How to Audit Amazon Ads Targeting Reports

Amazon describes the Sponsored Products targeting report as a report with sales and performance metrics for keywords, products, and categories that received at least one impression. That makes the target the primary unit of analysis. It is not the same unit as a customer's shopping query. Amazon's search term report identifies customer search terms that produced at least one ad click, while the advertised product report organizes performance around promoted products. Begin an audit by writing the question in the same grain as the report: which configured targets received delivery and what attributed outcomes were recorded for those targets during a fixed period?

Record the advertising account, marketplace, currency, campaign type, report type, requested date range, generated time, attribution behavior shown in the account, and any delay that could leave recent rows incomplete. Preserve the original export. Then map campaign and ad group identifiers to readable names without discarding the identifiers. Campaign names can change; stable IDs make the review reproducible. If a column definition is unclear, use the current console documentation for that account rather than guessing from a third-party template. A row is useful only when its scope and measurement rules are understood.

A target row without its campaign context can support the wrong action. Join the target to campaign status, budget, bidding strategy, placement adjustments, ad group, advertised ASIN, target expression, match type, current bid, and product availability. Amazon documents automatic and manual targeting options for Sponsored Products, including keyword and product targeting. Preserve that distinction. An automatic target expression, a manual keyword, a category target, and an individual product target may require different review logic even when the exported metrics share column names.

Add product and operating context next: selling price during the period, promotion dates, offer eligibility, stock status, fees, returns, and contribution guardrails when those records are available. Do not convert attributed sales into profit without the required cost records. Mark targets with incomplete spend, order, or product context as review-only. Also flag simultaneous changes to budgets, bids, placements, negatives, product content, price, or inventory. Those changes may explain a difference in delivery, but the targeting report alone does not establish which change caused it. The joined table should make uncertainty visible instead of hiding it.

Create decision classes that preserve sample size and business context. A practical ledger can include observe, investigate, bid review, targeting review, negative-review candidate, product-page review, inventory hold, and no action. Define the minimum evidence for each class before sorting. For example, a target with impressions but no clicks answers a delivery and relevance question, while a target with clicks but no attributed orders may need more complete time or product-page review. A target with attributed sales still needs contribution context before it earns more budget. Avoid universal click, order, or efficiency thresholds that are not grounded in the account's economics.

Use related reports to test the proposed decision. Amazon recommends reviewing search term reports from automatic targeting campaigns and describes match types as controls over which searches an ad may be eligible to show against. If the target appears inefficient, inspect the search terms, advertised products, and placements before adding a negative or reducing a bid. A broad target can contain both useful and irrelevant queries. Likewise, one search term can appear under different campaign structures. Keep the canonical owner for each intended query visible so the audit does not create duplicate targets that compete without a documented reason.

For each approved action, record the target identifier, prior setting, new setting, reason, supporting export, approver, change time, and rollback condition. Make one control change when possible: bid, target state, match type, negative, or campaign placement, not a bundle of untraceable edits. Use a bounded canary for changes that could materially reduce reach or increase spend. Preserve important branded, defensive, or product-specific targets until their role is understood. A low-volume row is not automatically waste, and a high-attributed-sales row is not automatically profitable.

Review the change after a later complete window that matches the original scope. Confirm delivery, spend, search terms, attributed orders and sales when available, product availability, and any concurrent edits. Label the result observed, inconclusive, or reversed. Do not claim the change improved performance merely because a later period is better; calendar demand, competition, price, inventory, and other controls may also have changed. The durable outcome of a targeting audit is a governed ledger connecting each target to evidence, an explicit decision, and a scheduled review. That system is more valuable than a one-time spreadsheet sorted by one metric.

Keep rejected actions in the ledger with their evidence and reason. That prevents the next reviewer from repeating an unsupported edit, preserves institutional context when campaign names change, and makes later reconsideration depend on new information rather than a different spreadsheet sort.