Keyword relevance can't be predicted from theory — not from Amazon's relevance algorithm and not from human search behavior. The rule (credited by one practitioner to 'Ritu'): commit a set dollar amount to a keyword before making a keep-or-kill call, rather than trying to reason your way to a verdict in advance. The spend threshold should scale with product price so that it produces at least one real sale-or-no-sale data point before judgment.
This reframes keyword testing as a paid research cost you're expected to incur, not a targeting mistake to avoid — you spend to learn, not to guess correctly upfront. It pairs with the Amazon Search Term Report and a 7-10 Day PPC Data-Collection Cooldown for gathering the data, and complements decision frameworks like "Right to Win" Keyword Selection Framework and Per-Keyword Rank Ceiling & SQP Competitor Conversion Benchmarking for what to do once the data is in.