Marketing mix modeling software estimates how much each of your marketing channels actually contributed to sales, from aggregate data like weekly spend and revenue, and tells you where the next dollar would do the most good. In October 2026 that covers four quite different kinds of product: self-serve SaaS platforms, platforms that come with a team attached, consultancies that build the model for you, and attribution tools that have bolted on something they call MMM. I’ve built marketing mix models for Monzo Bank, Booking.com, Time Out and a handful of DTC brands, and advised Google, Appsflyer and Recast on MMM, so I’ve seen this market from the builder’s side and the buyer’s side. (Since Recast appears below, consider that my disclosure: I have advised them, and I’ve tried to describe them exactly as plainly as everyone else.)
The vendor list has changed a lot in the last eighteen months. Nielsen’s MMM business now belongs to Circana, Rockerbox belongs to DoubleVerify, OptiMine belongs to Uptempo, and the experiment-first companies like Haus have all shipped their own MMMs. At the same time the open-source models from Google and Meta got good enough that a dozen agencies now sell Meridian as a service. If you haven’t looked since 2024, most of what you remember is out of date. If you’re new to the method itself, start with what marketing mix modeling is and come back.
I checked every vendor on this page against its own website, documentation, pricing page or press releases on 24 October 2026. Where a vendor’s public material doesn’t say something (whether its model is Bayesian, say), I’ve written “not stated” rather than guessing.
The four kinds of marketing mix modeling software
The word “software” hides the most important difference between these companies, which is how much of the work they do for you. A marketing mix model is part statistics and part judgment: somebody has to decide which controls go in, whether a channel’s result is believable, and what to do when the model and an experiment disagree. Every vendor makes a choice about who that somebody is.
SaaS platforms like Recast, Sellforte, Cassandra, Prescient AI, Keen, Stella and WorkMagic give you a hosted model that refreshes on a schedule, with dashboards, response curves and a budget optimizer. Most still have a customer-success or data-science team behind the scenes, and the self-serve ones vary a lot in how much hand-holding they offer. They suit a company that wants an always-on model without hiring a modeler, and can cope with a vendor’s opinions being baked into the product.
Platforms with a managed service (Measured, Haus, Lifesight, Paramark, MASS Analytics, Analytic Edge) put a named strategist or data scientist on your account, and price accordingly. In practice this is where most mid-market and enterprise buyers end up, because running the regression is the easy part of MMM and getting your finance team to believe the result is the hard part.
Consultancies (Analytic Partners, Ipsos MMA, Ekimetrics, Gain Theory, Circana, Kantar, Marketbridge, and the 52 agencies on Google’s Meridian partner list) build models as projects or ongoing engagements, usually for large advertisers with TV, retail and brand questions as well as digital ones. They’re the most expensive route and the most flexible, and the quality depends heavily on the team you get.
Attribution tools with an MMM feature (Northbeam, Triple Whale, Rockerbox, Fospha, Funnel, Adobe Mix Modeler) started by tracking users or moving data, and added a mix model later. Some of these are serious models and some are closer to a smoothed attribution report, so I’ve given them their own comparison page. The short version is that you should ask what the MMM is trained on, because a model trained on pixel data inherits the pixel’s blind spots.
The MMM vendor directory
Use the filters to narrow the list to what you’d actually consider, then tick two or three vendors to compare them side by side. “Runs experiments” means the vendor runs geo tests or holdouts inside its product or service, which matters because an MMM that can’t be checked against an experiment is an MMM you have to take on faith (there’s more on that in calibrating MMM with experiments).
Recast
SaaS platform (can be run by certified Operators)- Bayesian: Yes
- Experiments: Yes
- Sales-led
- Some pricing info
How it works: Proprietary Bayesian MMM with time-varying effects, refreshed regularly; geo tests use synthetic controls and their results feed back into the model.
Best fit: Consumer brands from scaled DTC to Fortune 500; names Rocket Money, Square, Canva and Hasbro.
Pricing: MMM is quote-only. Its separate GeoLift by Recast app is $100/month or $1,000/year after a six-month free trial, and included for MMM customers.
Worth knowing: Forecasting and budget optimizer, a reports library, public APIs and an MCP server for asking questions from Claude or ChatGPT.
Independent (Recast Labs, Inc.). $3.4M seed led by Lerer Hippeau, December 2022. Disclosure: Mike Taylor has advised Recast on MMM.
Measured
Platform with heavy service support- Bayesian: Not stated
- Experiments: Yes
- Sales-led
- Prices not published
How it works: Geo-matched and first-party audience-split tests, plus a causal MMM that uses the test results as priors. Since October 2025 you can choose which tests calibrate the model.
Best fit: Enterprise and omnichannel retail, CPG, financial services; names Dermalogica, Steve Madden and Mars.
Pricing: Not published; demo only.
Worth knowing: Media plan optimizer (June 2025), certified partners programme (March 2025), MCP server (May 2026).
Independent. $21M from Telescope Partners, May 2022.
Haus
Platform + measurement strategist- Bayesian: Not stated
- Experiments: Yes
- Sales-led
- Some pricing info
How it works: Experiments first: geo tests with synthetic control, time tests and fixed-geo tests. Causal MMM (generally available October 2025) is shaped by those test results.
Best fit: Shopify DTC up to enterprise, plus apps and betting; names Intuit, FanDuel, SharkNinja and Wayfair.
Pricing: Four tiers named on the pricing page (Basics, Core, Plus, Enterprise), no prices shown.
Worth knowing: Architect, an AI agent that recommends budget moves (debuted May 2026, live for Causal MMM September 2026). TV, OOH, podcast and influencer tests on higher tiers.
Independent, backed by Insight Partners; $20M led by 01 Advisors (July 2024).
Mutinex
SaaS platform- Bayesian: Not stated
- Experiments: Not stated
- Sales-led
- Prices not published
How it works: GrowthOS, which it calls a commercial mix model, with a data layer (DataOS) and an AI assistant (MAITE). Campaign-varying MMM added December 2025.
Best fit: Enterprise CMOs in telco, banking, beverages, retail and travel; names Optus, Samsung, Hershey's and ING.
Pricing: Not published. A self-serve GrowthOS On-Demand was in early access from May 2026.
Worth knowing: Publishes an open MMM validation framework. Signals (AI-visibility measurement) launched August 2026.
Independent, Australian, expanding in the US. A$17.5M led by Marbruck, October 2024.
Sellforte
SaaS platform- Bayesian: Yes
- Experiments: Yes
- Sales-led
- Prices published
How it works: Always-on "causal Bayesian" MMM at region level; geo lift, conversion lift and A/B tests enter as Bayesian priors; Bayesian synthetic control for geo tests.
Best fit: Mid-size to large retailers and ecommerce, strongest in Europe; names Lidl, bonprix, C&A and Douglas.
Pricing: Shown for a $300k/month media spend: Incrementality Testing $1,900/mo, Incremental Attribution $2,500/mo, Full-Funnel MMM $4,500/mo, Enterprise $8,500/mo (plus customizations).
Worth knowing: Consent-based pixel attribution at ad-set level, and Activate, which pushes recommendations into the ad platforms.
Independent (Finland). €4M round led by Sonae IM, March 2021.
Cassandra
Self-serve SaaS (managed tier available)- Bayesian: Not stated
- Experiments: Yes
- Self-serve start
- Prices published
How it works: MMM checked against live geo tests, with always-on incrementality readouts; the Essentials tier includes up to three models and five experiments a year.
Best fit: Small and mid-size brands and agencies (ecommerce, B2B, charities); names Gina Tricot, charity: water and Velasca.
Pricing: Essentials from 2,700/mo, Complete from 3,400/mo, Scale from 4,050/mo, Enterprise custom; prices rise with annual media spend, the currency isn't shown and a six-month minimum is listed.
Worth knowing: Has broadened into an "AI marketing platform" whose agents also run ads, creative and reports.
Independent (Milan). €2M seed led by Italian Founders Fund, March 2025.
Lifesight
Platform + optional managed measurement- Bayesian: Not stated
- Experiments: Yes
- Sales-led
- Some pricing info
How it works: Unified measurement: geo incrementality tests calibrate a causal MMM, which in turn calibrates attribution.
Best fit: Mid-market to enterprise retail, ecommerce, QSR, travel; logos include IKEA, KFC and New Balance.
Pricing: Tiers named (Performance, Precision, Enterprise); annual subscriptions scoped per business, no figures.
Worth knowing: Mia, an AI analyst agent (March 2026), and an MCP server.
Independent (Singapore).
Prescient AI
SaaS platform- Bayesian: Not stated
- Experiments: No
- Sales-led
- Some pricing info
How it works: Proprietary MMM retrained daily at campaign level, with halo effects across DTC, Amazon and retail. Validates against tests you run elsewhere rather than running them itself.
Best fit: DTC and omnichannel brands spending $100k+ a month on ads; names HexClad, Saatva and BrüMate.
Pricing: SMB, Growth and Enterprise tiers on annual contracts, priced by business size and sales channels; no figures.
Worth knowing: Retail attribution and "agnostic data ingestion" in a rebuilt model (July 2025).
Independent. $10M Series A co-led by Headline, March 2024.
Keen Decision Systems
SaaS platform- Bayesian: Yes
- Experiments: Not stated
- Self-serve start
- Prices not published
How it works: Bayesian, forward-looking MMM ("Marketing Elasticity Engine") with category benchmarks to give smaller brands priors.
Best fit: Mostly consumer packaged goods brands and their agencies; names Twinings, J.M. Smucker and Dramamine.
Pricing: Not published; free trial offered.
Worth knowing: Keen OS modules for Intelligence, Measure, Plan and Buy; a standalone planning module (September 2025).
Still independent: no acquisition found. $11M Series B led by Ballast Point Ventures, July 2022.
Analytic Partners
Consultancy + platform (GPS Enterprise)- Bayesian: Not stated
- Experiments: Yes
- Sales-led
- Prices not published
How it works: "Commercial analytics": MMM, attribution, experiment design, brand and pricing work, delivered by consultants on its GPS-E platform with an ROI Genome benchmark database.
Best fit: Global enterprises; names Beiersdorf, Kroger and VF Corp.
Pricing: Not published.
Worth knowing: Bought Magic Numbers (UK, 2024) and Analyx (Germany, 2025). Leader in the 2025 Gartner MMM Magic Quadrant.
Private; Onex Partners growth investment, 2022.
Circana (formerly Nielsen MMM)
Consultative or self-serve on Liquid Data- Bayesian: Not stated
- Experiments: Not stated
- Sales-led
- Prices not published
How it works: MMM built on store-level point-of-sale data with a large benchmark base; Marketing Foresight and Liquid Mix tools.
Best fit: Large CPG and retail advertisers who already buy point-of-sale data.
Pricing: Not published.
Worth knowing: Nielsen's MMM business became part of Circana when the deal closed on 21 August 2025; nielsen.com's MMM page now redirects there.
Owned by Circana (private).
Ipsos MMA
Consultancy + platform- Bayesian: Not stated
- Experiments: Yes
- Sales-led
- Prices not published
How it works: MMM combined with attribution ("unified marketing measurement") on its Activate platform, with in-market testing.
Best fit: Enterprise in auto, CPG, financial services, pharma and retail; names General Motors, Volvo Cars and McCormick.
Pricing: Not published.
Worth knowing: One of the longest-running MMM firms; Leader in the 2025 Gartner MMM Magic Quadrant.
Part of Ipsos (listed in Paris); Ipsos completed the purchase from Hunting Ridge in 2018.
Kantar (LIFT ROI)
Platform + research services- Bayesian: Not stated
- Experiments: Not stated
- Sales-led
- Prices not published
How it works: "AI-powered" MMM that brings brand strength and creative quality from Kantar's brand research into the model.
Best fit: Enterprise marketing and finance teams; names KFC, TSB Bank and Telenor.
Pricing: Not published.
Worth knowing: The pitch is the link to brand equity data, which most MMM vendors don't have.
Bain Capital-led ownership.
Ekimetrics
Consultancy + platform- Bayesian: Not stated
- Experiments: Not stated
- Sales-led
- Prices not published
How it works: A repeatable MMM core plus specialist modules, built with the client's data team rather than as a black box.
Best fit: Global enterprise in CPG, luxury and beauty, auto; names Nestlé, Estée Lauder and Renault.
Pricing: Not published.
Worth knowing: Certified Google Meridian partner; creative scores fed into MMM via a CreativeX partnership (2025).
Independent (founded in Paris).
Gain Theory
Consultancy + platform (HiFusion)- Bayesian: Not stated
- Experiments: Yes
- Sales-led
- Prices not published
How it works: MMM and unified measurement with incrementality testing; HiFusion measures several KPIs together and can sit alongside a client's own models.
Best fit: Large global brands.
Pricing: Not published.
Worth knowing: Named a Leader in the Forrester Wave for marketing measurement and optimization services, Q1 2026.
A WPP company.
MASS Analytics (MassTer)
Software with done-for-you to self-run tiers- Bayesian: Not stated
- Experiments: Not stated
- Sales-led
- Prices not published
How it works: MassTer software for building MMMs, with an AI agent (Maia) that builds and refreshes models, 150+ data connectors and a view that reconciles MMM, MTA and tests.
Best fit: CPG, retail and agencies; names Intel, Kellogg's and Monoprix.
Pricing: Not published; offers a fixed 60-day pilot.
Worth knowing: Runs an MMM Academy that trains and certifies in-house modelers, useful if you want to own the work eventually.
Independent.
Analytic Edge
Software + managed analytics- Bayesian: Not stated
- Experiments: Yes
- Sales-led
- Prices not published
How it works: Automated MMM ("Demand Drivers") plus pricing, forecasting and a test-and-learn tool (SynTest).
Best fit: CPG and consumer brands, strong in Asia-Pacific; logos include Unilever and Kimberly-Clark.
Pricing: Not published.
Worth knowing: Certified Google Meridian partner; TikTok measurement partner for MMM (April 2026).
Acquired by C5i, announced July 2024.
Marketbridge
Consultancy- Bayesian: Yes
- Experiments: Yes
- Sales-led
- Prices not published
How it works: Bayesian MMM combined with attribution, recalibrated frequently with tests; says the code is open and the client keeps it.
Best fit: B2B and technology companies, where long sales cycles break most consumer-brand MMM setups.
Pricing: Not published.
Worth knowing: One of few MMM shops built around B2B pipeline rather than ecommerce revenue.
Independent, backed by RTC Partners; acquired April Six (2025) and Revere (2025).
Paramark
Platform + growth advisor- Bayesian: Not stated
- Experiments: Yes
- Sales-led
- Prices published
How it works: MMM plus geo holdouts, audience splits and A/B tests, with forecasting and scenario planning; hierarchical models on the Advanced tier.
Best fit: Mid-market to enterprise consumer and B2B software and fintech; Greylock names Chime, Plaid and Square.
Pricing: Essentials $100k/yr (one MMM), Advanced $150k/yr (two MMMs, geo and hierarchical), Enterprise $220k+/yr; unlimited tests and users on every tier; startup pricing on request.
Worth knowing: The clearest published prices among the managed platforms, with a named human advisor working in your Slack.
Independent. $6M seed led by Greylock, May 2025.
Fospha
Daily MMM-style attribution- Bayesian: Partly
- Experiments: Yes
- Sales-led
- Some pricing info
How it works: "Media Mix Attribution": an always-on daily model at ad level that blends deterministic, correlative and causal pieces (including Bayesian saturation curves), with an MMM suite for planning and geo tests run by its science team.
Best fit: Ecommerce brands, especially fashion and beauty; names Huel, Gymshark and Dyson.
Pricing: Lite, Pro and Enterprise tiers per suite, no figures; built for brands spending over $150k a month.
Worth knowing: Halo measurement across DTC, Amazon and TikTok Shop.
A Blenheim Chalcot venture.
Northbeam
Attribution tool with MMM add-on- Bayesian: Not stated
- Experiments: Yes
- Self-serve start
- Prices published
How it works: First-party multi-touch attribution is the core. MMM+ is a daily media mix model that can take the MTA data as an input; Northbeam Incrementality (April 2026) runs automated geo lift tests, US Meta first.
Best fit: Shopify and DTC brands from seven figures up.
Pricing: Starter $1,500 (month-to-month, under $1.5M a year in ad spend); Professional $3,500/month; Growth and Enterprise custom. MMM+ and Incrementality are add-ons.
Worth knowing: Lift results feed back into the MTA and MMM+.
Independent.
Rockerbox
Attribution, MMM and testing suite- Bayesian: Not stated
- Experiments: Yes
- Sales-led
- Prices not published
How it works: De-duplicated user-level attribution, an MMM for long-run contribution and forecasts, and incrementality tests whose results calibrate both.
Best fit: Mid-market to enterprise omnichannel brands, including TV, podcast and direct mail spenders.
Pricing: Not published.
Worth knowing: Pairs with DoubleVerify's media-quality data since the acquisition.
Owned by DoubleVerify: announced February 2025 at $85M, closed March 2025.
Triple Whale
Shopify data platform with MMM in Compass- Bayesian: Partly
- Experiments: Yes
- Self-serve start
- Some pricing info
How it works: Pixel attribution plus Compass ("unified measurement"), where you choose daily, weekly or monthly MMM granularity and can use geo lift results and pixel attribution as priors. GeoLift tests use synthetic control.
Best fit: Shopify brands, from small stores on the free plan up.
Pricing: Free plan, then Foundation, Automate and Enterprise, priced by annual GMV with a slider; paid plans are 12-month contracts. Compass is included on Enterprise.
Worth knowing: Moby AI agents; geo tests and Meta Conversion Lift set-up inside the app on Enterprise.
Independent.
Funnel (Funnel Measure)
Data hub with MMM add-on- Bayesian: Not stated
- Experiments: Yes
- Self-serve start
- Some pricing info
How it works: MMM, attribution and incrementality tests combined by "triangulation"; MMM described as daily, at channel and campaign level, with adstock and saturation.
Best fit: Teams that already use Funnel to move marketing data, in retail, FMCG, travel and finance.
Pricing: Data hub from $300/month (Starter) or $600/month (Business), billed annually; Measure is an add-on on Business and Enterprise with no published price.
Worth knowing: If your data is already flowing through Funnel, the data-engineering part of the bill is mostly paid.
Independent (Sweden).
Adobe Mix Modeler
Enterprise SaaS (Adobe Experience Cloud)- Bayesian: Not stated
- Experiments: No
- Sales-led
- Prices not published
How it works: Ingests event and summary data into Adobe Experience Platform, harmonises it, and trains MMM and attribution models together, with optional priors from incrementality tests or other models.
Best fit: Enterprises already on Adobe Experience Platform.
Pricing: Not published; sold through Adobe.
Worth knowing: Its appeal is staying inside the Adobe stack; the tests themselves happen elsewhere.
Adobe product.
Google Meridian certified partners
Agencies and consultancies running open-source Meridian- Bayesian: Yes
- Experiments: Partly
- Sales-led
- Prices not published
How it works: Bayesian geo-level MMM on Google's Meridian, with experiment results as ROI priors; some partners also run Meridian GeoX geo tests.
Best fit: Anyone who wants an open, inspectable model but not the work of running it; 52 partners listed, from Adswerve, Tinuiti and Wpromote to Accenture, NIQ and Brainlabs.
Pricing: Set by each partner; Meridian itself is free (Apache 2.0).
Worth knowing: You keep a model you could take in-house later, because the code is public.
Google's list, filterable by region and by MMM or GeoX.
Arima
Data + modeling services- Bayesian: Not stated
- Experiments: Not stated
- Sales-led
- Prices not published
How it works: MMM built alongside its "Synthetic Society", a privacy-safe synthetic population dataset down to postal-code level.
Best fit: North American brands and agencies, particularly in Canada.
Pricing: Not published.
Worth knowing: Different angle from the rest: the data asset is the product, and the MMM sits on top.
Independent (Toronto).
Uptempo (OptiMine)
SaaS inside a marketing planning suite- Bayesian: Not stated
- Experiments: Yes
- Sales-led
- Some pricing info
How it works: OptiMine's MMM with what-if scenarios and saturation curves, plus a built-in media testing tool; uses no cookies or personal data.
Best fit: Enterprise marketers who plan budgets in Uptempo.
Pricing: Tiered by model granularity (channel-level up to full OptiMine models); no figures.
Worth knowing: OptiMine was acquired by Uptempo in August 2025; Uptempo Marketing Analytics, built on it, launched July 2026.
Owned by Uptempo.
Stella
Self-serve SaaS (consultancy option)- Bayesian: Not stated
- Experiments: Yes
- Self-serve start
- Prices published
How it works: Combines MTA, post-purchase surveys, regional holdouts and an MMM that the holdouts recalibrate, weighting each by strength of evidence.
Best fit: Consumer brands; pitches Starter at $5-10M a year in revenue and the full product at $300k+ a month in spend. Names PopSockets and Momofuku.
Pricing: Free (one MMM and one holdout a year on uploaded data); Starter $750/mo; Professional $3,000/mo (unlimited regional holdouts and MMMs); Scale $6,000/mo; 30-day terms.
Worth knowing: The cheapest published way to get an MMM and a holdout design from a vendor.
Independent.
WorkMagic
SaaS (Shopify app)- Bayesian: Not stated
- Experiments: Yes
- Self-serve start
- Prices not published
How it works: Automated geo tests on matched DMA or ZIP groups (typically three to four weeks), an incrementality-calibrated MMM and incrementality-adjusted attribution.
Best fit: Shopify and DTC brands; names Liquid I.V. and True Classic.
Pricing: Shopify app free to install; pricing custom, with a self-serve tier for brands under $10M GMV.
Worth knowing: Built to get a first test live within a week.
Independent.
No vendor matches every filter. Try removing one.
A couple of things the directory can’t show you. First, “not stated” on the Bayesian column is common among the big consultancies, and it doesn’t mean they aren’t Bayesian; it means they don’t say so on their site, so you should ask. Second, a few names you might expect are missing on purpose. Polar Analytics and Hyros are attribution and analytics tools without a shipped MMM product (Polar’s site says to talk to its team about scope), so they’re on the attribution comparison instead. INCRMNTAL measures incrementality without a classic MMM and lives on the incrementality testing tools page. I also looked for an MMM vendor called “Uptimize” and couldn’t find one; the name seems to be a mix-up of Uptempo and OptiMine.
What changed in the MMM vendor market in 2025 and 2026
The biggest shift is consolidation among the older names. Circana completed its purchase of Nielsen’s marketing mix modeling business on 21 August 2025, so if you have a “Nielsen MMM” contract, your vendor is now Circana (not to be confused with NIQ, the former NielsenIQ, which sells its own MMM). DoubleVerify bought Rockerbox in a deal announced in February 2025 at $85 million and closed in March. Uptempo bought OptiMine, announced in August 2025, and launched Uptempo Marketing Analytics on top of it in July 2026. C5i bought Analytic Edge in 2024, and Analytic Partners picked up Magic Numbers and Analyx. None of this means those products got worse, but it does mean the roadmap now answers to a parent company, which is worth asking about.
The second shift is that the experiment companies and the model companies have converged. Haus made its name on geo tests and launched Causal MMM in October 2025; Measured has always paired tests with a model and upgraded it the same month; Northbeam, an attribution company, launched automated geo lift tests in April 2026; and Recast, a modeling company, sells a standalone geo testing app. The pitch from nearly everyone is now the same loop (run experiments, use them to calibrate a model, use the model to plan), which is the right loop. The differences are in how seriously each one does each step.
The third shift is open source. Google’s Meridian now has 52 certified partners on its official list, from performance agencies like Tinuiti and Wpromote to Accenture and NIQ, and Google released Meridian GeoX for geo experiments in 2026. That makes “Meridian as a service from an agency” a real alternative to a proprietary SaaS model, with the advantage that the model is inspectable and you could take it in-house later. I’ve covered the libraries themselves on the open-source MMM page and Meridian separately.
Finally, almost every vendor shipped an AI agent or an MCP server (a connector that lets Claude or ChatGPT query your measurement data) between 2025 and 2026: Recast, Measured, Lifesight, Sellforte, Haus, Mutinex, Fospha and MASS Analytics all have one. These are useful for getting answers out of a model you already trust. They don’t change whether the model deserves that trust, which is the subject of the rest of this page.
Choosing marketing mix modeling tools by situation
If you’re a Shopify brand spending tens of thousands a month, mostly on Meta and Google, you probably don’t need an enterprise MMM yet. The cheapest credible starts are Stella’s free and $750-a-month tiers, Cassandra’s self-serve plans, or the MMM inside the attribution tool you may already pay for, as long as you treat it as a hypothesis and test the big calls. Prescient AI and Fospha both say they’re built for brands spending over $100,000 and $150,000 a month on ads respectively, so below that you’re not their customer.
For scaled DTC and omnichannel consumer brands (the eight-figure brands that sell on Amazon and in stores as well as their own site), the shortlist is usually Recast, Haus, Measured, Prescient AI, Sellforte, Lifesight, WorkMagic and Fospha. The question that splits them is whether you want experiments and the model from the same vendor. Haus and Measured lead with tests; Recast and Prescient lead with the model; Sellforte and Lifesight try to do both in one product. In my experience the first MMM a growing brand buys is mostly valuable for the arguments it starts, so pick the one whose outputs your finance lead will actually read.
Enterprise CPG and retail is a different market, where the outcome is often sales through retailers you don’t control and the data includes distribution and price. Circana, Analytic Partners, Ipsos MMA, Kantar, Ekimetrics, Gain Theory, MASS Analytics, Analytic Edge and Keen are the names here, along with the Meridian partners attached to the big holding companies. Ask about retail sales data and promotions before you ask about digital channels.
Subscription apps, fintech and B2B have their own quirks: conversion happens long after the ad, and the outcome you care about (a funded account, a qualified pipeline opportunity) is rarer than an ecommerce order. Paramark is built around software and fintech, Marketbridge around B2B, and INCRMNTAL around apps. Whoever you choose, make sure the model is fit on the outcome the business cares about, not on whatever is easiest to count, which is the same point I make in the MMM measurement plan.
And if you have a data scientist with time to spare, the honest answer may be no vendor at all. A Meridian or PyMC-Marketing model built in-house costs you salary and time rather than a license, and the build vs buy page has a calculator for comparing the two.
What the published prices tell you
Most MMM vendors don’t publish prices, which tells you something on its own: the price depends on your spend, your channels and how much help you need. The handful who do publish are worth knowing, because they anchor every other negotiation. These were the figures on each vendor’s own pricing page on 24 October 2026:
| Vendor | What’s published | Notes |
|---|---|---|
| Stella | Free; $750, $3,000 and $6,000 a month | 30-day terms. Regional holdouts and MMMs on connected data start at $3,000. |
| GeoLift by Recast | $100 a month or $1,000 a year | After a six-month free trial; geo testing only, included for Recast MMM customers. |
| Sellforte | $1,900 to $8,500 a month | Shown for $300k average monthly media spend; full-funnel MMM is $4,500. |
| Cassandra | From 2,700, 3,400 and 4,050 a month | Rises with annual media spend; currency not shown; six-month minimum listed. |
| Paramark | $100k, $150k and $220k+ a year | Billed annually; unlimited tests and users on every tier. |
| Northbeam | $1,500 (Starter) and $3,500 a month (Professional) | MMM+ and incrementality are add-ons on higher tiers. |
| Funnel | From $300 or $600 a month | That’s the data hub; the Measure (MMM) add-on isn’t priced publicly. |
The spread here, from free to over $200,000 a year, mostly reflects how much human time is included. A $750 plan is software that you operate yourself; a $150,000 plan includes bi-weekly reviews with a measurement expert. When you compare quotes, put the people cost on your side of the ledger too, because the cheap option often needs an analyst you’d otherwise not hire.
How to evaluate an MMM vendor
Every vendor demo looks good, because the demo data was chosen to look good. The way to evaluate one is to ask questions whose answers are hard to fake, ideally on your own data during a paid pilot. These are the ones I’d ask, grouped roughly in the order I’d ask them:
- Show me the model. Ask for the specification in writing: how adstock and saturation are modeled, which controls are included, what the priors are if it’s Bayesian. A vendor that can’t describe its model beyond “proprietary AI” is asking you to defend numbers to your CFO that you can’t explain. (Adstock and saturation curves cover what you’re listening for.)
- Show me the uncertainty. Every channel ROI should come with a range. If two channels’ intervals overlap heavily, the honest recommendation is “we can’t tell them apart yet”, and you want a vendor that says so.
- Show me out-of-sample accuracy. In-sample fit is easy to inflate. Ask how the model does on weeks it never saw, and how much the channel ROIs move when it’s refit with one more month of data. Stability matters more than an impressive R-squared; the model validation page explains why.
- Show me what happens with an experiment. Ask how a geo test result changes the model. The good answer involves priors or calibration; the weak answer is a slide next to the dashboard. Better still, ask whether the model has ever predicted a test result before the test ran.
- Show me correlated channels. Ask what the model does when two channels always run together (brand search and TV, for example). The answer tells you a lot about whether they understand multicollinearity or paper over it.
- Tell me who does the work. Find out who builds the data pipeline, who reviews each refresh, and how many accounts that person covers.
- Tell me what I keep. Can you export the inputs, outputs and, if it’s built on open source, the code, if you leave?
Red flags in an MMM vendor pitch
Some warning signs come up often enough that I’d treat any one of them as a reason to slow down. The first is a model that only reports point estimates. Marketing mix models are uncertain by nature (you’re estimating several channel effects from a hundred or so weekly observations), and a vendor that hides that uncertainty is no more accurate than one that shows it, only less honest about the error bars.
The second is the black box. “Our proprietary AI” is a fine answer to how a model is optimized, but not to what it assumes. If you can’t get an answer to whether price and promotions are in the model, you can’t know whether the TV result is really a promotion result (the confounding variables page shows how easily that happens). The third is no experiment calibration, or an experiments feature that’s purely decorative. A model that never meets ground truth will drift toward whatever story its priors or its training data tell.
The last two are about results that are too convenient. If every channel comes out profitable, be suspicious, because real budgets nearly always include something that isn’t pulling its weight. And if the MMM’s channel credit looks almost identical to the ad platforms’ own reporting, the model may be learning from that reporting rather than from sales. Neither is proof of a bad model, but each deserves a pointed follow-up question before you sign anything.
If you’d rather understand the method well enough to ask these questions without notes, the Vexpower MMM course walks through building, reviewing and challenging a model on an invented brand, Northstar. Knowing how a model can go wrong is the most useful thing you can bring to a vendor meeting.
