How triangulation works
No single measurement of marketing is complete, so triangulation stops trying to find one. It reads three views of the same activity and looks for agreement. Platform reporting is fast, detailed and biased towards the platform that produced it. Experiments such as holdout and geo tests are causal and trustworthy but slow, costly and only available for channels with enough volume. Marketing mix modelling covers everything, including offline and brand activity, but works on long aggregate history and cannot say anything about last week.
Each view is strong exactly where the others are weak. When all three point the same way, you can act with confidence. When they disagree, the disagreement itself is the finding, and it usually points at a specific fault: a tagging break, double counting between platforms, or a channel taking credit for demand it did not create.
Why triangulation matters
Every measurement method has a known direction of error. Platform attribution over-credits channels that sit close to the purchase. Analytics under-credits anything that lost its cookie or arrived through a messaging app. Modelling flatters spend that follows demand rather than creating it. If you rely on one method, you inherit its bias silently and never see it.
Using several views makes the bias visible. It also changes the conversation with a client or a board, because the question stops being “whose dashboard is right” and becomes “what would have to be true for these views to agree”. That is a question with an answer.
Where triangulation goes wrong
The most common failure is shopping for the friendliest number. Once you have several sources, it is easy to quote whichever one supports the decision already made, which is worse than having a single flawed source honestly applied.
The second is comparing views that were never comparable: a platform report on a click-dated window against analytics on the date of purchase, over different periods, with different definitions of a conversion. Line the definitions up first or the disagreement is meaningless. The third is treating a rough model and a controlled experiment as equal evidence. They are not, and when they conflict the experiment should usually win.
How to act on it
Decide in advance which view answers which question. Use platform data for daily and weekly optimisation, since only it knows which creative and audience did the work. Use business-level blended figures as the monthly headline, because that is where profit actually appears. Use experiments to settle any argument big enough to change a budget, and modelling for annual or quarterly planning if the business has the history for it.
Write that hierarchy down before the numbers disagree, so nobody is choosing a method after seeing the result. For most small and mid-sized advertisers, including almost every business I work with in Nepal, the practical version is modest: consistent tracking, a blended monthly figure from real sales records, a thank-you page question about how customers heard of you, and the occasional on-off test on one channel. That is enough to catch the errors that matter.
