Method · score-v1.0.0

Six signals, one formula, no model

Every ASIN gets a 0–100 score, a coverage figure and one headline flag. The score is reproducible: the same CSVs and the same version always produce the same order.

The signals and their weights

SignalWeightInput
Monthly unit sales30%Monthly unit sales on the product sheet
Commission per sale20%Price × the best open campaign commission rate
Creator competition20%Creator videos already reviewing the ASIN
Category adjacency15%Share of realised commission in that category
Campaign budget health10%Remaining budget, spend already used, days left
Sample likelihood5%Whether a funded campaign is still open

Normalisation

Each raw value is mapped to 0–1 before weighting, so no single signal can dominate because of its units.

  • Sales — logarithmic to 5,000 units/month: log10(1+units) / log10(1+5,000). Logarithmic because the difference between 10 and 100 units matters more than between 5,000 and 5,100.
  • Commission per sale — price × rate, capped at $60 per sale. A $2 commission on a $10 product and $2 on a $200 product are worth the same to this ranking.
  • Competition — inverse logarithmic: 0 creator videos scores 1.0, 300 or more scores 0.0.
  • Adjacency — the category's share of realised commission. Categories that paid nothing score 0; that is a real signal, not missing data.
  • Budget — 1.0 when a campaign is open, has at least 60 days left and has already spent something; 0.5 when only one of those holds; 0.3 when it is open but unproven; 0.0 when exhausted or ended.
  • Sample — 1.0 for an open funded campaign, 0.5 for a campaign that exists but is closed.

Missing evidence is not zero

If a product sheet has no sales figure, or a campaign row has no budget, that signal is dropped from the weighted mean and the remaining weights are renormalised. The share of weight that had real input is stored as coverage. Anything below 60% coverage is flagged data_missing even when its score looks strong, because a high score on two signals is not a recommendation.

This is deliberate: counting a missing value as zero would bury a good product that simply was not enriched, and silently imputing a value would invent evidence.

Flags

FlagRule
dead_weightKnown monthly sales of zero, or every campaign for the product has spent nothing and none is open
data_missingCoverage below 60%
high_potentialScore ≥ 70, 500+ monthly units, fewer than 50 creator videos, an open funded campaign
low_hanging_fruit200+ monthly units and fewer than 12 creator videos
on_my_storefrontThis ASIN already paid commission
variation_i_ownA variation of a parent ASIN that already pays
watchNothing above applies yet

The headline flag follows a fixed precedence — dead weight, data missing, high potential, low-hanging fruit, on my storefront, variation I own, watch — so a row cannot be read two ways.

Measuring the ranking

A score is a prediction. The desk records the prediction and the outreach state, then joins realised commission back onto the same ASIN 30, 60 and 90 days later. The number reported is precision at the top of the list — how many of the first 30 filmed products earned more than the account's median — not the number of rows produced.

Weights change only as a version bump, with the before/after precision in the run log. A weight change without a measurement is not allowed to look like an improvement.

Known limits

  • Sales and creator-video counts are only as fresh as the product sheet that was imported. The tool shows the import date and never pretends the figures are live.
  • Category adjacency is only as good as the category column; products with no category lose that signal entirely.
  • Creator Connections budgets move. A campaign can be exhausted between imports.
  • Nothing here predicts Amazon's own ranking. It ranks the products worth a video, which is a different question.
  • The score says nothing about whether a product is safe, legal to promote, or a good fit for the audience. That judgement stays human.