About Meridian

Marketing mix modeling (MMM) is a statistical analysis technique that measures the impact of marketing campaigns and activities to guide budget planning decisions and improve overall media effectiveness. MMM uses aggregated data to measure impact across marketing channels and account for non-marketing factors that impact revenue and other key performance indicators (KPIs). MMM is privacy-safe and does not use any cookie or user-level information.

Meridian is an MMM framework that enables advertisers to set up and run their own in-house models. Meridian helps you answer key questions such as:

  • How did the marketing channels drive my revenue or other KPI?
  • What was my marketing return on investment (ROI1)?
  • How do I optimize my marketing budget allocation for the future?

Meridian is a highly customizable modeling framework that is based on Bayesian causal inference. It is capable of handling large scale geo-level data, which is encouraged if available, but it can also be used for national-level modeling. Meridian provides clear insights and visualizations to inform business decisions around marketing budget and planning. Additionally, Meridian provides methodologies to support calibration of MMM with experiments and other prior information, and to optimize target ad frequency by utilizing reach and frequency data.

Key features

Meridian supports all major MMM use cases by providing modeling and optimization methodologies. For more information about Meridian methodologies, see Model specification and The Meridian model section.

Additionally, the key features include:

  • Hierarchical geo-level modeling: Meridian's hierarchical geo-level model lets you make use of geo-level marketing data, which potentially contains much more information about your marketing effectiveness than national-level data. Additionally, you can examine the effectiveness of marketing efforts at a local or regional level. The hierarchical approach often yields tighter credible intervals on metrics such as ROI. For more information, see Geo-level Bayesian Hierarchical Media Mix Modeling.

    Meridian supports fully Bayesian models with 50+ geos and 2-3 years of weekly data utilizing Tensorflow Probability and its XLA compiler. GPU hardware, available using Google Colab Pro+ or other tools, can further optimize speed performance.

    The standard national level approach is supported if you don't have geo-level data available.

  • Incorporating prior knowledge about media performance: Meridian's Bayesian model lets you incorporate existing knowledge about your media performance through the use of ROI priors. In this model, ROI is a model parameter which can take any prior distribution—no additional calculations are needed to translate prior ROI information to the model parameters. Knowledge can be derived from any available source such as past experiments, past MMM results, industry expertise, or industry benchmarks.

    The Bayesian method is flexible because you can control the degree to which priors influence the posterior distribution. Priors can be used to estimate a parameter when the signal in the current data is weak. Meridian quantifies uncertainty for all model parameters, ROI, and marginal ROI. For more information, see Media Mix Model Calibration With Bayesian Priors.

  • Accounting for media saturation and lagged effects: Saturation and lagged effects for paid and organic media are modeled using parametric transformation functions. Saturation is modeled using a Hill function, which captures diminishing marginal returns. Lagged effects are modeled using an adstock function with geometric decay. Meridian utilizes Bayesian Markov Chain Monte Carlo (MCMC) sampling methods to jointly estimate all model parameters, including these transformation parameters. For more information, see Bayesian Methods for Media Mix Modeling with Carryover and Shape Effects.

  • Optional use of reach and frequency data for additional insights: In addition to using impressions, Meridian provides the option to use reach and frequency data as model inputs to provide additional insights. Reach is the number of unique viewers within each time period, and frequency is the corresponding average number of impressions per viewer. This provides a better prediction of how each media channel might perform with a change in spending. For more information, see Bayesian Hierarchical Media Mix Model Incorporating Reach and Frequency Data.

  • Modeling lower funnel channels (such as paid search): Meridian is designed based on causal inference theory to support rational decision-making efforts. Model assumptions required for valid causal inference are made fully transparent. Specifically, Meridian provides an option to use Google Query Volume (GQV) as a control variable when measuring the impact of paid search.

  • Media budget optimization: The optimization phase determines the optimal budget allocation across channels based on your overall budget. There is also an option for Meridian to suggest the optimal overall budget based on your advertising goals. Additionally, Meridian provides frequency optimization for any channel with reach and frequency data.

  • Estimation using what-if scenarios: With your fitted model, you can estimate what your ROI would have been under different hypothetical media scenarios, such as increasing or decreasing advertising spending on a specific channel or re-allocating budget across channels.

  • Evaluating and reporting model goodness of fit: Meridian reports model fit statistics, both within-sample and out-of-sample. You can use this to compare different model configurations, such as prior distributions and parameterizations.

  • Optional inclusion of non-media treatment variables: Non-media treatments, such as changes to price and promotions, can optionally be included to estimate the effectiveness of non-media marketing actions.

  1. "ROI" and "iROAS" are being used synonymously throughout the documents, both denoting the measurement of the incremental return on investment. ↩

Next
Rationale for causal inference and Bayesian modeling