A Worked Example, Start to Finish
One question taken through data, checks, analysis and reporting, with the reasoning shown at each step including the parts that produced a weaker answer than hoped.
An illustration of the whole method applied to a single realistic question.
The question
A competitor's product appears to be gaining. Should we respond on price?
What decision this informs: whether to cut price, which costs margin immediately.
What would change the answer: if the gain is real and sustained, a response may be warranted. If it is a promotion, a stock-out elsewhere, or seasonal, it is not.
The data
Watch list: our product, the competitor, four other comparable products as a basket, the category leader.
Period: eight weeks.
Cadence: daily, fixed time.
Fields: overall rank, category rank with the category recorded, price, availability, review count, title, featured seller.
The checks, before any analysis
Pipeline health over the period: two gap days, both affecting all products, from a collection outage. Excluded.
Manual reconciliation: done in week two, ranks and prices matched.
Identifiers stable: titles unchanged throughout.
Rank type consistent: overall rank used for all comparisons.
All checks passed, which is worth stating in the output because it is what makes the rest credible.
What the data showed
The competitor's rank improved substantially over weeks three to six, then partially retreated.
Their price fell 18 percent at the start of week three and returned to its previous level at the end of week six.
The basket median was flat throughout, so this was not market movement.
Their availability was continuous.
Their review velocity rose modestly in weeks four to seven, consistent with a sales increase lagged by the review delay.
Our own rank worsened slightly during the same period, more than the basket.
The interpretation
The gain coincides exactly with a temporary price reduction and reverses when it ends.
The post-promotion level is slightly better than their pre-promotion level, but within the noise floor established in weeks one and two, so the durable effect is not demonstrated.
Our own decline during the period is real and modest, consistent with losing some price-sensitive volume to them temporarily.
Conclusion: this was a promotion, not a sustained competitive gain. A price response is not warranted on this evidence.
What was not established
Whether their promotion was profitable. Not observable.
Whether they intend to repeat it. Not observable.
How much volume we actually lost. Our own first-party data answers this and rank does not.
Whether their small post-promotion improvement is real. The window is too short relative to the noise floor; another month would settle it.
The report
Over eight weeks of daily observation, the competitor's rank improvement (weeks 3–6) coincided precisely with an 18 percent price reduction and reversed when the price returned. The comparison basket was flat, both products were available throughout, and no data gaps affected the period beyond two excluded days.
Their settled position after the promotion is marginally better than before, within the established noise range, so a durable gain is not demonstrated on this evidence.
Our own position declined slightly more than the basket during the promotion, consistent with temporary volume loss; the magnitude should be read from our own sales data rather than estimated from rank.
Recommendation: no price response. Re-examine in four weeks to test whether their post-promotion level holds.
What made this defensible
The basket, which ruled out market movement.
The price capture, which supplied the explanation.
The availability capture, which ruled out a supply artefact.
The noise floor, which prevented reading a small residual as a gain.
The explicit list of what was not established, which is what stops the conclusion being over-read.
None of it is sophisticated. It is the same five checks applied in order, and the whole analysis took an afternoon because the data had been collected properly for eight weeks beforehand.
What a different result would have looked like
The example concluded that a competitor's gain was promotional. Two other outcomes were possible and worth describing.
If the basket had risen with them, the conclusion would have been a category-wide movement, and the appropriate response is to understand the category rather than the competitor.
If their price had been unchanged, the gain would have needed another explanation — a listing change, advertising, external publicity, or a competitor's outage — and the next step would have been checking each rather than concluding a demand shift.
If their post-promotion level had held clearly above the noise floor, the conclusion would have been a durable gain, and a price response would have been worth considering with the margin arithmetic.
If our data had shown gaps or an identifier change, the honest output would have been that the analysis could not be completed, with a date when it could.
Four branches, all reachable from the same five checks. The method is not clever; it is applied consistently, and that is what makes the conclusion trustworthy in each branch.