics that applied to each included order. Fees, shipping, refunds and cost changes must have a documented basis. Today’s cost is not a substitute for a missing historical snapshot. Avoid summing currencies without an explicit exchange-rate method.
Period contribution margin = total included contribution ÷ total included revenue
A simple average of percentages can mislead. A $100 revenue line earning $20 and a $900 revenue line earning $90 combine to an 11% contribution margin, not the 15% average of their individual percentages. See the free workbench for scenario calculations.
4. Distinguish improvement from attribution
A before-and-after chart can describe improvement without proving that repricing caused it. Promotions, seasonality, availability, competitor exits and changes in product mix may also affect the result.
A more useful evaluation defines the primary metric and comparable control/treatment groups in advance. Avoid splitting the same competitive listing across opposing treatments. Record stockouts and strategy changes, allow an appropriate adjustment period, and report uncertainty. Do not choose a winning time window only after looking at the result.
When a comparison lacks enough evidence, “incremental contribution not yet measured” is more informative than a made-up uplift. A replay can test whether a strategy would respect its limits; it cannot prove what shoppers or competitors would have done.
5. Test difficult decisions, not just favorable examples
Ask each provider to explain a price cut, an upward test, a floor hold and an acknowledged request without a matching readback. Review how stale inventory, changed costs, a manual edit and a rejected update appear to the seller.
Raising prices is not a unique feature by itself: Repricer publicly describes upward/downward actions and margin controls. Evaluate the actual bounds, recovery behavior, evidence retention and contribution measurement rather than count feature labels. Repricer’s published feature list.
What would a controlled upward test establish?
It should start from a fresh confirmed price and valid economics, remain within your approved limits, observe the subsequent marketplace state and stop or recover when evidence is incomplete. A higher request alone establishes neither continued Buy Box ownership nor additional profit.
Your strategy should serve your own business goals using permitted marketplace evidence. It should not coordinate prices with other sellers or use another customer’s private cost data.
6. Give your team a review they can repeat
Start with the exact store and SKU, choose the period, inspect exceptions and then open the supporting records. Keep current snapshots separate from period totals. An observation-based Buy Box rate should expose its denominator and should not be relabelled as shopper-impression share.
The downloadable scorecard records questions and evidence gaps; it does not award a security certificate or rank a provider automatically. Fill it in with your own listing examples during an evaluation.
Provider references were reviewed on September 5, 2026. Features, plans and availability can change. Vendor descriptions are attributed claims, not independent performance verification.