MMM Studio

Step three · before it runs

What do you expect each channel to return? We’ll tell you if that’s even possible.

Every model starts from a belief, whether the tool admits it or not. Yours is on this screen where you can see it. Drag a dial and watch the number at the side: that is how much of what you actually sold your beliefs are claiming media produced. If it exceeds what you sold, the beliefs are wrong before any modelling starts.

Channel · total spend

$0Revenue back per $1 spent$12

You are claiming

TV — Broadcast
$10.6M
you think $1.22 · could plausibly be anywhere from $0.28 to $5.37
$12.9M
of revenue
fairly sure · change this, or attach a test you have run
How sure are you?
This sets the width of the pale band, nothing else. The model will move any of these if your data disagrees — a wide band just lets it move further.
Let the data decide how quickly this channel tires outOnly worth ticking where you have spent at a lot of different levels.
If you have actually tested this channel
Switched it off in some regions, or ran a lift study? Put the result here and the model starts from what you measured instead of what you assumed. The dial closes, because typing over a measurement throws it away.
Extra sales it madeExtra money it costHow precise that was
Paid Social
$7.3M
you think $1.22 · could plausibly be anywhere from $0.28 to $5.37
$8.9M
fairly sure · adjust
How sure are you?
This sets the width of the pale band, nothing else. The model will move any of these if your data disagrees — a wide band just lets it move further.
Let the data decide how quickly this channel tires outOnly worth ticking where you have spent at a lot of different levels.
If you have actually tested this channel
Switched it off in some regions, or ran a lift study? Put the result here and the model starts from what you measured instead of what you assumed. The dial closes, because typing over a measurement throws it away.
Extra sales it madeExtra money it costHow precise that was
Paid Search
$6.3M
you think $1.22 · could plausibly be anywhere from $0.28 to $5.37
$7.7M
fairly sure · adjust
How sure are you?
This sets the width of the pale band, nothing else. The model will move any of these if your data disagrees — a wide band just lets it move further.
Let the data decide how quickly this channel tires outOnly worth ticking where you have spent at a lot of different levels.
If you have actually tested this channel
Switched it off in some regions, or ran a lift study? Put the result here and the model starts from what you measured instead of what you assumed. The dial closes, because typing over a measurement throws it away.
Extra sales it madeExtra money it costHow precise that was
Programmatic Display
$4.1M
you think $1.22 · could plausibly be anywhere from $0.28 to $5.37
$5.0M
fairly sure · adjust
How sure are you?
This sets the width of the pale band, nothing else. The model will move any of these if your data disagrees — a wide band just lets it move further.
Let the data decide how quickly this channel tires outOnly worth ticking where you have spent at a lot of different levels.
If you have actually tested this channel
Switched it off in some regions, or ran a lift study? Put the result here and the model starts from what you measured instead of what you assumed. The dial closes, because typing over a measurement throws it away.
Extra sales it madeExtra money it costHow precise that was
Affiliate
$2.2M
you think $1.22 · could plausibly be anywhere from $0.28 to $5.37
$2.7M
fairly sure · adjust
How sure are you?
This sets the width of the pale band, nothing else. The model will move any of these if your data disagrees — a wide band just lets it move further.
Let the data decide how quickly this channel tires outOnly worth ticking where you have spent at a lot of different levels.
If you have actually tested this channel
Switched it off in some regions, or ran a lift study? Put the result here and the model starts from what you measured instead of what you assumed. The dial closes, because typing over a measurement throws it away.
Extra sales it madeExtra money it costHow precise that was
Retail Media
$1.4M
you think $1.22 · could plausibly be anywhere from $0.28 to $5.37
$1.7M
fairly sure · adjust
How sure are you?
This sets the width of the pale band, nothing else. The model will move any of these if your data disagrees — a wide band just lets it move further.
Let the data decide how quickly this channel tires outOnly worth ticking where you have spent at a lot of different levels.
If you have actually tested this channel
Switched it off in some regions, or ran a lift study? Put the result here and the model starts from what you measured instead of what you assumed. The dial closes, because typing over a measurement throws it away.
Extra sales it madeExtra money it costHow precise that was

Each dial sets a belief, not a fact — the pale band behind it is how far you are willing to be wrong. Wide band means “convince me”; narrow band means “I already know this”. The model will move any of these if the data disagrees, and after the run the evidence screen shows you exactly which ones it moved.

So this is what you are saying

Add up your six claims against what you actually sold over the same period.

8% of revenue from media. Plausible.

General guidance · not from your data

A business spending 6.7% of revenue on media typically finds somewhere between 10% and 35% of sales traceable to it. That range is published experience, not a measurement of this dataset — you are below it, which is a modest set of beliefs rather than a wrong one.

With your uncertainty carried through, the engine puts that share between 5% and 24% across 4,000 draws.

Why this screen exists at all

This is the cheapest disagreement in the whole product. It costs two minutes here and a quarter of credibility later. It is also the only screen where your own number appears next to the model’s, which is what makes the evidence screen legible: “the data moved you” only means something if you were somewhere first.

How it runs, if you want to change it

Weeks to hidekept back as a testName this runoptional

Starting pointsmore = a stricter checkEffort per attemptmore = slower, steadier

Afterwards, try to break the answerAdds a few minutes and tells you which parts you can rely on.

Quick preview

about 30 seconds

A rough answer, fast, so you can see whether your dials are in the right area. It cuts corners in one specific way: it always sounds more certain than it should. Use it to steer, never to decide.

If you have run this exact thing before, you get that run back instead of waiting for it again. Nothing here is charged twice.