Honest numbers come with a range.

A single number hides how sure the model is. Odins reports "most likely 12%, probably between 10 and 15", so you know when to act and when to test first.

Book a demo Why it matters

TV contribution to sales

last 12 months

"TV drove 10 to 15% of sales, most likely 12%" tells you more than "TV drove 12%".

The approach

Start with what you know. Let evidence move it.

Bayesian modeling treats your team's knowledge as the starting point and every month of data as an update.

Priors: documented assumptions about each channel, set with your team in 2 to 3 sessions
Evidence: each month of data confirms some assumptions and corrects others
Posterior: what the model believes now, with the uncertainty stated
Over time the data takes over: after about a year, the priors have largely faded

Why not a point estimate

A number without a range is a guess.

Two models can both say TV drove 12%. One has evidence, one is guessing. The range is what tells them apart.

Wide range: the data can't resolve it yet, so we propose a structured test
Tight range: reallocate with confidence and defend it to the board
Built for single-market data: useful ranges without dozens of regions to compare

Channel contribution, 80% ranges

from the posterior

Search 14 to 16% optimize

Social 9 to 13% optimize

Display 2 to 9% test first

A tight range is a decision. A wide range is a test plan, with cost boundaries.

Where the ranges show up.

Every part of the platform reports uncertainty the same way.

Scenario planning

Every forecast comes as a range, so you see the risk before committing.

See scenario planning

Recommendations

Uncertain channels get structured tests, not guesses.

See recommendations

Reporting and insights

Contributions and marginal returns, with intervals.

See reporting

The model

Everything the model estimates, and how to inspect it.

See the methodology

Bayesian is table stakes

Most modern MMM is Bayesian, including the open-source frameworks. The difference is the discipline in how the priors get set.

How we set priors

FAQ

Frequently asked questions.

What is a prior?

A starting assumption about a channel, written down and agreed with your team. For example: search saturates early, TV works with a lag.

Do we need a statistician to set them?

No. It is a guided conversation about your own marketing, typically 2 to 3 sessions. We translate it into the model.

What if our assumptions are wrong?

The data has the final say. Priors set the starting point; the evidence moves the model wherever it disagrees.

Is Bayesian modeling new?

The theory is decades old. Modern computing made it practical, which is why most serious MMM now runs on it.

Why not classical regression?

With roughly 150 weekly observations and ten or more channels, a blank-slate regression overfits. Priors give the model structure the data alone can't.

Can we see the assumptions?

Yes. Every prior is documented and approved by your team before the model runs, and you can revisit them at any time.

What does an 80% range mean in practice?

Eight times out of ten, the outcome should land inside it. Narrow means act; wide means test first. Either way, you know before you spend.