Google updates Meridian MMM with pricing variables and new priors

Google announced significant updates to Meridian, its open-source Marketing Mix Modeling tool, on September 30, 2025. The enhancements enable marketers to measure the impact of non-media variables and obtain more accurate return on investment calculations through improved modeling capabilities.

Meridian now supports the inclusion of non-media variables such as pricing and promotions to measure their impact on sales more precisely. The update introduces channel-level contribution priors, which allow users to guide the MMM with their own business knowledge. These priors help translate domain expertise into more actionable insights.

The modeling framework has been enhanced with binomial adstock decay functions that can measure the longer-term effects of upper-funnel media. These functions account for brand recall and influence that may drive purchases weeks after initial exposure. When the Hill function is concave—for example when its slope parameter equals one—channels without reach and frequency data will always have a higher overall ROI than their marginal ROI.

Marginal ROI (mROI) based priors have been added as an alternative to ROI priors for paid media channels. The mROI of a channel is defined as the expected return on one additional monetary unit of spend. The additional monetary unit is allocated across geographic regions and time periods by scaling up the reach while holding the average frequency fixed.

The choice between ROI and mROI priors carries important implications, particularly for creating prior parity across channels. Both ROI and mROI have a prior distribution. If the ROI prior is specified, then an mROI prior is induced. Conversely, if the mROI prior is specified, then an ROI prior is induced. An induced prior does not belong to a parametric family, and it is typically not independent of other model parameters.

A common ROI prior can be used for all channels to create prior ROI parity. As the prior strength increases—meaning the standard deviation decreases—posterior ROI distributions will shrink toward a common value. Channel-specific ROI priors can incorporate prior knowledge, such as experiment results. Although ROI priors don't control optimization budget shifts as effectively as mROI priors, optimization spend constraints can limit the amount of budget shift proposed for any given channel.

Alternatively, a common mROI can be used for all channels to create prior mROI parity. As the prior strength increases, posterior mROI values will shrink toward a common value. Prior mROI parity generally results in less significant optimization budget shifts. If the same mROI prior is used for all channels, then the prior optimal budget allocation will equal historical spending patterns.

For reach and frequency channels, the marginal ROI by reach equals the ROI. This occurs because the marginal ROI prior is applied to the marginal ROI by reach, where the next monetary unit spent increases the reach without changing the average frequency. Under the Meridian model specification, media effects are linear in reach. Therefore, the choice between an ROI and a marginal ROI prior parameterization has no impact on the prior for reach and frequency channels.

The adstock decay specification parameter now allows users to define which decay function is used for each channel. Meridian provides two decay curves: geometric and binomial. The rate at which media effects taper off is governed by the choice of function along with the learned parameter alpha.

Geometric decay is parameterized where the weight decreases exponentially with lag. Binomial decay uses a different functional form that can persist longer before decaying. The binomial curve is convex if alpha is less than 0.5, linear if alpha equals 0.5, and concave if alpha is greater than 0.5. Its x-intercept is always at max_lag plus one.

Selecting binomial decay makes sense when a channel has a significant proportion of effects in the latter half of the effects window. Otherwise, geometric decay is recommended. The decay curve impacts the relative weights of lagged media. Increasing the relative weight of later time periods necessarily decreases the relative weight of earlier time periods.

Meridian recommends using binomial decay with max_lag values of four to 20 time periods, while geometric decay works well with two to 10 time periods. The binomial decay curve defines weights that decay to zero more gradually than the geometric curve. Therefore, the binomial decay curve encourages a larger proportion of a channel's total media effect to happen in later time periods.

When outcome is not revenue and revenue_per_kpi is not passed to InputData, users can set custom priors using three approaches: a custom total paid media contribution prior, a custom incremental KPI per cost prior, or channel-level contribution priors.

For the total paid media contribution approach, users can set a common ROI prior on all channels such that the total media contribution specific prior mean and standard deviation align with intuition about the total proportion of KPI that is incremental due to paid media. For channel-level contribution priors, users can set media_prior_type='contribution' and rf_prior_type='contribution' in the ModelSpec and customize the prior distributions.

Google expanded the Meridian partner network to 30 certified global partners who can support implementation. The partners include firms such as Publicis Media, dentsu, Monks, Adswerve, KINESSO, Accenture, NIQ, and others across various geographic regions including the Americas, APAC, Europe, and Latin America.

A Discord community was launched to provide real-time discussions, peer support, and direct interaction with the Meridian team. The community offers resources where users can ask questions and share implementation experiences.

The updates address measurement challenges for campaigns where ROI clarity is increasingly critical. By incorporating non-media variables like pricing and promotions, marketers can isolate the true impact of media spending from other business factors affecting sales.

Why this matters

The Meridian updates arrive as marketers face growing pressure to demonstrate clear ROI from advertising investments. Traditional attribution models often struggle to account for the complex interplay between media exposure and eventual conversions, particularly for upper-funnel campaigns where effects manifest over weeks rather than days.

Marketing mix modeling has gained traction as privacy changes limit deterministic attribution. With third-party cookies being phased out and platform tracking becoming more restricted, MMM provides an aggregate approach to measuring marketing effectiveness without relying on individual user tracking.

The addition of non-media variables addresses a longstanding limitation in MMM implementations. Price promotions and changes can significantly impact sales, yet many models attribute these effects to concurrent media campaigns. By explicitly modeling these variables, marketers can avoid over-crediting media for sales lifts actually driven by promotional pricing.

Channel-level contribution priors represent a methodological advancement that bridges the gap between statistical modeling and business intuition. Marketing teams often possess qualitative knowledge about relative channel performance based on years of campaign management. These priors allow that expertise to inform the model without dictating results, creating a balance between data-driven insights and human judgment.

The enhanced binomial adstock decay functions tackle the challenge of measuring brand-building activities. Upper-funnel campaigns often aim to build awareness and consideration that manifests in conversions long after initial exposure. Traditional models with faster decay rates may underestimate these longer-term effects, potentially leading to underinvestment in brand-building channels.

Marginal ROI priors address budget allocation questions more directly than traditional ROI metrics. When planning budget changes, the relevant question is not average historical return but rather the expected return on the next dollar spent. Channels operating at saturation may show strong average ROI but poor marginal ROI, while underinvested channels might deliver higher returns on incremental spend.

The expansion to 30 certified partners lowers implementation barriers for organizations lacking in-house data science capabilities. MMM traditionally required significant statistical expertise, limiting adoption primarily to large advertisers with dedicated analytics teams. Certified partners provide implementation support that makes the methodology accessible to mid-market advertisers.

Google's investment in Discord community resources signals recognition that open-source tool adoption depends on robust support ecosystems. While the codebase is freely available on GitHub, successful implementation requires guidance on model specification, prior selection, and result interpretation.

The timing coincides with broader industry movement toward incrementality testing and causal measurement. Platforms including Meta and TikTok have invested in conversion lift studies and geo-experiments. MMM complements these approaches by providing continuous measurement rather than periodic tests, though at the cost of more assumptions about causal relationships.

For advertisers using multiple measurement approaches, Meridian can serve as a fact-checker against multi-touch attribution. Attribution models provide directional signals for day-to-day optimization but may suffer from systematic biases. MMM offers an alternative perspective that can validate or challenge attribution-based conclusions about channel effectiveness.

The open-source nature differentiates Meridian from commercial MMM vendors who typically operate as black boxes. Transparency in model specification and estimation allows users to understand exactly how conclusions are derived and customize the approach for their specific business context.

Timeline

• September 30, 2025: Google announces Meridian updates including non-media variables, channel-level contribution priors, and enhanced binomial adstock decay functions
• September 2025: Meridian version 1.2.1 released with updates to output_model_results_summary and additional helper functions
• July 2025: Meridian version 1.1.7 adds organic RF support and optimization improvements
• June 2025: Version 1.1.2 introduces mROI priors and channel constraints parameters
• May 2025: Discord community launched for Meridian users

Summary

Who: Google, Meridian users, certified implementation partners, marketing teams, data scientists

What: Updates to Meridian open-source Marketing Mix Model adding non-media variables (pricing, promotions), channel-level contribution priors, binomial adstock decay functions, marginal ROI-based priors, and 30 certified global partners

When: Announced September 30, 2025

Where: Available globally through GitHub, with certified partners across Americas, APAC, Europe, Latin America, and Middle East

Why: To help marketers make more accurate budget decisions by measuring ROI more precisely, accounting for non-media factors, capturing longer-term brand effects, and providing better optimization recommendations based on marginal rather than average returns