Amazon Marketing Mix Modeling: data feed, MMM API and BYOM in AMC
In short: Amazon offers three MMM delivery paths: the data feed, the asynchronous MMM API and Bring Your Own Models (BYOM) in Amazon Marketing Cloud. All deliver Amazon signals with up to three years of history. The data feed and API are free; BYOM in AMC is free to access, but AWS compute costs apply.
What a marketing mix model is
A marketing mix model answers an allocation question: how much of your total budget belongs on which channel? Amazon describes MMMs as privacy-friendly, data-driven statistical tools that quantify the holistic incremental sales impact and ROI of both marketing and non-marketing activities, taking into account factors beyond media such as product pricing, promotional events and competitor activities. That view of non-media factors is what separates an MMM from a plain channel readout.
How strongly MMMs work in practice, Amazon quantifies with its own internal data: marketing mix models inform over 60% of cross-channel budget allocation decisions, and three out of four marketers use MMMs as a source of truth for the sales impact of each channel and publisher. Those figures come from Amazon Internal Data (US, 2019) and are not an independent market study; we name them here because they explain why Amazon ships those signals via MMM, API and AMC.
The three delivery paths compared
Amazon bundles the MMM signals into three solutions. The Amazon MMM Data Feed delivers ready-to-analyze data, the Amazon MMM API exposes the same signals programmatically through 9 endpoints, and Amazon BYOM in AMC runs your own models with Amazon signals in a secure AWS environment. All three deliver media and retail signals with up to 3 years of daily/weekly data.
The difference is not the data but the operating model: one-off, automated or self-run. That decides latency, costs and who keeps the methodology. The table below summarises the paths; the details follow in the sections after it.
| Path | Access | Delivery | Costs |
|---|---|---|---|
| Amazon MMM Data Feed | Manager account or S3 | typically 10 business days | free |
| Amazon MMM API | Amazon Ads API and MMM registration | within 24 hours | free |
| Amazon BYOM in AMC | AMC instance, BYOM in beta | in the clean room, compute on AWS | free to access, AWS compute costs for your own models |
The data feed: registration and delivery
The data feed is the classic entry point and runs through a self-service portal in your manager account. The flow Amazon documents: you register for the MMM service and sign the Data Usage Agreement, create a report with advertiser, geography and data frequency, validate the auto-populated ASINs and campaigns (DSP and sponsored ads) and submit the report for review.
Delivery then happens via the manager account or straight into your S3 bucket, typically within 10 business days. Amazon points to the path in the Ads Console under Measurement & Reporting > Marketing Mix Models. Important for planning: the 10 business days sit between request and delivery, not between model build and result.
The MMM API: asynchronous and automated
The Amazon MMM API is an asynchronous interface: you submit a request, Amazon prepares the report in the background, and you collect it later. The report contains aggregated Amazon sales and media performance signals used for modelling. Amazon provides 9 endpoints for it, which let you automate the data pipeline and the model refresh.
Amazon names two requirements: onboarding with the Amazon Ads API and registration for the MMM service through a manager account. The advantage over the feed is repeatability: a refresh runs without the manual portal path, typically with delivery within 24 hours instead of the 10 business days of the feed.
BYOM in AMC: your own models in the cloud
Amazon BYOM in AMC is the path when the methodology stays with you. You build and deploy your MMM with Amazon signals, for example organic search, and your own methodology, with full control of data, models and measurement. The environment is a secure AWS environment; your modelling logic does not leave the protected space.
The status matters: BYOM in AMC is available in beta in the following locales: United States, Canada, France, Germany, Italy, Spain, and United Kingdom. BYOM in AMC is therefore a beta feature in seven markets, while the data feed and the API are available in many more. If you plan around BYOM, you plan a beta function with fewer locales.
Data scope, grain and granularity
The scope is the reason Amazon offers these signals at all: Advertisers can access up to three years of historical, geo-level data globally, and all three solutions deliver up to 3 years of daily/weekly data. The package covers advertising performance, retail sales, promotional activity, add-to-carts and further proprietary BYOM signals such as ratings and reviews and organic search.
You choose the grain when you request the report. Amazon introduced daily-grain access on April 15, 2024: Select daily or weekly grain from within the MMM report request settings. Amazon's reasoning: a daily grain allows models to run over calendar months without missing or adding weeks that span month boundaries, and makes the signals comparable across publishers.
Who can use MMM, and in which markets
Access is broader than the name suggests: Amazon Marketing Mix Models solutions are available to all active Amazon Ads advertisers, including sellers, vendors, agencies, and advertisers that don’t sell products in the Amazon store. Advertisers that do not sell in the Amazon store are therefore part of the target group. The requirement is an Amazon Ads manager account.
One restriction remains: the total sales model is reserved for advertisers who sell their products 1P on Amazon, and therefore also for agencies and MMM providers that work with 1P vendors. For availability Amazon names different markets per path: the data feed and the API are available in 21 locales, BYOM in AMC only in the seven beta markets listed.
What Amazon MMM costs
The cost question has two answers, and mixing them builds the wrong budget. The Amazon MMM Data Feed and API are completely free for all eligible advertisers. So you pay nothing for the signals themselves.
BYOM in AMC is free to access, but running your own models costs money: BYOM in AMC is free to access; however, standard AWS compute costs apply when running proprietary models within the secure clean room environment. The data feed and API are free, access to BYOM in AMC is free, and you pay standard AWS compute costs for your own models in the clean room.
MMM, AMC and incrementality are not the same
Our assessment: these three tools answer three different questions, and an MMM does not replace a holdout test. An MMM answers the allocation question across the media mix. Amazon Marketing Cloud is the query and clean-room surface through which you analyse Amazon signals and run your own models. An incrementality test answers the causal question of how many purchases would not have happened without your advertising. If you want to estimate incremental impact, you need both: the model for allocation, the test for causality.
AMC Custom Models for Audiences is also not the same thing; Amazon launched it on October 31, 2025. That feature targets audience-based models such as CLTV or new-to-brand prediction and runs on AWS Clean Rooms; in its launch announcement Amazon highlights that model algorithms are never shared with AMC. The MMM BYOM in AMC follows the same security principle, but it is the path for your media mix models.
Which path fits when
Our assessment on sequencing: start with the data feed if you need a model once for a decision, or if you want to check first whether the Amazon signals support your question at all. That is the cheapest path and delivers a full data set in about 10 business days.
Move to the API once a refresh is needed regularly, because a budget cycle or a model validation cannot wait ten business days. BYOM in AMC pays off when your methodology is a differentiating asset but the data quality is already confirmed: it is the most demanding path, because compute time, model code and the beta status carry a permanent load in your planning.
Step by step
- Set up an Amazon Ads manager account. Access to all MMM solutions requires an Amazon Ads manager account. Without it there is neither the MMM portal nor the registration for the MMM service.
- Sign the data usage agreement. On the first run you sign the Data Usage Agreement and register your brand. The agreement is emailed to you and is the prerequisite for every report request.
- Create a report with advertiser, geography and frequency. In the Ads Console under Measurement & Reporting > Marketing Mix Models you choose advertiser, geography and data frequency. You set the grain yourself: Select daily or weekly grain from within the MMM report request settings.
- Validate ASINs and campaigns. Amazon pre-fills the list of ASINs and campaigns; you review it and correct it where needed. DSP campaigns and sponsored ads are covered. A wrong assignment travels straight into the model.
- Submit the report for review. The report goes into review. Only then does delivery start, which for the data feed typically takes 10 business days.
- Collect the data and model. You collect the data via the manager account, S3 delivery or the MMM API, which typically delivers within 24 hours. In the model you decide whether the Amazon signals carry it alone or get combined with your other channels.
Frequently asked questions
What does Amazon MMM cost?
The data feed and the MMM API are free for all eligible advertisers. BYOM in AMC is free to access, but standard AWS compute costs apply when you run your own models in the clean room. A budget therefore only needs its own line if you operate the models yourself.
How long does it take to get the MMM data?
With the data feed it is typically 10 business days until delivery via manager account or S3. The MMM API instead delivers within 24 hours. The waiting time therefore sits between request and data, not between model build and result.
Do I need Amazon Marketing Cloud for this?
Not for the data feed or the API: both hand the signals to your own model. You only need AMC for BYOM, because your models run there in the protected environment. BYOM in AMC is in beta and available in seven markets.
How much history is available?
All three paths deliver up to 3 years of daily/weekly data, and Amazon also names up to three years of historical, geo-level data globally. Since April 15, 2024 you can choose between daily and weekly grain in the report request.
What is the difference between MMM and an incrementality test?
An MMM allocates budget across the media mix, an incrementality test causally measures how many purchases would not have happened without your advertising. The model does not replace the test: for incremental impact you need both. Amazon DSP provides the holdout groups, AMC evaluates them.