加入新推出的
Discord 社区，展开实时讨论，获得同行支持，并直接与 Meridian 团队互动！
Meridian 简介
使用集合让一切井井有条
根据您的偏好保存内容并对其进行分类。
营销组合建模分析 (MMM) 是一种统计分析法，用于衡量营销广告系列和活动的影响，从而指导预算规划决策并提高整体媒体效果。MMM 使用汇总数据来衡量在各个营销渠道的影响，同时还会考虑影响收入和其他关键绩效指标 (KPI) 的非营销因素。MMM 注重隐私保护，不会使用任何 Cookie 或用户级信息。
Meridian 是一个 MMM 框架，广告客户可以借助此框架建立和运行自己的内部模型。Meridian 可帮助您解答以下关键问题：
- 营销渠道是如何帮助实现收入或其他 KPI 的？
- 我的营销投资回报率 (ROI1) 是多少？
- 如何优化未来的营销预算分配？
Meridian 是一个基于贝叶斯因果推理而又高度可自定义的建模框架。它能够处理大规模的地理位置级数据（建议在有此类数据的情况下使用），但也可以用于国家级建模。Meridian 可提供清晰的数据洞见和可视化图表，以便为有关营销预算和规划的业务决策提供依据。此外，Meridian 还提供各种方法，支持利用实验和其他先验信息对 MMM 进行校准，并利用覆盖面和频次数据优化广告的目标频次。
主要功能
Meridian 通过提供建模和优化方法，可满足所有主要 MMM 应用场景的需要。如需详细了解 Meridian 方法，请参阅模型规范和“Meridian 模型”部分。
此外，其主要功能包括：
分层地理位置级建模：借助 Meridian 的分层地理位置级模型，您可以利用地理位置级营销数据；相比国家级数据，地理位置级营销数据可能包含更丰富的营销效果信息。此外，您还可以检查本地或区域一级的营销工作成效。分层方法通常会在投资回报率等指标上产生更严格的可信区间。如需了解详情，请参阅“Geo-level Bayesian Hierarchical Media Mix Modeling”（地理位置级贝叶斯分层媒体组合建模）。
Meridian 利用 Tensorflow Probability 及其 XLA 编译器，完全支持使用 50 多个地理位置和 2-3 年每周数据的贝叶斯模型。使用 Google Colab Pro+ 或其他工具提供的 GPU 硬件能进一步优化速度性能。
如果您没有地理位置级数据，系统也支持标准的国家级方法。
整合有关媒体效果的先验知识：借助 Meridian 的贝叶斯模型，您可以使用投资回报率先验来整合有关媒体效果的现有知识。在此模型中，投资回报率是一个模型形参，可以采用任何先验分布，无需进行额外计算即可将先验投资回报率信息转化为模型形参。知识可以来自任何可用来源，例如过往实验、过往 MMM 结果、行业专业知识或行业基准数据。
贝叶斯方法非常灵活，因为您可以控制先验对后验分布的影响程度。如果当前数据中的信号较弱，可以使用先验来估计形参。Meridian 会量化所有模型形参、投资回报率和边际投资回报率的不确定性。如需了解详情，请参阅“Media Mix Model Calibration With Bayesian Priors”（使用贝叶斯先验进行媒体组合模型校准）。
考虑媒体饱和度与滞后效应：付费媒体和自然媒体的饱和度与滞后效应使用形参转换函数建模。饱和度使用 Hill 函数建模，该函数可捕获边际回报递减的情况。滞后效应使用具有几何衰减的 Adstock 函数建模。Meridian 利用贝叶斯马尔可夫链蒙特卡洛 (MCMC) 抽样方法来联合估计所有模型形参，包括这些转换形参。如需了解详情，请参阅“Bayesian Methods for Media Mix Modeling with Carryover and Shape Effects”（利用延滞效应和形状效应进行媒体组合建模的贝叶斯方法）。
可选择使用覆盖面和频次数据，获取更多数据洞见：除了使用展示次数之外，Meridian 还提供将覆盖面和频次数据作为模型输入的选项，可为用户提供更多数据洞见。覆盖面是指每个时间段内的唯一身份浏览者数量，而频次是指向每位浏览者展示广告的平均次数。这样可以更好地预测每个媒体渠道在支出发生变化时可能有何表现。如需了解详情，请参阅整合了覆盖面和频次数据的贝叶斯分层媒体组合模型。
对漏斗下端渠道（例如付费搜索）进行建模：Meridian 的设计基于因果推理理论，旨在支持理性决策。有效因果推理所需的模型假设完全透明。具体而言，Meridian 提供了在衡量付费搜索效果时将 Google 搜索查询量 (GQV) 用作控制变量的选项。
媒体预算优化：优化阶段会根据您的总预算确定各个渠道的最佳预算分配。Meridian 还可以根据您的广告目标推荐最佳总预算。此外，Meridian 还会为任何具有覆盖面和频次数据的渠道提供频次优化。
利用假设情景进行估计：借助拟合后的模型，您可以估计不同假设媒体情景下的投资回报率，如增加或减少特定渠道的广告支出，或重新分配各渠道的预算。
评估和报告模型拟合优度：Meridian 可报告样本内和样本外的模型拟合统计信息。您可以借此来比较不同的模型配置，例如先验分布和形参化。
可选择包含非媒体处理变量：可选择包含非媒体处理（例如价格和促销活动的变化），以估算非媒体营销行动的效果。
如未另行说明，那么本页面中的内容已根据知识共享署名 4.0 许可获得了许可，并且代码示例已根据 Apache 2.0 许可获得了许可。有关详情，请参阅 Google 开发者网站政策。Java 是 Oracle 和/或其关联公司的注册商标。
最后更新时间 (UTC)：2025-08-04。
[null,null,["最后更新时间 (UTC)：2025-08-04。"],[[["\u003cp\u003eMarketing mix modeling (MMM) uses aggregated data to measure marketing campaign impact across channels, informing budget planning and improving media effectiveness while maintaining user privacy.\u003c/p\u003e\n"],["\u003cp\u003eMeridian is an open-source MMM framework that enables advertisers to build and run their own in-house models to understand marketing ROI, channel performance, and budget optimization.\u003c/p\u003e\n"],["\u003cp\u003eMeridian utilizes Bayesian causal inference, handles large-scale geo-level data, incorporates prior knowledge about media performance, and accounts for media saturation and lagged effects for accurate insights.\u003c/p\u003e\n"],["\u003cp\u003eThis framework offers advanced features including reach and frequency data integration, lower-funnel channel modeling, media budget optimization, and what-if scenario estimations to support comprehensive marketing analysis.\u003c/p\u003e\n"],["\u003cp\u003eMeridian facilitates robust model evaluation and reporting, including goodness of fit statistics and the optional inclusion of non-media treatment variables for a holistic understanding of marketing performance.\u003c/p\u003e\n"]]],[],null,["# About Meridian\n\nMarketing mix modeling (MMM) is a statistical analysis technique that measures\nthe impact of marketing campaigns and activities to guide budget planning\ndecisions and improve overall media effectiveness. MMM uses aggregated data to\nmeasure impact across marketing channels and account for non-marketing factors\nthat impact revenue and other key performance indicators (KPIs). MMM is\nprivacy-safe and does not use any cookie or user-level information.\n\nMeridian is an MMM framework that enables advertisers to set up and run\ntheir own in-house models. Meridian helps you answer key questions such\nas:\n\n- How did the marketing channels drive my revenue or other KPI?\n- What was my marketing return on investment (ROI^[1](#fn1)^)?\n- How do I optimize my marketing budget allocation for the future?\n\nMeridian is a highly customizable modeling framework that is based on\n[Bayesian causal inference](/meridian/docs/basics/rationale-for-causal-inference-and-bayesian-modeling). It is\ncapable of handling large scale geo-level data, which is encouraged if\navailable, but it can also be used for national-level modeling. Meridian\nprovides clear insights and visualizations to inform business decisions around\nmarketing budget and planning. Additionally, Meridian provides\nmethodologies to support calibration of MMM with experiments and other prior\ninformation, and to optimize target ad frequency by utilizing reach and\nfrequency data.\n\nKey features\n------------\n\nMeridian supports all major MMM use cases by providing modeling and\noptimization methodologies. For more information about Meridian\nmethodologies, see [Model specification](/meridian/docs/basics/model-spec) and\n*The Meridian model* section.\n\nAdditionally, the key features include:\n\n- **Hierarchical geo-level modeling:** Meridian's hierarchical\n geo-level model lets you make use of geo-level marketing data, which\n potentially contains much more information about your marketing\n effectiveness than national-level data. Additionally, you can examine the\n effectiveness of marketing efforts at a local or regional level. The\n hierarchical approach often yields tighter credible intervals on metrics\n such as ROI. For more information, see [Geo-level Bayesian Hierarchical\n Media Mix\n Modeling](https://research.google/pubs/geo-level-bayesian-hierarchical-media-mix-modeling/).\n\n Meridian supports fully Bayesian models with 50+ geos and 2-3 years\n of weekly data utilizing [Tensorflow\n Probability](https://www.tensorflow.org/probability/overview) and its [XLA\n compiler](https://www.tensorflow.org/xla). GPU hardware, available using\n Google Colab Pro+ or other tools, can further optimize speed performance.\n\n The standard national level approach is supported if you don't have\n geo-level data available.\n- **Incorporating prior knowledge about media performance:**\n Meridian's Bayesian model lets you incorporate existing knowledge\n about your media performance through the use of ROI priors. In this model,\n ROI is a model parameter which can take any prior distribution---no additional\n calculations are needed to translate prior ROI information to the model\n parameters. Knowledge can be derived from any available source such as past\n experiments, past MMM results, industry expertise, or industry benchmarks.\n\n The Bayesian method is flexible because you can control the degree to which\n priors influence the posterior distribution. Priors can be used to estimate\n a parameter when the signal in the current data is weak. Meridian\n quantifies uncertainty for all model parameters, ROI, and marginal ROI. For\n more information, see [Media Mix Model Calibration With Bayesian\n Priors](https://research.google/pubs/media-mix-model-calibration-with-bayesian-priors/).\n | **Note:** If you don't have experiment priors and want to explore an open source option to get this data, you can try GeoX. GeoX experiments help address the typical technical issues encountered in analyzing randomized paired geo experiments. For more information about GeoX, see the [google/trimmed_match](https://github.com/google/trimmed_match) and [google/matched markets](https://github.com/google/matched_markets) repositories in GitHub.\n- **Accounting for media saturation and lagged effects:** Saturation and\n lagged effects for paid and organic media are modeled using parametric\n transformation functions. Saturation is modeled using a Hill function, which\n captures diminishing marginal returns. Lagged effects are modeled using an\n adstock function with geometric decay. Meridian utilizes Bayesian\n Markov Chain Monte Carlo (MCMC) sampling methods to jointly estimate all\n model parameters, including these transformation parameters. For more\n information, see [Bayesian Methods for Media Mix Modeling with Carryover and\n Shape\n Effects](https://research.google/pubs/bayesian-methods-for-media-mix-modeling-with-carryover-and-shape-effects/).\n\n- **Optional use of reach and frequency data for additional insights:** In\n addition to using impressions, Meridian provides the option to use\n reach and frequency data as model inputs to provide additional insights.\n Reach is the number of unique viewers within each time period, and frequency\n is the corresponding average number of impressions per viewer. This provides\n a better prediction of how each media channel might perform with a change in\n spending. For more information, see [Bayesian Hierarchical Media Mix Model\n Incorporating Reach and Frequency\n Data](https://research.google/pubs/bayesian-hierarchical-media-mix-model-incorporating-reach-and-frequency-data/).\n\n- **Modeling lower funnel channels (such as paid search):** Meridian\n is designed based on causal inference theory to support rational\n decision-making efforts. Model assumptions required for valid causal\n inference are made fully transparent. Specifically, Meridian\n provides an option to use Google Query Volume (GQV) as a control variable\n when measuring the impact of paid search.\n\n- **Media budget optimization:** The optimization phase determines the optimal\n budget allocation across channels based on your overall budget. There is\n also an option for Meridian to suggest the optimal overall budget\n based on your advertising goals. Additionally, Meridian provides\n frequency optimization for any channel with reach and frequency data.\n\n- **Estimation using what-if scenarios:** With your fitted model, you can\n estimate what your ROI would have been under different hypothetical media\n scenarios, such as increasing or decreasing advertising spending on a\n specific channel or re-allocating budget across channels.\n\n- **Evaluating and reporting model goodness of fit:** Meridian reports\n model fit statistics, both within-sample and out-of-sample. You can use this\n to compare different model configurations, such as prior distributions and\n parameterizations.\n\n- **Optional inclusion of non-media treatment variables:** Non-media\n treatments, such as changes to price and promotions, can optionally be\n included to estimate the effectiveness of non-media marketing actions.\n\n*** ** * ** ***\n\n1. \"ROI\" and \"iROAS\" are being used synonymously throughout the documents, both denoting the measurement of the incremental return on investment. [↩](#fnref1)"]]
