Haus incrementality testing means running geo experiments through Haus’s platform: it turns advertising up or down in some regions, holds others steady, and uses synthetic control to estimate how much extra revenue the ads caused. Measured incrementality works on the same idea with a different design (matched markets plus audience holdouts), and both now feed those results into their own marketing mix models. However, they’re two of about twenty ways to run an incrementality test in 2026, and several of the alternatives are free.
My agency spent over $10 million a year on Facebook ads, and at that scale the question that matters is some version of “would these sales have happened anyway?” That’s what an incrementality test answers, by creating a comparison group that didn’t see the ads. The tools below differ in who designs the test, what unit gets randomized (users or regions), which channels they can handle, and what they cost. I checked each one against its own site, docs or repository on 24 October 2026. One disclosure before we start: I’ve advised Recast, which makes one of the tools here.
The three ways to run an incrementality test
Every tool on this page belongs to one of three routes, and choosing the route matters more than choosing the brand. The first is a platform lift study, where Meta or Google randomly withholds your ads from a slice of users and compares conversions. These are the most statistically powerful tests for a single online channel, because randomizing individual users gives you millions of units instead of a few dozen regions. However, they only measure that platform, they only count conversions that platform can see, and the platform is grading its own homework. The conversion lift studies page covers how to read one.
The second route is a DIY geo test with open-source software. You pick regions, change spend in some of them, and use a package like Meta’s GeoLift or Google’s Meridian GeoX to build a counterfactual from the rest. This works for any channel, including TV and outdoor, and it measures total sales wherever they happen. It’s free, but someone on your team has to own the design, the power analysis and the readout, which is where most DIY tests fail.
The third route is a vendor, which ranges from a $100-a-month self-serve app to an enterprise platform with a measurement strategist. You pay for market selection, power analysis, test operations and, increasingly, a model that keeps learning from every test. At larger budgets this is usually money well spent, because a test that moves 10 percent of a seven-figure budget is worth far more than the fee.
How Haus incrementality testing works
Haus runs geo experiments as its core product, with three test types: standard geo tests, time tests and fixed geo tests. Its blog explains why it uses synthetic control (a weighted blend of untreated regions built to match the test regions before the test) rather than matched markets, and claims the approach is more precise. On the higher tiers it covers TV, out-of-home, podcast and influencer tests as well as digital channels, and it reads the result on total sales, including Shopify and Amazon on its entry tier.
Since October 2025 Haus has also sold a Causal MMM, which uses those experiment results to shape the model’s response curves, and in 2026 it added Architect, an AI agent that recommends budget moves from the model. Its pricing page lists four tiers (Basics, Core, Plus and Enterprise) with no prices; Basics is self-serve US geo testing for Shopify brands, and Enterprise adds a dedicated data scientist. The customer logos skew towards well-known consumer brands (Intuit, FanDuel, SharkNinja, Wayfair), and the most recent round it has announced on its own blog is $20 million led by 01 Advisors, in July 2024.
Haus vs Measured
Measured is the other name most people shortlist, and the two have converged on the same loop: test, calibrate a model, plan. The differences are in emphasis. Measured describes its tests as “geo-matched and first-party split tests”, which means matched-market geo tests plus audience holdouts using your own customer lists, and it pairs them with a causal MMM that treats the test results as priors (since October 2025 you can choose which tests calibrate it). It sells mostly to enterprise and omnichannel retail, CPG and financial services, and doesn’t publish prices.
If I were choosing between them, I’d ask each to design the same test on my data and compare the minimum detectable effect, the number of regions they’d hold out and the spend at stake. The vendor whose design detects a smaller lift for the same cost is giving you a better test, regardless of whose method sounds more sophisticated. I’d also ask both what they do when a test result disagrees with their own model, because how a vendor handles that disagreement is the whole value of the loop.
Incrementality testing tools side by side
The table covers the tools that run or analyze incrementality tests, as of 24 October 2026. “Published price” means a figure on the vendor’s own pricing page; most enterprise tools don’t publish one.
| Tool | Route | Method | Best fit | Published price or minimum |
|---|---|---|---|---|
| Haus | Vendor | Geo tests with synthetic control; time and fixed-geo tests; Causal MMM | Shopify DTC to enterprise, apps, betting | Tiers named, no prices |
| Measured | Vendor | Matched-market geo tests and first-party audience holdouts; causal MMM | Enterprise, omnichannel retail, CPG | Not published |
| GeoLift by Recast | Self-serve app | Synthetic control with power analysis for market selection | Brands that want to run their own geo tests | $100/month or $1,000/year after a six-month free trial |
| Lifesight | Vendor | Geo holdout tests calibrating an MMM and attribution | Mid-market to enterprise retail and DTC | Tiers named, no prices |
| Stella | Self-serve / consultancy | Regional holdouts recalibrating an MMM, plus MTA and surveys | DTC and consumer brands | Free; $750, $3,000 (regional holdouts) and $6,000 a month |
| WorkMagic | Vendor (Shopify app) | Geo-matched DMA or ZIP groups, three to four weeks typical | Shopify DTC | Custom; self-serve tier under $10M GMV |
| INCRMNTAL | Vendor | No holdouts: causal inference on natural changes in spend | Mobile apps and gaming, now web and TV | Tiers named, license not tied to spend |
| Sellforte | Vendor | Geo lift, conversion lift and A/B tests as Bayesian priors | European retail and ecommerce | Incrementality testing $1,900/month at $300k spend |
| Cassandra | Self-serve SaaS | Geo tests checking its MMM | Small and mid-size brands | From 2,700/month (up to five tests a year) |
| Paramark | Vendor | Geo holdouts, audience splits, A/B, with MMM | Software, fintech, B2B | From $100k/year, unlimited tests |
| Northbeam Incrementality | Attribution add-on | Automated geo lift tests, US Meta first (April 2026) | Existing Northbeam customers | Add-on, not published |
| Triple Whale GeoLift | Attribution add-on | Synthetic control on census-based “Geo Pods”, 21 days minimum | Triple Whale Enterprise customers | Enterprise plan |
| Polar Causal Lift | Analytics add-on | Geo holdout with synthetic control and clustered regions | Shopify brands on Polar | $5,000 (bundled) or $20,000 (standalone) per test for a $50M brand |
| Meta Conversion Lift | Platform | Randomised user holdout on Meta | Meta advertisers with enough conversions | No fee; access via a Meta representative |
| Google Conversion Lift | Platform | User-based or geo-based holdout | Google Ads advertisers | No fee; 1,000 conversions and $5,000 campaign budget (user-based) |
| GeoLift (Meta) | Open source (R) | Augmented synthetic control, power analysis | Teams with R skills | Free (MIT) |
| Meridian GeoX (Google) | Open source | Time-based regression, stratified sampling, multi-cell designs | Teams on Meridian or wanting MMM calibration | Free |
| CausalPy (PyMC Labs) | Open source (Python) | Bayesian synthetic control, geo lift, difference-in-differences | Python teams, PyMC-Marketing users | Free (Apache 2.0) |
| CausalImpact (Google) | Open source (R; Python port) | Bayesian structural time series | One test region against a set of controls | Free |
| trimmed_match / matched_markets (Google) | Open source (Python) | Randomised paired geo experiments; matched-market design | Researchers; little recent maintenance | Free |
A few notes on the open-source rows. Meta’s GeoLift has no tagged release since v2.6.06 in May 2023, but the code is still being updated (its DESCRIPTION file reads 2.7.5) and you install it from GitHub. Google’s trimmed_match hasn’t had a commit since June 2023, so I’d treat it as reference code. CausalImpact is on CRAN at version 1.4.1, and the community Python port tfcausalimpact released 0.0.19 in September 2026. The geo lift testing page and synthetic control page explain the methods underneath.
Which incrementality testing platform fits your situation
If you spend most of your budget on Meta or Google and sell online, start with the platform’s own lift study. It’s free beyond the ad spend, it has the most statistical power, and it will tell you quickly whether your attribution numbers are wildly off. Google publishes its eligibility rules (at least 1,000 observed conversions and a $5,000 minimum campaign budget for user-based studies); Meta’s developer documentation still routes Conversion Lift through your Meta representative. Treat the result as a lower-risk first read rather than the final word, because neither platform can see sales that happen in a store or on Amazon.
If you sell through stores, Amazon or retailers, or you want to test TV, audio or outdoor, you need a geo test, because a geo test measures total sales in a region regardless of where the purchase happened. With an analyst who writes R or Python and regional sales data, GeoLift, Meridian GeoX or CausalPy will do the job for free. With a spreadsheet analyst, GeoLift by Recast or Stella’s regional holdouts give you market selection and synthetic control without code. Without anyone who has the time, buy the service.
If you want to test several channels a quarter and feed every result into a marketing mix model, that’s when a platform like Haus, Measured, Lifesight, Sellforte or Paramark earns its fee, because the hard part becomes scheduling tests so they don’t contaminate each other and keeping the model honest. Several MMM vendors now include tests in the package, which I’ve noted in the MMM software directory. And if you’re a mobile app that can’t easily run geo holdouts, INCRMNTAL’s no-holdout approach is designed for exactly that constraint, with the trade-off that it’s inferring causality from natural variation rather than a controlled experiment.
What to check before you trust any test result
The tools differ, but the ways a test goes wrong are the same everywhere, and a vendor dashboard can hide them as easily as a spreadsheet. I’d want to see each of these before acting on a result, whoever ran it:
The last item on that list is where incrementality tests pay off over time. A single test tells you about one channel in one period, but a test that becomes a prior in your model improves every decision the model makes afterwards, which is the subject of calibrating MMM with experiments. If you want to practise designing and reading tests before running a real one, the Vexpower MMM course includes a geo experiment on its invented Northstar brand, and the incrementality guide covers the counterfactual thinking underneath all of these tools. For the bigger picture of where testing fits alongside modeling, start with what marketing mix modeling is.
