There is a problem at the center of marketing measurement that most people work around rather than solve: you can never run a controlled experiment on your business.
You cannot pause all your advertising for six months and compare what would have happened. You cannot run two identical versions of your company side-by-side. Every measurement approach you use is, by definition, looking at an approximation of reality. The question is not whether your measurement is perfect, it is not, but whether the approximation is good enough to make better decisions.
Marketing triangulation is the practical answer to this constraint. It is not a software product, not a vendor's proprietary methodology. It is a principle: use multiple independent measurement methods, look for where they converge, and investigate where they diverge.
Confidence in a decision rises when multiple measurement methods agree
Where the Term Comes From
In navigation, triangulation means using multiple fixed reference points to determine your exact position. One landmark gives you a line you might be on. Two landmarks narrow it to a point. Three independent landmarks give you a precise fix, and let you catch errors in any single measurement.
The same logic applies to marketing measurement. Any single method has blind spots. A method that tracks individual users misses offline behavior, cross-device journeys, and the long-run brand effects that build over months. A method that models aggregate data is slow to update and cannot tell you which specific campaign drove a result this week. A method that relies on platform reporting has a conflict of interest baked in.
Combine multiple methods and you can stress-test your conclusions. Where they agree, you can have confidence. Where they diverge, you know something interesting is happening, and you should investigate before acting.
The Three Methods and Their Trade-Offs
Marketing triangulation draws on three core measurement approaches. Each has distinct strengths. Each has significant limitations. Understanding both is what makes the combination valuable.
Media Mix Modeling (MMM)
Media mix modeling uses statistical regression to explain business outcomes, sales, revenue, leads, app installs, as a function of media spend across channels, plus external variables like seasonality, price, competitive activity, and macroeconomic conditions. The model then decomposes how much of your revenue would have happened without any advertising (the baseline), and how much each channel contributed on top of that.
MMM works at the aggregate level. You feed it weekly or monthly data: total spend on TV, total spend on paid search, total revenue, promotions run, and so on. No individual user tracking is required. This makes it naturally privacy-safe, a significant advantage as third-party cookies disappear.
The main strengths of MMM: it captures the long-run effects of advertising (a brand campaign this quarter might drive sales next quarter, and MMM can model that lag); it includes channels that attribution tools cannot track, like TV, OOH, and podcast; and it provides a cross-channel view uncorrupted by each platform's self-interested attribution.
The limitations are real: MMM is slow. You typically need 2-3 years of historical data to build a robust model, and results reflect the past, not the present. It cannot tell you which specific campaign within a channel was driving the effect. And the output depends heavily on model assumptions, adstock rates, saturation curves, that require significant expertise to set correctly.
Typical cost: $50,000-$200,000 for a vendor-built model, with ongoing retaining costs. Open-source tools like Google Meridian and Meta Robyn are free but require real data science capability to use well.
Incrementality Testing
Incrementality testing uses controlled experiments to answer a specific question: if we had not shown these ads, would these conversions still have happened?
The core concept is counterfactual reasoning. You create a control group that does not see your ads and compare their conversion rate to the group that does. The difference is the incremental lift, the conversions your advertising actually caused. Everything else is organic demand that would have converted regardless.
There are three main approaches. Holdout tests (also called ghost tests or audience split tests) randomly split your audience and suppress ads to the holdout group. Geo experiments split geographic markets into test and control and measure business outcome differences. Ghost ads (PSA testing) show a control group an unrelated ad in the same placement, providing a cleaner counterfactual than simple suppression.
Incrementality testing is the only method that directly measures causality. MMM provides causal estimates but relies on model assumptions. Attribution provides correlation data. Holdout tests and geo experiments give you actual causal evidence from actual experiments.
The limitations: you cannot run incrementality tests on everything at once. Each test takes 2-8 weeks and covers one channel or one audience at a time. Running a comprehensive testing program requires patience and systematic prioritization. And some findings are uncomfortable, most brands discover their retargeting campaigns have significantly lower incremental value than platform-reported ROAS suggests.
Attribution Modeling
Attribution modeling tracks individual touchpoints in the customer journey and assigns credit for conversions to those touchpoints. It is the most widely used measurement approach because it is fast, granular, and built into every major ad platform.
Attribution models range from simple rules (last-click gives all credit to the final touchpoint; first-click gives all credit to the first) to algorithmic approaches (data-driven attribution uses machine learning to weight touchpoints based on which appear more often in converting paths versus non-converting paths).
The appeal of attribution is real-time feedback. You can see performance by campaign, by creative, by audience, by day. For within-channel optimization, comparing two Meta campaigns against each other, attribution signals are useful.
The problems with attribution are structural. It can only measure what it can track, and tracking is deeply incomplete. Cross-device journeys, incognito browsing, offline purchases, word-of-mouth, and podcast listening are all invisible to attribution tools. Studies consistently suggest 20-40% of the customer journey cannot be seen by digital tracking, and that number is growing as privacy protections tighten.
More fundamentally: every major ad platform runs its own attribution. Meta attributes conversions to Meta ads. Google attributes conversions to Google ads. TikTok attributes conversions to TikTok ads. Each platform's attribution window and logic is set by the platform itself, an entity with a direct financial interest in showing high performance. This is not fraud; it is a structural incentive problem that produces systematically optimistic numbers.
Why Each Method Alone Is Insufficient
MMM without incrementality testing: you have a cross-channel view and can make budget allocation decisions, but you cannot prove causality. The model's channel coefficients are correlational estimates, not causal proof. And MMM is too slow for campaign-level optimization, by the time the model updates, the moment has passed.
Incrementality testing without MMM: you can prove what works causally, but you can only test one thing at a time. A testing-only program cannot give you a complete picture of cross-channel ROI. You will always have channels you have not yet tested, and you cannot prioritize budget allocation across all channels simultaneously.
Attribution without the other two: you have fast feedback for within-channel optimization, but you are making budget allocation decisions based on data that is systematically biased toward bottom-funnel channels and that is produced by parties with conflicting interests. You will over-invest in retargeting, under-invest in brand and awareness, and have no way to detect how large the error is.
The three methods have complementary blind spots. Where one is weak, another is strong. That is the core logic of triangulation.
How Triangulation Works in Practice
Here is a concrete scenario that plays out at many brands.
A company runs MMM and finds that Facebook has weak ROI, the model shows Facebook driving approximately 0.8x revenue return on ad spend. Platform attribution shows Facebook at 3.2x ROAS. The marketing manager is skeptical of the MMM, it seems to contradict what the platform is reporting, and the attribution data looks clean.
They run an incrementality test: a holdout study on their Facebook retargeting audience, running for four weeks. Results: the holdout group converts at 74% the rate of the exposed group. That means roughly 74% of Facebook retargeting conversions were happening organically, the ads were not causing them. The true incremental ROAS is closer to 0.9x.
Now the picture is coherent. MMM flagged weak Facebook ROI. The incrementality test confirmed it. The platform attribution was wrong because it was giving Facebook credit for organic conversions from users who were already planning to buy. The decision is clear: reduce Facebook retargeting spend, shift budget to Facebook prospecting (where incrementality is higher because you are reaching cold audiences who would not have found you otherwise) or to other channels with higher incremental ROI.
This is triangulation in action. The MMM and incrementality data pointed in the same direction. The attribution data diverged. The divergence was a signal worth investigating, not a reason to dismiss one method and trust another.
When Triangulation Makes Business Sense
Marketing triangulation is not the right approach for every company. Be honest about the requirements.
Minimum ad spend: the rough threshold is $100,000-$200,000 per month in advertising spend. Below that level, the cost and time investment in comprehensive measurement often exceeds the expected savings from better allocation. At lower spend levels, focus on getting your tracking fundamentals right and running one well-designed incrementality test per year.
Data infrastructure: you need clean, consistent historical data, ideally 18 months to 3 years of weekly spend and revenue data, with reasonable breakdowns by channel. If your data is unreliable or incomplete, your models will be unreliable. Garbage in, garbage out applies more to MMM than almost any other analytical tool.
Time horizon: triangulation is not a quick answer. An incrementality test takes 4-8 weeks per channel. An MMM takes 3-6 months to build and calibrate. You need organizational patience for a measurement approach that produces results on this timeline.
Organizational buy-in: the hardest part of triangulation is not technical, it is getting internal agreement to act on results that contradict platform reporting. When an incrementality test shows Facebook's true ROI is a third of what the platform reports, the Facebook channel manager will push back. Leadership needs to be willing to take actions based on independent measurement, even when the platforms disagree.
Where to Start
Most companies should begin with incrementality testing, not MMM. Here is why: incrementality tests produce results in weeks, not months. They are cheaper to run than a full MMM build. And they produce the kind of concrete, surprising findings, "our retargeting campaign is only 30% incremental", that build internal credibility for better measurement. For a practical breakdown of all three methods side by side, see MMM vs. incrementality testing vs. attribution.
Start with your highest-spend channel. If retargeting is your largest line item, run a holdout test on retargeting first. The findings will almost certainly be informative, and the channel size means the financial implications of acting on those findings are substantial.
Once you have one or two incrementality test results, you have calibration data, a way to estimate how much your attribution is inflating or deflating performance for those channels. This calibration is useful even before you invest in MMM.
Layer in MMM when you have the data infrastructure, the budget, and the organizational appetite. MMM provides the cross-channel, long-run view that incrementality testing cannot, especially for brand channels like TV, YouTube, and podcast that are difficult to test incrementally but may be doing significant work.
The goal is not to run all three methods simultaneously from day one. It is to build a measurement program that improves over time, each method adding a perspective that the others cannot provide, each finding sharpening your understanding of what is actually driving your business results.