By: Atul Mohan, PhD
Why the channels that look most efficient in a dashboard may be capturing demand created somewhere else.
Atul Mohan, PhD, is a senior data and AI executive with more than 24 years of experience turning complex analytics into measurable business outcomes. A Princeton-trained data scientist, he has held leadership roles across American Express, Interpublic Group, Omnicom, Teachers Federal Credit Union, and Brainlabs, building and guiding teams at the intersection of data science, marketing analytics, governance, and AI strategy.
It looks like your best-performing channel. It is mostly taking credit for work done somewhere else.
Pull up almost any paid media report, and brand search will be sitting near the top. Low cost per acquisition. High conversion rate. Excellent return on ad spend. Whatever optimization system you point at that report is going to recommend moving money toward it.
Do that, and your business gets worse. Not because the measurement is sloppy, but because the model is structured wrong in a way that no amount of cleaner data will fix.
Start with what a brand search actually is. Somebody types your company name into a search engine. That person has already decided to consider you. Something made that happen. A television flight, a social campaign, a friend’s recommendation, a product experience from two years ago that they are only now acting on.
The brand search click is where that intent surfaces. It is not where the intent was created.
In causal language, brand search sits on the path between your upper funnel activity and the eventual conversion. It is a mediator. And putting a mediator into a model as an independent variable, right alongside the things that cause it, is a specification error with a well-understood consequence. The mediator absorbs the effect of everything upstream.
Your model is not wrong that brand search converts well. It is wrong about what caused the conversion.
What makes this genuinely dangerous, as opposed to merely incorrect, is the way the damage unfolds over time.
Budget shifts toward brand search. Upper funnel spend gets trimmed. And for a while nothing bad appears to happen, because the demand your previous upper funnel work created is still moving through the system. Brand search volume holds steady. Efficiency metrics look better than they have in years. Somebody gets promoted on the strength of it.
Then, with a lag that depends entirely on your category and purchase cycle, the volume of people searching your brand starts to slide. By the time that shows up clearly in the reporting, the spend that would have prevented it was cut three or four quarters ago, and the person who cut it has a different job and a strong story about the efficiency gains they delivered.
I have watched this cycle run more than once. It is remarkably hard to diagnose from inside the company, because every intermediate metric is improving right up until it is not. There is no moment where a dashboard turns red, and someone raises a hand. There is a slow drift, and then a quarter where the acquisition number misses, and then a search for explanations that lands on the market, the category, the macro environment, anything except a budget decision made a year earlier that looked correct at the time.
There are a few ways to handle it, and none of them are free.
The cleanest is experimental. Turn brand search off in a set of matched geographies and watch what happens to total conversions, not to search conversions. The results tend to be uncomfortable for whoever owns the channel, because a meaningful share of that traffic converts anyway through the organic result sitting directly below the ad.
This test costs very little. It takes a few weeks. The fact that so few advertisers run it says more about organizational incentives than about methodology. Nobody wants to run an experiment whose most likely outcome is that a line item they are responsible for turns out to be less valuable than the report says.
In a mix model, you can specify the structure honestly, treating brand search volume as an intermediate outcome driven by your upper funnel spend rather than as another channel competing on the same footing. This is more work, and it requires you to take an explicit position on the causal graph, which some teams find uncomfortable.
That discomfort is appropriate and worth sitting with. You were taking a position already. Leaving brand search in as an independent input is a causal claim. It is just an unexamined one, and unexamined claims have a way of being wrong in whichever direction is most convenient.
The pragmatic middle path, if the full restructure is beyond what your organization will tolerate, is to split brand from non-brand completely, report them separately, and refuse to let allocation decisions treat them as substitutes for one another. It does not solve the attribution question, but it does stop the slow bleed of budget out of demand creation.
What I would push back on hardest is the most common response of all, which is to acknowledge the problem in a footnote and then carry on reporting brand search as a top-performing channel.
Everybody in the room knows the number is not what it appears to be. The number keeps driving decisions anyway. A caveat that does not change anyone’s behavior is decoration, and I would argue it is worse than no caveat at all, because it lets the organization feel sophisticated about a mistake it is still making. It also gives everyone cover. When the acquisition number eventually misses, several people can point to the footnote and say they had flagged it.
There is a version of this problem that shows up in defensive language, and it is worth addressing because it sounds reasonable. The argument goes that you have to bid on your own brand terms, otherwise competitors will, and you will pay more later to recover traffic you used to get free.
Sometimes that is true. In categories with aggressive conquesting, it can be clearly true. But it is an empirical question with a knowable answer, and most organizations treat it as an article of faith rather than testing it. The geographic holdout answers it directly. You will find out whether your organic result holds the traffic, in which categories, and at what competitive intensity. Running the test converts a belief into a number.
There is a broader lesson here that goes past search.
Any channel that sits close to the conversion will look efficient. Retargeting has the same property. So does email to an existing list. So does organic traffic from a comparison site that only exists because somebody built the brand that people are comparing.
Measurement systems reward proximity to the sale. The further upstream a channel sits, the worse it will look in a report and the more likely it is to be cut first when budgets tighten. Which means the structural bias in your measurement runs in the same direction as the structural bias in a difficult quarter, and those two pressures compound.
That is not a measurement problem you can solve with better tracking. It is structural; it will get worse as tracking signal continues to erode, and the only reliable defense is experimentation. Not occasional experimentation when somebody gets curious, but a standing program with a budget line, run on a cadence, with results that feed back into how the models are specified.
The uncomfortable truth is that the channel your dashboard likes most is frequently the one contributing least. Somebody in your organization should be willing to say that out loud in a budget meeting, and be prepared to prove it.
About The Author
Atul Mohan, PhD, is a data and AI executive whose work spans data science, marketing analytics, measurement strategy, and the translation of complex technical systems into business decisions.
