Amazon Marketing Stream

Amazon Marketing Stream (AMS) is a push-based, near-real-time data feed that delivers hourly advertising metrics to advertisers via AWS. It is Amazon's structural replacement for the slow, pull-based, once-per-day reporting cadence of the legacy Ads API and Ads Console. Where the old world meant working with yesterday's data this morning and acting on it tomorrow, AMS collapses that loop to roughly one hour from the moment an impression or click happens.

For serious Amazon PPC operators, AMS is the single biggest infrastructural change Amazon has shipped in the last five years. It doesn't add a new ad type or a new placement. It changes what is operationally possible on every existing ad type.

What AMS actually delivers

Rather than the daily-aggregated CSVs of the Reports API, AMS pushes JSON event datasets to a customer-owned AWS SQS queue. Each dataset is keyed by the hour the data was generated and contains incremental updates — new impressions, new clicks, new conversions, new spend — broken out by campaign, ad group, target, search term, and placement.

The datasets currently available include (subject to ongoing expansion):

• sp-traffic — Sponsored Products impressions, clicks, spend, hourly.
• sp-conversion — Sponsored Products orders, units, sales, hourly, broken out by attribution window (1d/7d/14d/30d) where available.
• sb-traffic / sb-conversion — Sponsored Brands equivalents.
• sd-traffic / sd-conversion — Sponsored Display equivalents, including view-through metrics.
• budget-usage — hourly campaign budget consumption, including out-of-budget events.
• bid-recommendations — Amazon's suggested bids, updated frequently.
• campaign / ad-group / ad / target / negative-target / keyword — change-data-capture streams for every structural change to the account.

The "near real-time" in the name is accurate: latency is typically 30–90 minutes from event to delivery. That's not stock-exchange real-time, but it is more than fast enough to operate PPC as an intraday trading function rather than a daily batch job.

What changes operationally with hourly data

Hour-of-day bidding

Most categories have a strong intraday demand pattern: B2B office supplies peak mid-morning; consumer electronics peak in the evening; baby products spike around 9–11 PM. Without AMS, the only available bid is the 24-hour average, which over-bids during low-CVR hours and under-bids during high-CVR hours. With AMS, bids can be adjusted by hour of day, by day of week, and (with enough history) by hour × day-of-week × placement.

The lift from intraday bidding alone — separate from any other optimization — typically lands in the 8–18% efficiency-at-constant-spend range, depending on category seasonality.

Budget pacing and stockout avoidance

Without AMS, an out-of-budget campaign goes dark at 3 PM and the operator finds out tomorrow morning. With AMS, an out-of-budget event is delivered within the hour and budget can be topped up while the high-converting evening window is still live. The same plumbing supports proactive pacing: top up the budget of a campaign tracking to deplete by 5 PM, throttle the budget of a campaign converting poorly that day.

Placement intelligence

The legacy Placement Report aggregates to the day. AMS provides hourly placement performance — so the question "is my +50% top-of-search modifier still earning its premium at 11 PM on a Tuesday?" becomes answerable and actionable.

Anomaly detection

Search terms stop converting for many reasons: a competitor dropped price, a competitor's new product launched at the top of search, the listing image got flagged, a review went viral, an out-of-stock variation suppressed the parent. In the daily-reporting world, the anomaly is visible 24+ hours after it started, and the wasted spend is gone. With AMS, statistical anomaly detection — "this search term's CVR collapsed compared to its trailing 14-day baseline" — fires within 1–2 hours.

Re-thinking the search-term harvest loop

The harvest loop described under Sponsored Products historically ran weekly because the Search Term Report only refreshed daily and required statistical significance over several days. AMS pushes search-term-level events hourly, which means the loop can run continuously: harvest new converting terms within hours of their first conversion, route spend away from converting auto-campaign matches before the duplicate-spend window opens, and adapt to demand shifts in near real time.

Why most accounts under-use AMS

Hourly data is firehose volume. A modestly-sized account generates millions of AMS events per month. Three things have to be in place to extract value from it:

1. Ingestion plumbing — an AWS account, SQS queue, consumer process, a data warehouse (S3/Athena, Snowflake, or Postgres-with-discipline) to land and query the events.
2. Modeling layer — incremental aggregation by campaign × target × placement × hour, with idempotent re-processing for late-arriving events.
3. Action layer — an engine that turns the modeled data into bid changes, budget changes, and negative-keyword additions via the Ads API, with guardrails and audit logs.

Without all three, AMS is just expensive S3 storage. With all three, it is the foundation for 24/7 automated PPC operation.

AMALYZE's Amazon Ads module consumes AMS directly. The ingestion, modeling, and action layers are built in, so the operator works at the strategy and exception level rather than the data-engineering level.

What AMS does not solve

• It is not a new attribution model. Attribution windows (1/7/14/30 day) still apply, and ACOS still depends on which window is read.
• It does not surface organic data. AMS is an advertising stream. Organic search volume, organic rank, and PDP traffic still come from Brand Analytics, Search Query Performance, and SP-API reports.
• It does not eliminate the need for daily and weekly batch reconciliation. Late conversions still arrive against earlier clicks; the hourly stream is incremental and must be reconciled against the canonical daily reports.

Eligibility and access

AMS is available to advertisers via the Amazon Ads API. Eligibility has expanded substantially since launch; most active advertising accounts in mature marketplaces can access at least the traffic and conversion datasets. New datasets and new marketplace coverage are added on a rolling basis.

The strategic question is no longer "should we connect AMS?" — it's "do we have the engineering to use it, or do we route through a platform that already does?"