How to Govern Amazon DSP Frequency Caps

Inventory the advertiser accounts, campaigns, ad groups, frequency groups, audiences, publishers, deal types, devices, creatives, markets, and flights that can contribute exposure. Amazon's current help content says frequency can be controlled at several levels and that a frequency group applies a shared limit across multiple campaigns. Draw the hierarchy before changing a setting. A campaign cap layered with stricter ad-group caps can produce a different delivery pattern than a shared group spanning several campaigns.

Record whether each rule counts a user or household, the number of impressions, the time unit, start and end dates, owner, approval, and last change. Amazon documents daily, hourly, and shorter interval options in the DSP workflow. Do not assume the same option is available in every account, market, campaign type, or interface version. Verify the actual controls shown for the intended entity and capture the effective configuration after saving.

Write what the campaign is trying to achieve and why repeated exposure may help or hurt. A reach campaign, sequential creative plan, limited promotion, product-detail retargeting campaign, and long-consideration purchase do not share one defensible cap. Define the primary audience, expected flight, creative rotation, inventory breadth, and measurement period. Use prior observed reach and frequency as context, not as a promise that the same setting will reproduce an outcome.

Choose a starting cap that the team can explain and revise. State the number, time window, counting basis, level where it is applied, related group controls, and conditions that would trigger review. Amazon notes that caps can reduce repetitive impressions and redistribute opportunity toward additional unique users in supported delivery. That is a delivery mechanism, not evidence that a lower cap always improves conversions or that a higher cap always creates fatigue.

Amazon documents cross-publisher frequency capping with a unified counter across supported supply sources. Confirm which campaigns and publishers are inside the same counter before interpreting the result. A shopper exposed on one publisher may consume part of the cap before another publisher can serve. Preserve the publisher, campaign, ad group, deal, and group membership that was effective during each measurement window so a later configuration change does not rewrite the meaning of historical frequency.

Review deal limitations. Amazon says some programmatic guaranteed impressions may count toward a cap even when delivery on that guaranteed line cannot be suppressed, while enforcement can affect other eligible lines. Separate counted exposure from suppressible exposure in the operating record. If delivery looks inconsistent, inspect deal type, group membership, cap hierarchy, audience identity, and publisher support before assuming the platform ignored the rule.

Monitor reach, average frequency, frequency distribution, impressions, spend, delivery pace, viewability, creative-level engagement, and the campaign's approved outcome metric over a complete period. Break out important audiences and creative sets only when privacy thresholds and sample size support the view. A stable average can hide a long tail of heavily exposed people, while a lower click rate can reflect creative, supply, audience, placement, or timing changes rather than frequency alone.

Change one layer at a time and record the prior value, new value, effective time, approver, expected observation window, and rollback trigger. Pause experiments when creatives, targeting, supply, flight, conversion tracking, or budget materially change because those shifts confound the frequency comparison. Keep the previous configuration available for restoration. The goal is an explainable exposure policy that operators can audit across campaigns, not a single magic cap applied to every Amazon DSP program.

Hold a recurring review with media, creative, analytics, and account owners. Bring the exact effective caps, group memberships, campaign flights, audience definitions, publisher mix, creative versions, reach and frequency distributions, delivery changes, and approved outcome metrics. Start with configuration drift and data completeness before debating performance. A cap that was correct on paper but never saved, inherited from a group, or changed mid-window cannot support a clean comparison.

Record decisions as keep, lower, raise, split, consolidate, or investigate, with evidence, owner, effective time, and next review. Avoid rapid adjustments based on one incomplete day or one small audience cell. When the team cannot separate frequency from creative fatigue, supply change, targeting change, or tracking loss, hold the current setting and design a narrower observation. The review should make exposure policy legible across the account and surface contradictions before another flight reuses them.

Keep a historical register of every cap and group membership so analysts can reconstruct the setting that governed each reporting period. Include deleted groups and ended campaigns. Without that record, a later dashboard can show exposure outcomes while the configuration that produced them has disappeared. Treat unexplained gaps as missing evidence rather than backfilling a number from memory.