The Platform
find out what your media mix actually caused — and which parts of it are moving the needle.
marginal combines bayesian marketing mix modelling with incrementality testing so you can answer the two questions your current measurement stack can't: what did our spend cause, and how did our channels work together to produce that outcome.
bayesian marketing mix modelling
see your full media mix in context — not channel by channel in isolation.
most measurement tools look at channels one at a time. marginal runs bayesian mmm across your entire media mix simultaneously, so you see contribution in context — paid search, paid social, display, audio, video, brand, and above-the-line spend, all in the same model.
the output is a probabilistic distribution of contribution, not a single number dressed up as certainty. you get a range, a confidence level, and a decomposition that reflects how your channels are actually structured — not how your attribution tool wishes they were.
we use google meridian v1.5.3 as the model engine. we chose it because it's open-source, academically rigorous, and has no commercial incentive to tell you any particular answer.
incrementality testing
run controlled experiments that tell you what your spend actually caused.
mmm tells you contribution. incrementality testing tells you causation. the two are different questions, and you need both.
marginal designs and runs geo-holdout tests and conversion lift tests — controlled experiments that withhold spend from a defined group, measure the difference in outcome, and give you a causal estimate of what that channel or campaign actually drove.
this is the test your board can believe. not a model estimate, not a platform attribution figure — a controlled experiment with a treatment group and a control group.
the calibration loop
incrementality results feed back into the model. the model compounds over time.
the two methodologies aren't separate reports. incrementality test results are used to calibrate the mmm — anchoring the model's priors to real experimental evidence. each test makes the next model run more accurate. each model run surfaces where to run the next test.
over time you build a measurement system that learns. not a static snapshot, not a quarterly report — a compounding loop that gets more useful the longer you run it.
independent by design
built for brands. no platform affiliation. no commercial incentive to favour any channel.
every major attribution tool is owned by or deeply integrated with an ad platform. they have a structural incentive to report strong performance for the channels that feed them revenue.
marginal has no such incentive. we don't run media. we don't sell ad inventory. we don't have commercial relationships with the platforms whose performance we measure. our only commercial relationship is with you — the brand trying to get an honest answer.
works with your stack
ready to find out what your media mix is actually doing?
the first three model runs are free. no contract, no commitment — just a rigorous answer to the question your current measurement stack can't give you.
start your pilot — first 3 runs free