Every company that spends money on advertising is asking the same question: is it working? The answer to that question is harder to find than it looks. Platforms report results that don't match reality. Different methods produce different numbers. What "working" even means varies by business, channel, and time horizon.
Marketing measurement is the set of methods and tools used to answer that question honestly. This guide covers all of them: what each method measures, where it breaks down, and how to build a measurement approach that actually improves your decisions.
A complete measurement approach combines three methods across different time horizons
Why marketing measurement is hard
The fundamental challenge is that you can't run a control version of your business. You can't pause all your advertising for six months and compare what would have happened. You can't split your entire customer base into two identical groups and show ads to one half. Every measurement method is an approximation of a counterfactual that doesn't exist.
The goal is not perfect measurement. It's measurement that's good enough to make better decisions than you'd make without it.
Three structural problems make this harder than it looks.
The attribution problem. When a customer buys, they've usually interacted with multiple channels: a YouTube ad, an organic search result, a retargeting banner, a branded search click. Which one caused the purchase? Attribution models assign credit to those touchpoints, but the platforms reporting that attribution have a direct financial interest in claiming it. Meta's attribution says Meta drove the sale. Google's attribution says Google drove the sale. Both are reporting from inside their own data silo, and neither has an incentive toward accuracy.
The tracking problem. A large share of the customer journey is invisible to any measurement system. Users switch devices. They browse in incognito mode. iOS 14.5 and subsequent Apple privacy changes removed identifiers that advertisers used to track users across apps. Roughly 40% of desktop users run some form of ad blocker. Word of mouth, podcast listening, and offline conversations leave no digital trace at all. The customer journey you can measure is a partial map, not the full territory.
The causality problem. Correlation is easy to find and easy to misinterpret. If you spent $100k on YouTube in Q2 and Q2 sales were up 18%, that doesn't mean YouTube caused the increase. Sales might have risen due to seasonality, a PR mention, a competitor going offline, or organic demand growth. Attribution models measure co-occurrence, not causation. That distinction matters enormously when you're making budget allocation decisions based on the numbers.
The four main measurement approaches
Platform attribution
Platform attribution is what each ad platform reports in its native dashboard. When you open Meta Ads Manager or Google Ads, you're looking at platform attribution. It's fast, granular, and free. It's also the most systematically biased form of marketing measurement available.
Each platform uses its own attribution window. Meta's default is 7-day click and 1-day view: any conversion within 7 days of a click or 1 day of an ad impression gets credited to Meta. Google Ads defaults to a 30-day click window. TikTok uses 7-day click and 1-day view. Because these windows overlap, and because both platforms are attributing conversions from the same customer base, you routinely see platform-reported revenue that exceeds your actual backend revenue by 40-80%.
Platform attribution is appropriate for one thing: within-platform optimization. Comparing two Facebook creatives using Facebook's attribution is fine because both are subject to the same biases. Using Facebook's attribution to compare Facebook against Google is not, because the methodologies differ and both platforms grade their own homework.
Multi-touch attribution (MTA)
Multi-touch attribution solves one real problem with platform attribution: it gives you a unified cross-channel view. Instead of reading four separate platform dashboards that can't be reconciled, an MTA tool collects first-party event data from your site and creates a single view of the customer journey across channels.
Tools in this category include GA4, Northbeam, Triple Whale, and Rockerbox. The value is in the data layer: you can see that a customer touched Facebook, then read a blog post from organic search, then clicked a Google Shopping ad before converting. Platform dashboards can't show you that sequence because each one only sees itself.
The limitations are important to understand. MTA still can't track what it can't see. Cross-device journeys where users switch from mobile to desktop, incognito browsing, and any offline touchpoint remain invisible. Studies consistently suggest that 30-50% of customer interactions are not trackable by any pixel-based system. The journey you're seeing in your MTA tool is real, but it's incomplete.
More fundamentally, MTA is still a correlation-based system. It can tell you which touchpoints appeared in converting journeys. It cannot tell you which touchpoints caused the conversion.
Incrementality testing
Incrementality testing is the only method in marketing measurement that directly measures causality. The logic is simple: divide your audience or your markets into two groups, show ads to one group and not the other, then compare business outcomes. The difference in outcomes between the two groups is the incremental lift from your advertising.
There are three main forms of incrementality testing:
Holdout tests suppress ads to a randomly selected portion of your audience (typically 15-20%) while running normally to the rest. After 4-6 weeks, you compare conversion rates between the exposed and unexposed groups in your backend data. This is the cleanest way to measure whether a specific digital channel is driving real incremental conversions.
Geo experiments split geographic markets (cities, DMAs, countries) into test and control groups. You run your campaign in test markets and pause or reduce it in control markets. Because geographic markets are relatively self-contained, contamination between groups is limited. Geo experiments work well for channels that can't suppress ads at the audience level, including TV, out-of-home, and sometimes search.
Ghost ads (also called PSA testing) show a public service announcement or unrelated ad to the control group in the exact same placements your real ad appears in. This controls for any effect of simply being exposed to an ad impression, and isolates the effect of your specific message.
The trade-offs with incrementality testing are real. Each test takes 4-8 weeks to produce reliable results. You can't run tests on every channel simultaneously. The holdout group represents real potential customers you're not advertising to, which has a cost. And tests need to be designed carefully: underpowered tests (too small a holdout, too short a duration) produce results that look meaningful but aren't statistically reliable.
Despite these constraints, incrementality testing is the most valuable measurement investment most companies can make. The findings consistently surprise teams who have been making decisions based on platform attribution.
Media mix modeling (MMM)
Media mix modeling uses statistical regression to decompose business outcomes (revenue, sales, units) into their contributing factors: media spend across channels, baseline organic demand, seasonality, pricing, promotions, and external economic factors. Rather than tracking individual customer journeys, MMM works at the aggregate level: it observes that when Facebook spend goes up, sales tend to go up by a certain amount over a certain lag period, and estimates the contribution of each channel to overall business performance.
MMM does not rely on user tracking, cookies, or pixels. It works from aggregated historical data. This makes it uniquely suited to the post-cookie measurement environment and to channels that leave no digital trace (TV, radio, out-of-home, podcast).
The historical data requirement is significant. A good MMM typically needs at least one year of weekly data, and ideally two or more years, to distinguish media effects from seasonal variation and baseline demand. You also need variation in your spend: if you've been spending the exact same amount on Facebook every week for a year, the model can't estimate the relationship between Facebook spend and revenue.
Modern MMM implementations use Bayesian statistics (Google's Meridian and Meta's Robyn are both Bayesian) rather than classical regression. Bayesian MMM lets you incorporate prior knowledge about marketing dynamics (for example, that TV has a longer carryover effect than search) and produces probability distributions over estimates rather than single point estimates. This uncertainty quantification is important: it tells you how confident the model is in each channel's contribution.
What MMM is good for: strategic budget allocation, understanding brand and reach channels that don't show up in attribution, measuring the carryover effects of advertising, and working in a privacy-safe way without user-level tracking.
What MMM is not good for: real-time optimization, understanding performance at the campaign or creative level, or companies with small budgets where the spend signal is too noisy to model reliably.
How the methods compare
| Platform Attribution | MTA | Incrementality | MMM | |
|---|---|---|---|---|
| Speed | Real-time | Near real-time | 4-8 weeks | Monthly/quarterly |
| Causal? | No | No | Yes | Partial |
| Cookie-free? | No | Partial | Yes | Yes |
| Granularity | Campaign/ad level | Touchpoint level | Channel/campaign | Channel level |
| Cost | Free (built-in) | $300-$5k/mo | Time and holdout cost | $0 open source to $500k/yr vendor |
The concept of marketing triangulation
No single method gives you a complete picture. Platform attribution is fast and granular but biased and siloed. MTA gives cross-channel visibility but can't prove causality. Incrementality testing provides causal proof but is slow and can't cover everything. MMM is comprehensive and cookie-free but lagging and strategic rather than tactical.
Marketing triangulation uses all three together. The name comes from navigation: a navigator uses multiple fixed reference points to determine exact position, because each individual reference point has some uncertainty. Three independent measurements that agree give far more confidence than any single measurement, regardless of how sophisticated it is.
In practice, triangulation means using MMM for strategic budget allocation, incrementality testing to validate and calibrate that allocation, and attribution for operational optimization signals within campaigns. Where the methods agree, you can have high confidence. Where they diverge, that divergence is data: something interesting is happening that's worth understanding before acting on.
A common example of useful divergence: your MMM says Facebook contributes 9% of revenue with a 1.8x ROI. Your Facebook-attributed ROAS is 4.2x. You run a holdout test on Facebook retargeting and find 35% incremental lift. The incrementality data and the MMM directionally agree that Facebook's true contribution is much lower than attributed. That convergence of two independent methods gives you confidence to reallocate budget, even without perfectly reconciling the exact numbers.
How to build your measurement approach
The right measurement stack depends on your current spend level. More measurement is not always better: the complexity and cost of sophisticated measurement tools need to be justified by the scale of the decisions they inform.
Stage 1: under $200k/month in total ad spend. Start with clean data infrastructure and a properly configured GA4. Fix your conversion tracking if it's broken. Pull your backend revenue data (from Shopify, your CRM, or your payment processor) into a simple spreadsheet and track blended MER weekly. Run one or two incrementality tests per year on your largest channels. This setup costs almost nothing and gives you far more reliable data than relying on platform dashboards.
Stage 2: $200k-$1M/month. Add an MTA tool for cross-channel visibility. Run incrementality tests quarterly on rotating channels. Evaluate open-source MMM using Google Meridian or Meta Robyn if you have internal data science capability. If you don't, start evaluating vendor MMM at the higher end of this range. The investment in measurement infrastructure starts to pay for itself in improved allocation decisions.
Stage 3: over $1M/month. The full triangulation stack is justified at this spend level. Invest in vendor MMM or a robust open-source implementation calibrated against your incrementality data. Run a continuous incrementality testing program that rotates through your major channels on a quarterly cycle. Use attribution for operational optimization only, not for strategic budget decisions. The measurement complexity is proportional to the magnitude of the decisions it informs.
Common measurement mistakes
Using platform-reported ROAS to allocate budget across channels. Each platform's ROAS is reported using different methodology, different attribution windows, and data the other platforms can't see. Comparing Meta ROAS to Google ROAS is comparing numbers that aren't measuring the same thing.
Running incrementality tests for less than four weeks. Short tests capture day-of-week variation and random noise rather than true lift. Four weeks is a minimum; six to eight weeks is better for most channels.
Building measurement around the data you have rather than the questions you need to answer. If your most important question is whether YouTube is driving incremental awareness, build a geo experiment to measure that. Don't look at YouTube's attributed conversions and conclude it's not working because the numbers are low.
Treating attribution as proof of causality. Attribution tells you which touchpoints were present in customer journeys. It does not tell you which ones caused the journey to end in a conversion. These are fundamentally different questions.
Measuring marketing performance in isolation from business outcomes. Clicks, impressions, and attributed conversions are not business metrics. Revenue, contribution margin, customer lifetime value, and new customer acquisition cost are. Measurement systems that optimize for marketing metrics rather than business outcomes will lead you toward campaigns that look good in the dashboard and produce poor business results.