From 8fe077621f6d98e3f5486c850d56da91cca02857 Mon Sep 17 00:00:00 2001 From: juanitorduz Date: Sun, 8 Sep 2024 22:20:20 +0200 Subject: [PATCH] add more info about pymc marketing features --- docs/source/guide/mmm/mmm_intro.md | 55 ++++++++++++++++++++++++++++-- 1 file changed, 52 insertions(+), 3 deletions(-) diff --git a/docs/source/guide/mmm/mmm_intro.md b/docs/source/guide/mmm/mmm_intro.md index daf49dd6..299101aa 100644 --- a/docs/source/guide/mmm/mmm_intro.md +++ b/docs/source/guide/mmm/mmm_intro.md @@ -6,6 +6,26 @@ One approach might be to use heuristics, i.e. sensible rules of thumb, about wha Fortunately, with Bayesian modeling, we can do better than this! So-called Media Mix Modeling (MMM) can estimate how effective each advertising channel is in driving our outcome measure of interest, whether that is sales, new customer acquisitions, or any other key performance indicator (KPI). Once we have estimated each channel's effectiveness we can optimize our budget allocation to maximize our KPI. +## What is Media Mix Modeling? + +Media Mix Modeling is a statistical analysis technique used in marketing to evaluate the impact of various marketing activities on sales or other business outcomes. It's a data-driven approach that helps marketers understand which marketing channels are most effective, how they interact with each other, and how to optimize marketing spend across these channels. + +The "mix" in Media Mix Modeling refers to the combination of different marketing channels and tactics used by a company. These could include traditional media like TV, radio, and print, as well as digital channels like social media, search engine marketing, and email campaigns. + +## Why is Media Mix Modeling Important? + +In today's complex marketing landscape, businesses are investing in multiple channels simultaneously. Understanding the individual and combined effects of these channels is crucial for several reasons: + +1. **Efficiency**: MMM helps allocate marketing budgets more efficiently by identifying which channels provide the best return on investment (ROI). + +2. **Accountability**: It provides a way to measure and justify marketing spend to stakeholders. + +3. **Optimization**: By understanding the effectiveness of each channel, marketers can optimize their strategies for better results. + +4. **Forecasting**: MMM can be used to predict the potential impact of future marketing activities. + +5. **Holistic View**: It considers both online and offline marketing efforts, providing a comprehensive view of marketing effectiveness. + ## What can you do with Media Mix Modeling? Media Mix Modeling gives rich insights and is used in many ways, but here are some of the highlights: @@ -16,10 +36,11 @@ Media Mix Modeling gives rich insights and is used in many ways, but here are so 4. Optimize future marketing decisions. Rather than just inform future budget spending decisions, it is actually possible to optimize these spending decisions. For example, it is possible to calculate budgets across media channels that maximize our KPI for a given total budget. See this blog post on Bayesian decision-making for more information. 5. Inspire marketing experiments. If there is uncertainty about the effectiveness or saturation of channels, we can intelligently respond to this by running lift or incrementality tests to resolve some of this uncertainty. 6. Validate your understanding through predictions. We gain confidence in our knowledge of the world by making predictions and comparing them to what happens. MMM also generates forecasts that we can check against reality. As a result, we can improve our understanding and modeling iteratively to become more accurate over time. +7. Understand synergies between channels. MMM can help identify how different marketing channels interact with each other, revealing potential synergies or cannibalization effects. ![](bayesian_mmm_workflow2.png) -## How does Media Mix Modeling work? +## How does the classic Media Mix Modeling work? In simple terms, we can understand MMMs as regression modeling applied to business data. The goal is to estimate the impact of marketing activities and other drivers on a metric of interest, such as the number of new customers per week. @@ -45,6 +66,34 @@ Thus we can summarize the full MMM with this image: ![](bayesian_mmm.png) -## Putting it all together +## What can you do with PyMC-Marketing? + +PyMC-Marketing offers a comprehensive suite of features for Marketing Mix Modeling (MMM), designed to provide deep insights and powerful capabilities for marketing analytics: + +1. **Custom Priors and Likelihoods**: This feature allows users to incorporate domain-specific knowledge into their models through customizable prior distributions. By tailoring priors to reflect business-specific insights or expert opinions, users can create more accurate and relevant models. + +2. **Adstock Transformation**: The adstock feature optimizes the analysis of carry-over effects in marketing channels. This is crucial for understanding how the impact of marketing activities extends over time, helping to capture both immediate and delayed effects of campaigns. + +3. **Saturation Effects**: This functionality models the diminishing returns in media investments, a critical aspect of marketing strategy. It helps users understand at what point additional spending on a channel starts to yield less incremental benefit. + +4. **Customizable Adstock and Saturation Functions**: PyMC-Marketing provides flexibility in choosing and implementing adstock and saturation functions. Users can select from various pre-defined functions or implement their own custom functions to best fit their specific marketing scenarios. + +5. **Time-varying Intercept**: This advanced feature captures time-varying baseline contributions in the model using efficient Gaussian process approximation methods. It allows for modeling of changing baseline effects over time, which is essential for long-term analysis. + +6. **Time-varying Media Contribution**: Similar to the time-varying intercept, this feature models how media efficiency changes over time, again using Gaussian process approximation methods. This is particularly useful for understanding evolving media effectiveness. + +7. **Visualization and Model Diagnostics**: The package offers comprehensive tools for visualizing model performance and insights. This aids in interpreting results and communicating findings to stakeholders. + +8. **Multiple Inference Algorithms**: Users have the flexibility to choose from various NUTS (No-U-Turn Sampler) implementations, including BlackJax, NumPyro, and Nutpie. This allows for optimization of the inference process based on specific model requirements or computational preferences. + +9. **Out-of-sample Predictions**: The ability to forecast future marketing performance with credible intervals is a key feature. This supports scenario planning and simulations, enabling more informed decision-making. + +10. **Budget Optimization**: PyMC-Marketing includes tools for efficient allocation of marketing spend across various channels to maximize ROI. This feature is crucial for strategic budget planning and optimization. + +11. **Experiment Calibration**: The package supports fine-tuning models based on empirical experiments, allowing for a more unified and accurate view of marketing effectiveness. This includes capabilities for integrating lift test results and analyzing unobserved confounders and Return on Ad Spend (ROAS). + +These features collectively provide a robust framework for conducting sophisticated Marketing Mix Modeling, enabling marketers and analysts to gain deep insights into marketing effectiveness, optimize spending, and make data-driven decisions. + +## How to get started? -To see how all these different components come together, you can review the {ref}`MMM Example notebook `. +To see how all these different components come together, you can start from our {ref}`MMM Example notebook `.