---
canonical: "https://www.softatlas.io/list/incrementality-testing"
title: "Best Incrementality Testing Platforms for 2026"
locale: "en"
market: "International"
tool_count: 11
updated: "2026-10-12"
---

# Best Incrementality Testing Platforms for 2026

The 2026 field splits into three families. Dedicated platforms: Haus (widest range of test designs), INCRMNTAL (no holdouts needed) and Stella (public pricing, free plan). Measurement suites with a testing product: Cassandra, Lifesight, Measured, Polar Analytics and Recast. Free options: Meta's open-source GeoLift, and the lift studies built into Meta and Google Ads.

## What this category covers
Incrementality testing platforms answer one question: what would have happened without this spend? They answer it with experiments rather than attribution. The standard design is a geo test. Advertising is paused, or increased, in some regions while comparable regions carry on as before, and the difference in sales is the lift. Because no two regions are identical, most tools now build a synthetic control, a weighted blend of untouched regions that tracks the history of the test regions, and read the result against it. Tests typically run two to six weeks.  Two things have changed since 2024. Experiments have become a product rather than a consulting project, with self-serve design tools, public prices in a few cases and a free plan at one vendor. And they have merged with marketing mix modeling: almost every vendor here feeds test results into a model, on the principle that an experiment tells the truth about one channel for one month, and a model spreads that truth across the whole budget. If the model is what you are choosing, the Soft Atlas guide to marketing mix modeling tools covers that side.

## Methodology
We assessed each tool on four criteria: (1) experiment methods, meaning the test designs offered and how results are read, (2) channel coverage, including offline media and retail sales, (3) MMM calibration, that is whether test results feed a marketing mix model, and (4) how easy it is for a marketing team to get a test run, alone or with the vendor. Assessments are based on each vendor's public product pages, pricing pages and documentation, read in October 2026, and for the free options on Meta's and Google's own help pages and the GeoLift repository. Prices are published list prices at that date, in the currency the vendor displays. Pricing is described rather than scored. Tools are grouped by family (dedicated platforms, measurement suites with a testing product, free and native options) and listed alphabetically within each family; the order is not a ranking. Vendors can suggest an entry or a correction; entries are written and scored by Soft Atlas from public sources, and no placement is paid. Last updated October 2026.

## The tools

### 1. [Haus](/haus.md)
Haus, founded in San Francisco in 2021, has become the reference for geo experiments among US consumer brands: Oura, AG1, FanDuel, Sonos and Wayfair appear on its site. It offers three designs. GeoLift assigns test and control regions by stratified random sampling. Fixed Geo Tests let you pick the regions yourself, which is what an outdoor or radio campaign usually requires. Time Tests compare a period with a forecast, for channels that cannot be split by geography, such as linear TV or influencer work. Haus builds **synthetic controls** rather than matching markets one to one.  On top of the experiments sit a Causal MMM, refreshed weekly and built on the test results, and a causal attribution product. The platform is self-serve on the entry plan; a measurement strategist and access to data scientists come with the Core plan.  - **Strengths:** three test designs covering digital, TV, outdoor and retail; synthetic controls; an MMM built on the experiments; a strong roster of consumer brands; warehouse connections from the Core plan. - **Limitations:** no public price on any of the four plans; the entry plan is for Shopify brands, measures Shopify and Amazon only, and is limited to the United States; time tests, fixed geo tests and the MMM sit on higher plans.

### 2. [INCRMNTAL](/incrmntal.md)
INCRMNTAL, based near Tel Aviv, takes the opposite position to everyone else in this guide: it argues that you should not have to run experiments at all. Its model treats the changes you already make (a budget raised here, a campaign paused there) as micro-experiments and estimates their incremental effect, with insights available within days of an activity. It works on aggregated spend and KPI data, accepts no user-level data and needs no SDK, pixel or tag. Channel coverage is broad, from mobile and web to TV, connected TV, outdoor and podcasts, and customers include app and game publishers such as Fingersoft and Hopper.  The stance has a cost in comparability. INCRMNTAL states that it is **not an MMM provider** and dismisses calibrating a model with lift tests, so it will not slot into the test-then-model workflow the other vendors share. Pricing is by plan and unrelated to ad spend, but the figures are not public.  - **Strengths:** no holdouts to plan or to pay for in lost sales; works on aggregated data only; wide channel coverage; pricing unrelated to ad spend; onboarding in about two weeks. - **Limitations:** a model-based read, not a controlled experiment; no MMM calibration; prices not public; TV, connected TV and podcast are listed on the Enterprise plan only; a three-month proof of concept that renews for twelve months.

### 3. [Stella](/stella-2.md)
Stella is the easiest tool here to try. It is not venture-backed, it publishes its prices, and its free plan includes one holdout analysis and one MMM a year on data you upload. Paid plans run from $750 a month (Starter) to $3,000 (Professional, with unlimited holdouts and MMMs) and $6,000 (Scale), on **30-day terms** rather than annual contracts.  The method is careful for the price. A geo holdout turns ads off in some markets and compares them with closely matched ones; Stella then fits four candidate models (two synthetic control variants, a Bayesian structural time series and difference-in-differences) and keeps the one that best fitted the period before the test. An always-on module reads qualifying spend changes over 15 days without a holdout, and Stella itself calls that read directional. Holdout results anchor the in-house MMM. Tests typically run 20 to 28 days.  - **Strengths:** public prices and a free plan; monthly terms; several models compared for each experiment; an MMM anchored by holdouts; usable self-serve or run by Stella's team. - **Limitations:** the free and Starter plans run on uploaded data, with one test a year; aimed at brands spending $300K or more a month on paid media; radio, outdoor and store sales are not documented; the countries covered are not stated; a small, self-funded company.

### 4. [Cassandra](/cassandra.md)
Cassandra, founded in Milan in 2023, is the European entry in this guide and the one with the most automation around the test itself. It designs the geo experiment, launches it in your ad accounts within rules you set, and reads the result. An always-on module then checks every channel against real sales each day, and both feed the MMM and the attribution automatically.  Pricing is public and scales with annual media spend, **from €500 a month** for smaller budgets. At €5M of yearly spend the page shows Essentials from €2,500 a month with up to five experiments a year, Complete from €3,150 with ten and the always-on module, and Scale at €4,500, fully managed, with 25. The minimum commitment is six months. Customers include Gina Tricot, Velasca, North Sails and Home24.  - **Strengths:** experiments designed, launched and read for you; automatic calibration of MMM and attribution; public pricing in euros; a European customer base; a fully managed option. - **Limitations:** the number of experiments is capped by plan; Meta is the only channel named for automated execution; the product pages do not say whether tests use matched markets or synthetic control; six-month minimum commitment.

### 5. [Lifesight](/lifesight.md)
Lifesight sells incrementality testing as one module of a suite that also holds a causal MMM and causal attribution, under a single annual subscription. Its testing module documents more designs than any other here: matched markets, stepped-wedge (also called rolling thunder), central-control and synthetic control, plus split and time tests. Design is guided, and the company says no data scientist is required. Experiments can be deployed automatically to Meta and Google; other channels follow a manual guide. Typical tests run four to six weeks across 8 to 30 matched regions.  Calibration is the strong suit. A validated lift recalibrates the MMM curves and the attribution multipliers, and the model also accepts conversion lift studies run on the ad platforms. Mind the packaging: geo testing is in the entry Performance plan, but **the MMM that the tests calibrate starts at the Precision plan**.  - **Strengths:** widest documented set of test designs; automated deployment to Meta and Google; detailed calibration, platform lift studies included; public documentation; logos include Decathlon, IKEA and Reckitt. - **Limitations:** no public pricing; annual subscription; automated deployment limited to Meta and Google; MMM calibration requires the Precision plan; managed measurement is a paid add-on.

### 6. [Measured](/measured.md)
Measured has built its platform on experiments since 2017, and incrementality testing remains its centre. It runs geo-matched tests and first-party audience split tests, named synthetic control in its April 2026 release of a redesigned Geo Designer, and says it can run up to eight tests at once without contamination. The media mix model ingests those tests as causal priors, so the model starts from measured lift rather than from assumptions.  It suits brands that sell in several places at once: outcomes are measured across online, retail and marketplace sales, Amazon included, and the platform lists more than 300 managed integrations. Test design is guided and self-serve, with strategists available through a services team. Customers include The North Face, Steve Madden, Dermalogica and Overstock.  - **Strengths:** an experiment-first platform; geo and audience tests; several tests in parallel; tests feed the MMM as priors; omnichannel sales measurement. - **Limitations:** no public pricing and no stated minimum spend; whether strategist time is included in every contract is not stated; its own pages give different recommended test lengths, two to four weeks in one place and never under four in another.

### 7. [Polar Analytics Causal Lift](/polar-analytics.md)
Polar Analytics is a Paris and New York company best known for its e-commerce analytics. Causal Lift, launched in February 2025, is its testing product, and the model is done for you: each experiment is scoped, executed and interpreted by a Polar data scientist. The method is a geo experiment read against a synthetic control, a blend of holdout regions weighted to mirror the history of the test group. Covered channels are Google Ads, YouTube, Meta, TV and, in the United States, TikTok, with sales measured across Shopify, Amazon and physical stores.  Unusually, **the price is per test** and scales with revenue. For a brand under $5M in annual sales the page lists $3,200 for a first test and $4,000 for each one after; between $10M and $50M, $4,800 and $6,000. Polar has no MMM, so results inform decisions directly rather than calibrating a model.  - **Strengths:** a data scientist runs every test; pay per test; e-commerce connectors already in place; store and Amazon sales included. - **Limitations:** not self-serve; no MMM to calibrate; TikTok tests in the United States only; sold outside the core analytics plan; the pricing table is ambiguous on whether figures are per test or per month, so confirm before you sign.

### 8. [GeoLift by Recast](/recast.md)
GeoLift by Recast is the cheapest serious way into geo testing. It is free for six months with no credit card, then **$100 a month or $1,000 a year** per seat, and it is included for customers of Recast's MMM. You upload a file with dates, locations and a KPI; there are no connectors. The tool designs spend-up, spend-down or go-dark tests on anything you can target by geography, from TV by media market to retail by store, and Recast says a first design takes under 15 minutes.  The statistics come from the team behind the MMM: a fully Bayesian synthetic control, alongside a basic mode built on Meta's open-source package. The documentation is frank about limits. It advises against tests shorter than two weeks and warns that a spend-down test on a channel under $1M a year may struggle to detect anything. For MMM customers, every test is entered as evidence, significant or not.  - **Strengths:** a public, very low price with a long free period; self-serve; Bayesian synthetic control; honest documentation; a direct path into Recast's MMM. - **Limitations:** no data connectors, files are uploaded by hand; always-on testing and audience holdouts are not documented; estimates describe one point in time; the MMM itself has no public price.

### 9. [GeoLift (open source)](/meta-platforms-inc.md)
GeoLift is the R package released by Meta's marketing science team for geo experiments, and at least two commercial tools in this guide build on it or compare themselves to it. It covers the whole workflow: power analysis, market selection, and inference with augmented synthetic control. It needs R 4.0 or later, at least 25 periods of history before the test, more than 20 regions, and someone who can read the output. Meta labels it research software, not a Meta product, and itself recommends people-based tests where they are feasible.  Maintenance has slowed: the last tagged release dates from May 2023, with occasional fixes since.  - **Strengths:** free; a transparent, published method; full control over design and data; widely used as a baseline. - **Limitations:** R and statistical skills required; no support; slow maintenance since 2023; the data pipeline and the reporting are yours to build.

### 10. [Google Ads Conversion Lift](/google-ads.md)
Google's lift study comes in two forms. The user-based version splits people; the geography-based version splits regions and can read offline conversions aggregated by postal code or city, across Search, Shopping, Performance Max, Video, Demand Gen, Display and App campaigns. Reported metrics include incremental conversions, incremental value and incremental return on ad spend, with a confidence interval.  Access is the obstacle. Conversion Lift is not available to all accounts and usually goes through a Google representative. The geography-based version is a beta limited to holdback and go-dark designs within a single country, and Google notes that it tends to need more budget than a user-based study.  - **Strengths:** free; both user and geo designs; covers most Google campaign types; offline conversions supported in the geo version. - **Limitations:** not open to every account; measures Google only, by Google; the geo version is a beta, in a single country; a campaign can sit in only one study at a time.

### 11. [Meta Conversion Lift](/meta-platforms-inc.md)
Conversion Lift is the experiment built into Meta's ads platform. It splits people at random into a group eligible to see your ads and a group that is not, and reports the incremental conversions, whether purchases, sign-ups, app installs or offline sales. Randomising individuals gives more statistical power than a geo test, at no extra cost. As a guide, Meta asks for a campaign with at least $5,000 of spend and 500 conversions over the past year, and self-serve setup requires the Conversions API.  - **Strengths:** free; user-level randomisation; available self-serve in Experiments; usable as calibration input for an MMM, which Lifesight accepts for example. - **Limitations:** measures Meta only, by Meta; spend, conversion and data-quality thresholds; one self-serve path is unavailable to advertisers mainly targeting countries under the ePrivacy directive.

Source: https://www.softatlas.io/list/incrementality-testing