The Ecomm Analyst

Every vendor in this category will tell you that you need all three. Multi-touch attribution for the daily decisions, marketing mix modeling for the budget, incrementality testing to keep both honest. It is a tidy story and it is not wrong. It is just priced for someone else.

The useful question is not which method is most rigorous. It is which method earns its cost at your spend level. So here is how I actually think about it, by ad spend.

The rule that governs everything below

Measurement has to be cheap relative to the money it moves. My working ceiling is one to two percent of media spend on measurement tooling. Above that, you are spending real margin to find out whether you are spending real margin well, and the recursion stops being funny.

Run the arithmetic before you run the demo. A $1,500 per month attribution platform against $40,000 per month in ad spend is 3.75 percent. That tool needs to find you a lot of waste to break even, and it needs to find it every month, forever.

Under $50k per month in ad spend

You need multi-touch attribution and a post-purchase survey. That is the entire list.

At this spend level you have two or three channels that matter, and the decisions in front of you are coarse. Should I put more into Meta or Google. Is TikTok doing anything. Is my email revenue real or is Klaviyo claiming credit for people who were going to buy anyway. None of those questions require a model. They require clean data and an honest lookback window.

The survey matters more than people think. It is the only signal you have that is not derived from a cookie or a pixel, which makes it the one thing that can tell you when your attribution model is wrong rather than just precise. I have written about using post-purchase surveys as a real attribution signal in more detail.

Budget here should be in the low hundreds per month. Tools in this bracket run from roughly $99 to $300. ThoughtMetric, which sponsors this site, sits at the bottom of that range and bundles the survey. Whatever you pick, do not buy a platform that costs more than one percent of what you are spending on ads.

$50k to $250k per month

Add incrementality testing, but run it yourself.

This is the range where the questions get expensive to get wrong. Is brand search actually incremental or are you paying to intercept people who typed your name. Is retargeting driving purchases or harvesting them. Those are causal questions, and no attribution model, however sophisticated, can answer them. Attribution assigns credit. It does not establish cause.

The good news is that a geo holdout costs you almost nothing but discipline. Turn a channel off in a set of matched regions, leave it on elsewhere, wait four weeks, compare revenue. I wrote a walkthrough of running a cheap geo holdout that assumes no data team and no vendor.

You do not need an incrementality vendor yet. Platforms like Haus and INCRMNTAL do this properly, and if you have the budget they will do it better than you will. But at $100k per month in spend, a managed incrementality product priced in the low thousands per month is a meaningful slice of your measurement budget for something you can approximate yourself two or three times a year.

$250k to $1M per month

Now marketing mix modeling starts to pay for itself, and now you can afford a managed incrementality program.

MMM is worth being blunt about, because it is oversold to brands that cannot feed it. It needs two to three years of weekly history and, more importantly, it needs variation. If you have spent roughly the same amount on the same channels every week for two years, the model has nothing to learn from. It will produce confident-looking coefficients built on noise.

What makes MMM valuable at this tier is not that the math got better. It is that you now have enough channels, enough seasonality, and enough deliberate budget shifts that there is real signal in the history. You also now have upper-funnel spend that multi-touch attribution structurally undercounts, which is exactly the gap MMM fills.

Expect real money. Attribution platforms aimed at this tier start around $1,500 per month and go up. Managed incrementality testing is often priced per test, with first tests in the low thousands. Unified platforms that bundle attribution, MMM, and testing tend to be quote-only, and buyer-disclosed data puts brands spending $100k to $500k per month at roughly $40,000 to $90,000 in annual contracts. At $250k per month in spend, a $60,000 annual contract is two percent. That is defensible. At $60k per month in spend, the same contract is eight percent, and it is not.

Above $1M per month

Run all three, and run them against each other.

The point of holding three methods at once is not to average them into one number. It is to watch where they disagree. When MTA says a channel is working and a geo holdout says it is not, that gap is the most valuable thing in your reporting stack. It usually means the channel is harvesting demand that another channel created.

Use the tests to calibrate the models. Use the models to decide what to test next. Use MTA for the daily and weekly optimizations where being roughly right quickly beats being precisely right in six weeks.

What this really comes down to

Most brands under $250k per month in ad spend are being sold measurement infrastructure designed for brands ten times their size, and they are buying it because the deck is convincing and nobody wants to admit that their measurement problem is mostly a UTM hygiene problem.

Fix the tracking. Add a survey. Run a holdout a couple of times a year. Then, and only then, start pricing the models.