Just like any other attribution solution, Alvie regression-based attribution (RBA) aims to measure the impact of various marketing activities on a business KPI, such as revenue or the number of conversions.
RBA is a modern approach to attribution that utilizes advanced Bayesian methodology, similar to Google's meridian, which was released in March 2024. It leverages recent advancements in probabilistic programming (PyMC), robust data infrastructure, and the automation of marketing expert knowledge and intuition.
Core concepts of RBA
Regression analysis: Regression is a statistical method used to determine the relationship between a dependent variable (such as a business KPI) and one or more independent variables (such as advertising spend on marketing channels). By modeling this relationship, regression helps predict the value of the dependent variable based on the values of the independent variables.
Saturation and Carryover effects: Saturation effects occur when additional increments in advertising spending result in diminishing returns, while carryover effects reflect the prolonged influence of marketing activities on a business KPI. By incorporating these dynamics into regression models through transformations of the advertising spend data, we can address some inherent limitations of raw linear regression, such as the assumption of a constant and immediate impact of advertising spend.
Bayesian framework: Bayesian inference allows us to estimate the parameters of a regression model, including those for saturation and carryover effects. It involves starting with initial guesses (prior distributions), such as expected carryover period and ROAS based on incrementality tests, and refining these guesses as new data becomes available (likelihood). This approach combines previous experience and new evidence, resulting in a range of probable values (posterior distribution) for each parameter. This helps build a more accurate and reliable model. Read more about Bayesian Inference here.
Main benefits of RBA
Not dependent on active BigQuery export: RBA does not rely on user paths and can directly retrieve conversion and revenue data from the Google Analytics 4 API. This direct access streamlines data integration and simplifies the onboarding process.
Sophisticated use of prior knowledge: By incorporating automated marketing expertise and intuition as priors within the model, RBA allows the inclusion of test results to improve model performance.
Handling complex relationships: Bayesian regression's adaptability enables it to effectively model complex interactions, such as the interplay between different marketing strategies or how their impacts evolve over time.
Understanding uncertainty: Bayesian methods provide a spectrum of potential outcomes, rather than just a single-point estimate. This comprehensive view of uncertainty helps in making more nuanced and informed decisions.
Independence from platform data manipulation: The model operates on reliable data sources, such as advertising spend and selected KPIs, independent of any data manipulation by advertising platforms. However, it currently relies on Google Analytics 4 for information on unpaid channels and derives current targets from Google Analytics 4 data (revenue and number of conversions).
Future-proof and privacy-friendly: The model's independence from tracking individual user paths enhances its compliance with privacy regulations, making it a more sustainable and privacy-conscious choice.
Limitations of RBA
Targets: Currently, the targets are limited to data available from the Google Analytics 4 API, specifically the number of conversions and revenue. However, the team is actively working on expanding the range of available targets.