Amazon Marketing Stream: Hourly PPC Data Guide (2026)

Marcus sells electrolyte powder. Good margins, crowded category, about $22,000 a month in ad spend. Last spring he got Amazon Marketing Stream wired into his data warehouse, and for six weeks he had the cleanest hourly view of his advertising anyone in his niche had.

He did nothing with it.

Not because he was lazy. Because at 4 PM on a Tuesday, staring at a chart showing his best campaign had burned through its daily budget at 1:47 PM for the ninth day running, he had no mechanism to respond. He could log into the console and raise the budget tomorrow, manually, if he remembered. By then the chart had already moved on.

That is the honest state of Amazon Marketing Stream for most sellers. The data is genuinely excellent. The decision layer behind it usually is not. This guide covers what Marketing Stream actually is, what datasets it ships, how to turn it on, the four plays where hourly data changes real money, and the three traps that make people distrust it.

What Amazon Marketing Stream Actually Is

Amazon Marketing Stream is a push-based data feed for Amazon Advertising. Instead of you asking Amazon for a report and waiting, Amazon sends advertising performance data to a destination you control, in near real time, roughly once an hour.

That word, push, is the whole story. Every reporting workflow you have used on Amazon until now is pull-based. You request a search term report, Amazon queues it, you poll for it, you download it, you parse it. The data arrives in daily buckets and lands hours after the day closed.

Marketing Stream inverts it. You subscribe once. Amazon delivers hourly messages to an AWS destination for as long as the subscription lives. Nobody requests anything.

Two consequences matter:

• Granularity: Records are stamped by hour, not by day. You can finally see that your Sponsored Products CPC is $1.12 at 6 AM and $2.40 at 8 PM instead of seeing a blended $1.71 and guessing.
• Latency: Data for the 2 PM hour typically lands within an hour or two of 3 PM, not the next morning. The feedback loop drops from roughly 36 hours to roughly 90 minutes.

What Amazon Marketing Stream is not: a dashboard, a reporting UI, or anything you can use without engineering. There is no "Marketing Stream" tab in Seller Central or the ad console. It is plumbing. You either build on it, or you use a tool that already has.

What Datasets Marketing Stream Ships

Subscriptions are per dataset, per marketplace, per advertising account. The catalog keeps growing, but the ones that carry weight for a typical seller are these.

Two of these are quietly more valuable than the rest.

budget-usage tells you a campaign hit 100% of its daily budget at 1:47 PM. No report in the ad console gives you that timestamp. The daily report only shows spend equal to budget, which tells you a cap was hit but never when.

The metadata datasets give you a change log. If your ACoS jumped on Thursday, you can look up every bid and budget change that landed Thursday, including changes made by a tool, a VA, or Amazon's own automated rules. Most sellers have no such log and spend hours guessing.

How To Set Up Amazon Marketing Stream

Setup is an engineering task, not a settings toggle. Four prerequisites, in order.

1. Amazon Ads API access. You need a registered application with an approved Login with Amazon client, an access token, and API access to the advertising account. If you have never used the Amazon Ads API, this is the long pole: approval takes days, sometimes weeks.
2. An AWS account. Marketing Stream delivers to AWS. Amazon Kinesis Data Firehose is the common destination for analytics use cases; Amazon SQS is the common choice when you want a queue your own service consumes.
3. A destination and permissions. Create the Firehose delivery stream or SQS queue, then attach a resource policy that lets Amazon's Marketing Stream principal write to it. This is the step people get wrong most often: a policy that is too narrow fails silently and you sit waiting for data that never arrives.
4. Subscriptions. Call the Marketing Stream subscriptions endpoint once per dataset you want. Each subscription is scoped to one dataset, one marketplace, one profile. A seller in the US, UK, and Germany who wants traffic, conversion, and budget usage is creating nine subscriptions, not three.

After that, data starts flowing. Expect your first messages within a couple of hours, and expect roughly one message batch per hour per dataset from then on, forever, until you unsubscribe.

The Four Plays Where Hourly Data Changes Money

Hourly data is only worth what you do inside the hour. These are the four uses that survive contact with a real P&L.

1. Find and Fix Budget Exhaustion

This is the play with the fastest payback, and it does not require dayparting or machine learning.

Take Marcus. His top campaign ran a $400 daily budget and hit the cap at 1:47 PM. His sp-conversion data showed that the 6 PM to 10 PM window carried his highest conversion rate: 14.2% against a 9.1% daily average. He was structurally absent during his own best four hours, every single day.

The math on that is not subtle. Those four hours would have taken roughly $130 of additional spend at his CPC. At a 14.2% conversion rate and a $38 average order value, that is about $470 in sales, or a 28% ACoS on the incremental spend, better than his 31% target. He was declining profitable volume because a number set in January had not moved.

Use budget-usage to list every campaign that hits 100% before 8 PM. Cross-reference the hours it is missing against conversion rate by hour. Then either raise the budget or move spend from a campaign that never caps. Our Amazon budget rules guide covers how to schedule those increases so they only apply on the days that need them.

2. Dayparting You Can Actually Defend

Most dayparting advice is folklore. "Turn off ads between 1 AM and 5 AM" gets repeated endlessly by people who never checked whether those hours were unprofitable for their product.

Lena sells German home textiles and was about to pause overnight on exactly that advice. Her Amazon Marketing Stream data said the opposite: 2 AM to 5 AM was 4% of her spend and 7% of her orders, at a 19% ACoS against a 27% blended average. Cheap clicks, low competition, insomniac shoppers with a cart already open. Pausing would have removed her most efficient hours.

Hourly data turns dayparting from a rule of thumb into an argument with evidence. Pull at least four weeks of sp-traffic and sp-conversion, group by hour of day, and only act on hours with enough clicks to mean something. The full method is in our Amazon PPC dayparting guide: the analysis has not changed, but Marketing Stream makes the input data far better than what you can reconstruct from daily reports.

3. Shorten the Negative Keyword and Bid Feedback Loop

A search term that burns $90 with zero conversions is expensive on a daily report because you find out tomorrow. On hourly data you find out this afternoon.

The honest caveat: search term data does not arrive in Marketing Stream at the same granularity as targeting data, so this play works at the target and campaign level first. What you gain is speed on bid corrections. A target whose CPC doubled at noon because a competitor raised bids is visible at 1 PM, not Thursday. Our breakdown of how the Amazon PPC auction works explains why those intraday CPC shifts happen and what an appropriate response looks like.

4. Build a Real Change Log

Subscribe to the metadata datasets and you get a timestamped record of every bid, budget, status, and name change on the account. When performance moves, you stop theorizing. You look up what changed and when. For anyone running bulk operations across thousands of targets, this is the difference between a controlled process and a mystery.

Three Traps That Make Sellers Distrust the Data

Trap 1: Conversions Get Restated

Amazon attributes a conversion to the hour of the click, not the hour of the purchase. Someone clicks at 9 AM Monday and buys Wednesday night; that sale lands against Monday 9 AM and arrives in your feed days later.

So the 2 PM record you received at 3 PM is not final. It will be revised upward for up to 14 days. Every team that builds on Marketing Stream without understanding this makes the same mistake: they look at today's ACoS, see a catastrophe, and slash bids on data that is 40% incomplete.

Two rules fix it. Store records as append-only and take the latest version per key rather than overwriting. And never make a bid decision on data less than 72 hours old without inflating for expected backfill.

Trap 2: Single Hours Are Noisy

An hour with 11 clicks and one order shows a 9% conversion rate. An hour with 11 clicks and zero orders shows 0%. The underlying truth is identical; the sample is just tiny.

Set a minimum click threshold before you act. Thirty clicks in an aggregated hour bucket across four weeks is a reasonable floor for a mid-size account. Below that, aggregate into three or four dayparts instead of 24 hours and act on those.

Trap 3: It Costs Real Money and Real Attention

Firehose ingestion, S3 storage, warehouse compute, plus the engineering hours to build and maintain it. For an account spending $8,000 a month on ads, a data pipeline that costs $200 a month and a day of engineering time per quarter is hard to justify. For an account spending $150,000, it is trivially worth it.

Marketing Stream vs Ads API Reports vs AMC

Three Amazon data sources get confused constantly. They solve different problems.

They stack rather than compete. Marketing Stream tells you what is happening now. Ads API reports remain your source for search terms and standard reconciliations: see our Amazon advertising reports guide for which report answers which question. Amazon Marketing Cloud answers the strategic questions neither of the others can touch.

Who Should Actually Turn This On

Be honest about which row you are in.

• Under $10,000/month in ad spend: Skip the build. Your hours are too sparse to be statistically meaningful, and your time is better spent on listing conversion rate. Use a tool that handles intraday optimization for you.
• $10,000 to $50,000/month: budget-usage alone is worth it. Consider a minimal pipeline, or let software do it. Full warehouse builds rarely pay off here.
• $50,000+/month, multiple marketplaces: Build it or buy it, but do not go without. At this spend, one budget cap at 2 PM across a dozen campaigns costs more per month than the entire pipeline.
• Agencies and aggregators: Non-negotiable. The metadata change log alone justifies it across a portfolio of accounts.

Frequently Asked Questions

Is Amazon Marketing Stream free?

The feed itself has no fee from Amazon. You pay AWS for the destination, storage, and processing, plus whatever engineering time the build consumes. Budget $50 to $400 a month in AWS costs for a typical mid-size seller, depending on how many datasets and marketplaces you subscribe to.

Do I need an AWS account to use Amazon Marketing Stream?

For a direct subscription, yes. Delivery targets AWS destinations such as Kinesis Data Firehose or SQS. If you use a third-party tool that already ingests the feed, the AWS account is theirs, not yours.

How real-time is "real time"?

Hourly, not instant. Data for a given hour typically arrives within one to two hours of that hour closing. It is fast enough to respond the same day and far too slow for anything that needs second-level reaction.

Can I use Marketing Stream for dayparting?

Yes, and it is the best available input for it. But Marketing Stream only supplies the data. It cannot change a bid. You still need scheduling rules, budget rules, or automation software to act on what the data shows.

Does Marketing Stream include search term data?

Not at the same fidelity as the search term report. Treat Marketing Stream as your hourly view of targets, campaigns, and placements, and keep pulling the standard search term report for query-level decisions.

Why do my Marketing Stream numbers not match the ad console?

Usually attribution backfill. Console figures have had days to settle; fresh Stream records have not. Compare like for like by looking at data at least 14 days old, and the two should reconcile closely.

The Part That Actually Matters

Amazon Marketing Stream removed the excuse. Sellers used to say they could not optimize intraday because Amazon did not give them intraday data. Amazon gives it now, hourly, pushed to your own infrastructure, for free.

What replaced that excuse is a harder problem. Hourly data is only as valuable as the speed of the decision behind it. A feed that updates every hour feeding a person who reviews campaigns every Sunday produces a very well-documented account of money you already lost.

So before you open an AWS console, answer one question: when the data says your best campaign capped out at 1:47 PM, what happens next, and how long does it take? If the honest answer is "I notice it next week," fix the acting half of the loop first. Amazon Marketing Stream will still be there, and it will be far more useful once something is listening.

Marcus eventually did. He kept the pipeline for reporting, handed intraday bid and budget management to automation, and stopped watching charts he could not act on. His ACoS moved from 31% to 24% over the following quarter. The data had been telling him that the whole time.