Measurement got a lot more accessible for mid-market brands over the last few years. Google and Meta open-sourced their MMMs. Independent platforms like Haus, Measured and Recast made testing and modeling possible without an in-house data science team. Google cut the minimum budget for its lift tests from six figures to four. And growth marketers got more fluent along the way: most can explain why a prospecting conversion is more likely to be incremental than a brand search one. The advice that came with all of it was to triangulate across attribution, incrementality and MMM (you’ve seen the slide).

Triangulation is the measurement version of “eat a balanced diet.” Everyone agrees with it, and almost nobody tells you what to eat first. Most brands we talk to still steer on platform-reported numbers, report to the C-suite on CRM or Shopify numbers and throw their hands up at the gap between the two.

So here’s the order we’d follow for a mid-market brand starting from scratch, and how to know when you’re ready for the next step. There are exceptions, and we’ll get to them at the end.

1. Clean up your conversion data, server-side tracking and pixels.

2. Build a framework for acting on attribution within one channel, like Google Ads or Meta.

3. Lift-test channels where you have the biggest questions on impact.

4. Earn your way into MMM once you have the spend and history to justify it.

Make the platform’s numbers worth arguing with

Every ad platform and Google Analytics counts conversions and revenue its own way, so check that primary and secondary conversions are set up correctly, that server-side tracking is live without double-counting what the pixel already fires, and that attribution windows and models are as consistent as the platforms allow. Apply tagging, UTMs and naming conventions systematically on every ad. Then reconcile the platforms and analytics against your source of truth (CRM, Shopify or finance). They’ll never match. If platform conversions usually run 1.6x what your CRM reports and next month they jump to 2.8x, something broke.

They don’t have to agree. They have to move together.

Attribution is where a team learns to argue with the platform

We’re skeptical of most cross-channel attribution platforms outside Google Analytics. The data limitations, the implementation lift and the cost, in both time and money, make them hard to justify. In-platform and GA attribution are a different story: they’re where a team learns to argue with the platform. Reallocating credit across YouTube, PMax, non-brand and brand search inside Google Ads gets a team comfortable assigning different values to different activities, and being wrong here is cheap. Once your team has moved budget based on a view other than the platform default, and can explain why the two views disagreed, you’re ready for the next step.

Lift tests turn arguments into budget shifts

We’ve now had the amuse-bouche and the appetizer. It’s time for the entrée. Lift tests are the only tool here that establishes cause and effect directly: they show which marketing actually drives incremental revenue and which is just loitering near conversions that were going to happen anyway. Randomized control tests (user-level holdouts, geo-lift or geo-holdouts) are ideal. Staggered rollouts and on-off flighting are messier, but can still deliver useful insights.

Use incrementality testing to evaluate whether new channels worked, which of two new channels works better, or whether you’re overinvesting in non-incremental activity. Don’t overcomplicate it. New upper-funnel channels where it’s harder to tie back conversions and purchases make great candidates (CTV, podcast, streaming audio, TikTok awareness campaigns). Brand search and remarketing are almost always our two biggest priorities for holdouts. And we’ve repeatedly seen YouTube drive more lift than it got credit for, even in Google’s own attribution.

In every case, you should be trying to assign an incremental ROAS at the current investment level to inform a budget shift. We’ve heard way too many times, “We tested brand search a couple of years ago, and it was incremental. 5% of the conversions didn’t get picked up by other channels, and brand search has a 15:1 ROAS, so we’re never going to cut that.” Do the math. At 15:1 reported and 5% incremental, every $100,000 in brand spend buys $75,000 in revenue you wouldn’t otherwise have had. If $50,000 in CTV drove $125,000 in incremental revenue, the brand budget is the first place to look.

Lift tests also don’t need perfect pixels. A geo test compares total sales in markets that saw the ads against markets that didn’t, pulled straight from Shopify, your CRM or finance, with no attribution involved. Two caveats. Platform-run lift studies on Meta and Google count conversions through their own tracking, so they inherit whatever’s broken there. And your sales data has to be clean by market, with orders tied to a shipping ZIP or DMA.

The test doesn’t need clean pixels. Acting on the result does. Once you’ve run multiple lift tests, use them to apply new modifiers to platform- or analytics-reported numbers and shift budgets as a result. Those modifiers only hold while the platform numbers underneath them stay stable.

Earn your way into MMM

Meridian and Robyn, Google’s and Meta’s open-source models, made MMM cheap, and plenty of very smart people argue for starting top-down. Optimizing your overall media mix is usually the highest-leverage move you can make, and that’s exactly what MMM is built for.

Our counter: lift tests are faster, they start at $5,000 in Google Ads, and they’re what calibrate an MMM later. MMM is cheap to run now, but it still needs enough spend to learn from, which is why we wouldn’t consider one under about $5 million a year in media. An MMM built on spend that barely moved can tell you very little about where to shift budget, but it will usually still give you a precise-looking answer. Leading with lift tests also gets you to the easy budget reallocations much sooner.

That said, building and, more importantly, refining an MMM pays compounding dividends in getting your portfolio mix right. The woman who led analytics for us at a prior agency was one step shy of getting George Box’s line, “All models are wrong, but some are useful,” tattooed on her bicep. That applies to MMM more than the other two. The brands we’ve seen use MMM best go in with their eyes open. It isn’t a source of truth or a silver bullet. It’s a way to raise your confidence in major budget decisions.

When the tools disagree, the lift test wins

Once all three are running, each gets its own job and cadence. Attribution steers inside each channel week to week. Lift tests check the biggest channels and new bets every quarter, and the results become the modifiers you apply to platform numbers. MMM sets the annual budget split and gets recalibrated whenever a new lift result comes in.

When an MMM model said to cut Meta while a lift test showed it had the best incremental CPA of any channel, we put more money into Meta, then fed that result back into the model.

Each tool answers its own question. The lift test breaks the ties.

Some brands should start with MMM

Two kinds of brands should move MMM up, if not to the front. The first is CPG and other indirect sellers. Without reliable in-platform conversions, Step 1 barely exists for them, and an MMM’s top-down sales view is worth more. Geo tests on retail sales still work, so they skip a step rather than flip the order.

The second is any brand where lift tests can’t reach enough of the budget. That happens when spend is spread thin: on 14 channels with none above 25% of spend, you can’t test your way to an answer in a reasonable time. It also happens when a big share of spend sits in channels you can’t split by market, like national TV, host-read podcast sponsorships and influencer.

Two cases change Step 2 without moving MMM up. Under $100,000 a month in total budget, or with less than 10% of budget to put behind a new channel, skip extensive lift testing and wait until you start to hit diminishing returns on core channels. And in long-cycle B2B, lift tests need a proxy outcome. You won’t get a reliable read on closed-won deals from a 60-day test when your sales cycle is 180 days.

Pick a step and set a date

There are roughly 27,203 versions of the triangulation slide in circulation (it’s true, we lift tested it). They did their job: almost every growth marketer knows the three tools exist. None of them tells you where to start. Pick the step you’re actually on, write down what it takes to finish it, and give yourself six months to get one step further. You don’t eat a balanced diet in one sitting.