Definition
The Bayesian Brain is Zennith Agency's proprietary Bayesian Marketing Mix Model (MMM). It uses posterior probability distributions to calculate, with 98.1% accuracy, which marketing channels are most likely to exceed a client's target return before any budget is allocated.
Unlike platform-reported ROAS, which measures what has already happened, the Bayesian Brain predicts what will happen. This distinction is fundamental: platform-reported ROAS is a lagging indicator. The Bayesian Brain is a leading indicator.
The Problem
What Google tells you
Google Ads: 4.2x ROAS
Every platform reports its own attribution. Every platform inflates its own numbers. You are making budget decisions based on data that every channel has a financial incentive to distort.
What the model tells you
P(exceeding target) = 0.94
The Bayesian model produces a posterior distribution of expected returns for each channel: a probability-weighted estimate of what each pound is actually driving, independent of platform reporting.
The 98.1% Accuracy Claim
The 98.1% accuracy figure refers to the model's ability to correctly predict, before spend begins, whether a given channel allocation will exceed the client's target return on investment. This is measured by comparing the model's pre-campaign posterior probability distribution against the actual campaign outcome across Zennith's client portfolio.
Specifically: when the Bayesian Brain assigns a channel a probability greater than 80% of exceeding target, the channel exceeds target in 98.1% of cases. This is not a guarantee. It is a posterior probability, which means it is the best estimate given all available evidence. The model's confidence intervals are always communicated alongside its point estimates.
How It Works
01
Data Ingestion
The model ingests all available historical channel performance data: spend, impressions, clicks, conversions, and revenue by channel and time period. The richer the data, the tighter the posterior distributions.
02
Prior Distribution Setting
Based on industry benchmarks and the client's historical data, the model sets prior probability distributions for each channel's expected return. These priors encode what we know before seeing the client's specific data.
03
Bayesian Updating
As new data arrives, the model updates its probability distributions using Bayes' theorem, producing posterior distributions that reflect all available evidence. The model learns continuously: every campaign improves the next prediction.
04
Posterior Distribution Output
The model outputs a posterior distribution for each channel: not a single number, but a full probability curve showing the range of likely returns and the probability of exceeding the client's target.
05
Budget Recommendation
Budget is allocated to channels in proportion to their posterior probability of exceeding target. The client receives a recommendation expressed as: 'Channel X has a 94% probability of exceeding your target return at this spend level.'
Bayesian MMM vs. Platform-Reported ROAS
| Aspect | Platform-Reported ROAS | Bayesian Brain |
|---|---|---|
| What it measures | What has already happened (lagging indicator) | What will happen (leading indicator) |
| Output format | Single point estimate (e.g., 3.2x ROAS) | Posterior distribution (e.g., 94% probability of exceeding 3x ROAS) |
| Uncertainty handling | Ignored or averaged away | Explicitly modelled and communicated |
| Data requirement | Large historical dataset required | Works with limited data; improves as data grows |
| Budget decision basis | Past performance | Probability of future performance |
| Attribution model | Last-click or linear | Probabilistic, accounts for interaction effects |
What We Build
Bayesian Marketing Mix Model, built on your actual revenue and spend data
Posterior distribution reporting: probability of exceeding target by channel
Budget reallocation recommendations with documented confidence intervals
Stochastic optimisation: real-time budget adjustment under constraints
Channel attribution beyond platform-reported ROAS
Phase-Gate accountability: before and after revenue attribution audit
Frequently Asked
What is Bayesian Marketing Mix Modeling?
Bayesian Marketing Mix Modeling (Bayesian MMM) is a statistical framework that uses Bayesian inference to estimate the contribution of each marketing channel to business outcomes. Unlike platform-reported attribution (which every platform inflates in its own favour), Bayesian MMM produces a posterior distribution of expected returns for each channel: a probability-weighted estimate of what each pound or dollar is actually driving. The model updates as new data arrives, becoming more accurate over time.
How is this different from Google Analytics attribution?
Google Analytics attribution assigns credit to the last click, the first click, or a linear distribution across touchpoints, all of which are arbitrary rules, not statistical models. Bayesian MMM models the actual causal relationship between marketing spend and revenue using regression analysis, time-series data, and prior knowledge about channel behaviour. It accounts for factors that platform attribution ignores: seasonality, brand equity, competitor activity, and the lag between spend and conversion.
What does 98.1% accuracy mean in practice?
The 98.1% accuracy figure refers to the predictive accuracy of Bayesian models in forecasting customer behaviour across the channels in Zennith's client dataset. In practice, this means the model's predictions of which channels will exceed target ROAS are correct 98.1% of the time, compared to the industry average of roughly 60 to 70% for rule-based attribution models. This accuracy is what allows Zennith to make budget reallocation recommendations with documented confidence intervals rather than intuition.
What is a posterior distribution in marketing terms?
A posterior distribution is the model's output after combining prior knowledge (what we know about how this channel typically performs) with observed data (what we have actually measured). In marketing terms, it means the model gives each channel a probability of exceeding its target: not a single number, but a range with confidence levels. A channel with a posterior distribution showing 94% probability of exceeding target ROAS is a fundamentally different investment decision than one showing 45%.
Stop Guessing
Find out what your marketing spend is actually driving.
A free AI Visibility Audit includes a Bayesian Brain baseline: which of your current channels the model predicts will exceed target, and which are underperforming relative to their probability.