Marketing attribution software: the kinds, what each counts, and how to choose

"Attribution software" covers tools that do very different things. Some report what an ad platform credits to itself, others track shoppers across channels, and a few don't track shoppers at all and work from totals. Here's how they compare:

Several products now sell more than one kind. Triple Whale, Northbeam, Haus and Recast each describe a mix of attribution, modeling and testing on their sites. Judge each part by what it actually measures.

Every ad platform reports the sales it credits to its own ads, using its own window and rules. That's the first number most brands see, and it costs nothing extra. Meta Ads Manager reports the purchases Meta credits to Meta ads. Your Meta pixel runs on your own store, not on Amazon's pages, so an Amazon purchase never reaches that report.

Amazon Attribution fills part of that gap. Amazon describes it as a free measurement tool for the on-Amazon impact of marketing outside Amazon: search, social, display, video and email. You add an Attribution tag to each ad's link, and Amazon reports clicks, detail page views, add-to-carts, purchases, product sales and new-to-brand sales. Its reports use a 14-day attribution window.

Amazon says it's available to professional sellers enrolled in Brand Registry, vendors, KDP authors and agencies, through Amazon Ads Agent or tool providers using the Amazon Ads API. It counts shoppers who clicked a tagged link. Someone who sees an Instagram ad and later searches Amazon for your brand isn't in it. The Facebook ads for Amazon guide covers setup.

Google Analytics 4 credits key events on your website or app to the channels that brought the visitor. As of October 2026, Google's help center lists three models in GA4's Attribution reports: data-driven, paid and organic last click, and Google paid channels last click. Google says first click, linear, time decay and position-based models were removed in November 2023, and that all models leave out direct visits unless the whole path was direct.

Google describes its data-driven model as comparing the paths of people who converted with those who didn't, and says it's trained on data from randomized controlled trials of Google ad exposures. It still divides up events that happened on your site. A Meta ad that leads to an Amazon order never reaches GA4. The attribution models guide explains each rule and the windows Meta, Google and Amazon use.

Multi-touch tools put their own pixel on your store, connect to your ad accounts, and follow each shopper's clicks and views across channels. They then split each order's credit among those touches. Because one tool does the counting, the channels stop adding up to more than your sales, which is a real fix for a real problem.

Here's what two of them say, as of October 2026. Northbeam's homepage describes multi-touch attribution, incrementality and media mix modeling, plus a view-through model it calls Clicks + Deterministic Views. Triple Whale's homepage describes Compass, which it says combines attribution, marketing mix modeling and incrementality across DTC, Amazon and retail, with GeoLift tests. It also says its AI agent, Moby, can scale budgets, pause ads and launch campaigns once connected to your ad accounts.

A pixel sees touches, not intent. A shopper who would have bought anyway still leaves a trail of ad touches, and the tool still credits them. That's why several of these products now add tests or models next to their attribution.

A marketing mix model (MMM) doesn't track shoppers. It estimates each channel's effect from weekly spend and sales, plus price, promotions and season. Meta publishes Robyn and Google publishes Meridian as free, open-source MMMs, and paid vendors run models for you. The marketing mix modeling software guide covers the options and the data they need.

An incrementality test changes one thing on purpose: ads stay off for a holdout group of people or regions, and you compare sales. GeoLift is Meta Open Source's free package for geo-level lift, and its site says it's for research purposes and not a Meta product. Haus describes incrementality experiments alongside its Causal MMM and Causal Attribution products. Recast describes designing and analyzing GeoLift experiments whose findings feed back into its Bayesian MMM.

Every attribution report answers "which ads did buyers touch?" The budget question is different: "which sales wouldn't have happened without the ads?" That's incrementality, and only an experiment measures it directly. In large experiments at Facebook, Gordon and colleagues found that common non-experimental methods often missed the experiment's answer, sometimes by a lot.

A made-up brand spends $4,000 a month on Meta ads, and its attribution tool credits Meta with $12,000 of site sales. A geo holdout test finds the ads caused $6,000 of site sales and $2,500 of Amazon sales. That's $8,500 in all, an incremental ROAS of 2.13 against a credited ROAS of 3.0. The tool over-credited the site and couldn't see Amazon.

Neither number is useless. The test gives you a ratio of caused to credited sales, and you can use it to read the attribution report between tests. The attribution gap calculator sets the numbers side by side.

• Data sources, Amazon included. List where your sales happen and ask the vendor how each one gets in. If a share of your sales lands on Amazon, find out whether the tool reads Amazon orders, only Attribution-tagged clicks, or nothing.
• Method. Ask which of the five kinds each number comes from: a platform's own report, a rule, a learned model, a mix model or an experiment. A vendor should be able to name it and show you the documentation.
• Uncertainty shown. A single number with no range hides how sure the tool is. Prefer tools that report a range, and test designs that say in advance how big an effect they can detect.
• Cost of switching. Pixel tools start collecting on the day you install them, so history doesn't move with you. Check whether you can export your data and results, and what happens to them if you cancel.
• Who makes changes. Some tools can change budgets and campaigns in your ad accounts. Decide whether you want that, and keep a record of what the tool predicted each change would do.

Spillmark runs geo holdout tests for brands that advertise on Meta and sell on Amazon. It plans the test from your own order history and counts Shopify orders by ship-to state as well as Amazon's, so one test shows the lift on each store. It locks the design before the test starts and reports the range with every result.

Size a test first with the holdout test calculator, and see how lift is calculated for the method.