Something that happens in almost every performance-led marketing organisation that starts doing upper-funnel work.
The brand campaign runs. Afterwards, somebody has to report on it. The reporting infrastructure that exists is the performance one, so the campaign gets a D7 ROAS number. The number is low, because of course it is. And a campaign that may have worked gets filed as a failure with a decimal place attached.
The damage is not the bad decision on that campaign. It is that the number looked rigorous. Nobody argues with a measurement. They argue with opinions, and a number is not an opinion, so the verdict sticks.
Why the number is meaningless rather than just low
D7 ROAS answers a specific question: of the revenue attributable to this spend, how much arrived within seven days. It works for direct-response acquisition because the attribution chain is short, the intent is immediate, and the revenue lands fast.
Upper-funnel activity violates all three.
The attribution chain does not exist. Someone who sees a brand campaign and installs a fortnight later through a search or a store browse is not attributable to it by any mechanism you have. They will appear as organic, or worse, as attributable to whichever performance campaign happened to touch them last — which means your brand spend shows as zero while inflating the apparent performance of the channel that closed it.
The time horizon is wrong by an order of magnitude. Brand works on consideration, and consideration does not resolve in seven days.
Much of the effect is not direct at all. A large part of what upper funnel does is make performance marketing cheaper: better click-through, better install rate, lower effective cost per install, better conversion from the same audiences. That value appears entirely inside your performance numbers, where it gets credited to performance.
The size of that is easy to underrate. Open the calculator at its defaults with one change: cost per install at $1.00 instead of $1.10, standing in for that efficiency effect. Payback moves from day 207 to day 174. Every one of those 33 days shows up in the performance campaign's numbers and none of them in the brand campaign's D7 ROAS. Those are the calculator's published defaults with one input changed, not a customer result.
So a D7 ROAS of near zero on a brand campaign is not evidence the campaign failed. It is evidence you measured the wrong thing. Those are different findings and only one of them should change a budget.
The asymmetry that follows
This produces a predictable organisational bias.
Performance marketing has a measurable return. Upper funnel has a measurement that understates it. So budget flows to performance, not because anyone decided that was optimal, but because it is the only line with a defensible number attached.
Over time the organisation learns that upper funnel does not work. What it has actually learned is that upper funnel does not score well on a metric built for something else.
Pushing back on this with "brand is hard to measure" loses every time, because it sounds like special pleading from the side without evidence. The way out is not to argue the metric is unfair. It is to propose a different one and commit to it in advance.
What to commit to instead
Three things, agreed before the campaign runs rather than after it is scored.
Efficiency effects on the channels you can measure. If upper funnel is working, it should show up as improved click-through, improved install rate and lower effective cost per install on performance campaigns running against overlapping audiences. Those are measurable, they are in systems you already have, and they are where a large share of the value lands. Take a clean baseline before the campaign starts, because you cannot reconstruct one afterwards.
Organic and direct install volume against baseline. Crude, noisy, confounded by seasonality and store placement. Still better than nothing, provided you established the baseline first and you are honest about the confounds rather than claiming the whole delta.
Geographic or temporal holdout, if scale permits. The only method that gets you close to causal. Run the campaign in some markets and not others, compare. Expensive, often politically difficult, and the only one that settles an argument permanently.
None of these produce a ROAS figure. That is the point, and it needs saying out loud in the room where the budget is decided, before the money is spent.
The forecasting problem underneath
There is a harder version of this that the measurement conversation tends to skip.
You cannot forecast upper funnel from priors you do not have. If the organisation has not run this kind of activity before, there is no historical performance to train on, and the sophisticated approaches — mixed media modelling, incrementality testing, natural experiments — all need history you have not accumulated yet.
Which leaves you building a forecast from comparables and a chain of assumptions: impressions, into reach, into consideration, into installs, with a conversion rate at each step that is an estimate.
That is a legitimate thing to do, and it is worth being clear that it is what you are doing. The failure is not using assumed conversion rates. It is presenting the output as though it carried the same confidence as a performance forecast built on observed data.
State the assumptions, name the ones the answer turns on, and commit to what you will measure to find out whether they held. That is defensible. It also starts building the priors that make the next forecast better, which is the only way out of the trap.
Running it
Kadar models the full chain from spend through cost per install, installs, conversion, retention and revenue, with a ranked sensitivity showing which assumption is actually driving the result. Which is what you need when the inputs are estimates: not more precision, but clarity about which estimate the answer depends on.
Run your own numbers in the free calculator. No sign up.