Marketing Mix Modeling, explained with cannons on a beach

By: Roman Myskin - Oct. 13, 2026

Picture a beach lined with cannons, all firing out to sea at once. You're standing on the shore with a single device: it measures the total wave height at any given moment — the combined splash of every cannonball landing, all mixed together. You never see one wave per cannon. You see the sum.

Cannons on a beach firing out to sea

Each cannon is loaded with gunpowder, and each crew is doing its job well — they're all trying to push their cannonball as far out as they can. In marketing terms, every channel is already operating close to its own ideal ROAS. Nobody's crew is being lazy. But past a certain amount of powder, more doesn't mean farther — it just burns off as waste. Distance isn't only about effort, it's about hitting the efficient amount of powder for that specific cannon.

Here's the part that matters most: you're not standing at the cannons. You're standing far down the beach, and your position is the marker. The farther out a cannonball lands from you, the better that cannon's result — because a cannonball's true efficiency isn't just its raw distance, it's the combination of how far it landed from your position and how much weight it was carrying. A heavy cannonball that barely clears the marker isn't impressive. A light cannonball that sails past you, for the same powder, is.

You never get to watch an individual cannonball fly. But you're not powerless. Every day, you write down two things: the total wave height that day, and how much powder each cannon burned through that day. Do that for long enough — across enough days, with the crews varying their powder loads for their own reasons — and you can start untangling the mixed signal. Which cannon's powder is actually moving your marker-distance, and which cannon is just adding noise to the wave?

That's Marketing Mix Modeling, at its core:

Nothing on this beach is a single number

Here's where the picture needs one more layer, and it's an important one: nothing about this beach is fixed. No cannon fires the exact same distance every time, even with the exact same amount of powder.

Wind shifts. Sand shifts. A crew that's usually excellent has an off day. So if you plotted every shot from a single cannon on a given powder load, you wouldn't get one dot — you'd get a cloud of dots, clustered around some typical distance, with some shots landing a bit short and some a bit long. That cloud is a distribution.

This matters because it changes the question you're actually able to answer. You can't say "this channel produces exactly $40,000 of revenue for $10,000 of spend." What you can honestly say is "this channel's contribution is most likely somewhere in this range, and here's how confident we are about that range." Some cannons you've watched for years and know very well — their cloud of outcomes is tight. Others are newer or noisier — their cloud is wide, and the honest answer includes more uncertainty.

A good model doesn't pretend this uncertainty away. It reports the range of plausible efficiency for each cannon, not a single confident number. That's not a weakness of the method — it's the model being honest about what a mixed, noisy signal can and can't tell you.

Some waves take a while to reach you

One more wrinkle in the picture, and it's one of the most important for how these models actually get used: not every cannon is sitting right at the water's edge.

Some cannons are close to shore. Fire one of those, and the wave hits your measuring device almost immediately — you see the effect of that powder today. But other cannons are out on distant rocks, or firing from far down the coastline. Their cannonballs land, but the wave they create takes time to travel — it might reach you tomorrow, or spread out and arrive in smaller ripples across the whole week.

This is carry-over (adstock): a channel's effect on revenue doesn't always land the day you spend the money. A TV campaign or a brand-building push on social might not move today's wave much at all — its wave arrives gradually, over days or weeks, as the impression sits with people before they act on it. A search ad, by contrast, is often a cannon right at the shoreline — someone's already looking to buy, and the wave hits almost instantly. If you only ever looked at today's spend against today's revenue, you'd credit the close cannons with everything and miss the slow-arriving waves from the distant ones entirely. A good model accounts for that lag, spreading each cannon's contribution out over the days or weeks it actually takes to land.

There's a second wrinkle that lives alongside carry-over: saturation. Even a well-placed cannon, fired at exactly the right amount of powder, starts producing smaller and smaller waves per pound of powder as you keep adding more. The first loads of powder buy real distance. Beyond some point, doubling the powder doesn't come close to doubling the wave — you're throwing more into a cannon that's already firing close to as far as it can. This is the diminishing-returns curve every channel has: cheap, efficient reach early, and progressively more expensive, less efficient reach as you keep pouring in budget.

Put the two together and you get the real question a marketer needs answered — not "how big was today's wave," but "for this specific cannon, how long does its wave take to arrive, and how close is it already to the point where more powder stops paying off?" That's the combination that turns the model from a description of the past into an actual guide for where next quarter's budget should go.

Where this technique actually comes from

This isn't a new idea dressed up in a beach metaphor. Marketing Mix Modeling has been around for decades, and it was built for exactly this situation: a world where you have no way to trace an individual sale back to an individual ad.

Think about a company running TV spots, radio ads, billboards, and print all at once, decades before anyone could click a link. Nobody could tell you which specific viewer bought the product because they saw the TV ad on Tuesday. There was no cookie, no pixel, no login. All the company had was what they spent on each channel, week by week, and what came in the door, week by week. MMM grew out of that constraint — it's a way of finding the signal in the aggregate numbers when you can't observe anything at the individual level.

For a long stretch of the digital advertising era, that constraint seemed to disappear. Multi-touch attribution came along and promised something much more precise: track a person's path — this person saw the Instagram ad, then clicked the Google search ad, then bought — and credit each touch along the way. It felt like an upgrade from "watching wave heights" to "watching every individual cannonball in flight."

But that precision was never as complete as it looked, and it's gotten less complete every year. Cookie deprecation, iOS privacy changes, walled gardens that won't share user-level data across platforms, and consumers increasingly declining to be tracked at all — these have quietly eroded the individual-level trail that multi-touch attribution depends on. A growing share of the customer journey now happens in the dark: on channels that won't share data, on devices that won't be tracked, through people who've opted out.

Which is why Marketing Mix Modeling is back at the center of the conversation. It never needed the individual trail in the first place. It was built for a world where you only get to watch the waves — and that happens to be, once again, the world most marketers are actually operating in.



Home