If you spend real money on marketing, sooner or later someone asks which half is wasted. Two families of methods try to answer that, and they come at it from opposite ends. Choosing between them is mostly a matter of what data you have and what decision you're trying to make.
Attribution: follow the customer
Attribution traces individual journeys. Someone clicks an ad on Monday, reads an email on Wednesday, searches your brand on Friday and buys. The attribution model decides how to share the credit among those touches. Last click gives it all to the final one. Linear splits it evenly. Position-based gives most to the first and the last.
Its big strength is detail. You can see down to the campaign, the ad, sometimes the keyword. It updates daily, so you can act fast.
Its weakness is that it only sees what can be tracked. Privacy settings, browser restrictions, consent banners and people switching between phone and laptop all leave holes. Offline influences, like a billboard or word of mouth, are invisible. And a tracking-based model tends to flatter channels that sit close to the sale, such as brand search, because they're present at the end whether or not they caused the purchase.
Marketing mix modelling: look at the totals
Marketing mix modelling, or MMM, takes a different route. It doesn't follow anyone. It takes your sales over time, say weekly, and your spend in each channel over the same period, plus things that affect sales anyway, like seasonality, price changes, promotions and holidays. Then it uses a statistical model, usually a regression with some adjustments, to estimate how much each channel contributes.
It can include channels that can't be tracked at click level, like television or print. It doesn't depend on cookies or identifiers. It also helps with a question attribution handles badly: what happens to sales if I double this channel, or cut it?
The costs are real. It needs a fair amount of history, generally a couple of years of weekly data, and it needs variation. If you spent exactly the same amount on every channel every week, the model can't tell their effects apart. It updates slowly, so it won't tell you about last Tuesday's ad. And it gives an estimate with a range, not a single crisp number, which some people find unsatisfying. I think that's more honest than it sounds.
A rough guide to choosing
Mostly digital, a lot of tracking, quick decisions on campaigns: start with attribution. It's cheaper, faster, and gives you the detail to adjust creative and targeting. Just be aware of what it can't see.
Big budgets, several channels including offline, decisions about how to split a budget: MMM is worth the effort. You'll learn things that clicks can't show you.
Not much data, small budget: neither is likely to give you reliable answers. Simpler tools do the job. Compare periods, run a deliberate test such as switching a channel off in one region, and look at spend against sales on a chart before reaching for a model.
They work best together
The two methods check each other. If attribution says paid search drives 40 percent of your sales and the mix model says 15, that gap is telling you something, usually that search is claiming credit for demand created elsewhere. Neither is automatically right. The difference is where you should look.
A third tool is worth knowing: incrementality tests. You hold out a portion of an audience or a region, don't show them the ads, and compare. It is the cleanest evidence of cause and effect. It takes planning, but one good test can settle an argument that months of modelling couldn't.
A few cautions
Don't take a model's output as a command. Treat it as a well-informed opinion, and combine it with what you know about the business.
Check what the model assumes. Many tools quietly assume that advertising effects fade at a particular rate, or that returns diminish in a certain shape. Those assumptions drive the answer.
Be careful with data quality. The most common reason an MMM project goes sideways is not the statistics. It's that the spend figures, the sales figures and the promotion dates don't line up, or that someone changed how a metric was recorded halfway through.
And start by writing down the decision. "Should we move 20 percent of budget from channel A to B next quarter?" gives you a clear brief. "Tell us how marketing is doing" doesn't.