Every ad platform grades its own homework, and every one of them gives itself an A. The question that matters is not what the campaign reported. It is what would have happened if the campaign had never run.
Why reported ROAS is the wrong basis for a budget decision
Platform-reported ROAS is the most confidently wrong figure in marketing. It is
not fraud; it is a structural artifact. Each platform can only observe the
conversions it touched, so it reports on those and stays silent about the
counterfactual. Run the exercise of summing reported revenue across all your
channels and comparing it to the P&L, and the gap tells you exactly how much
double-counting is baked into your optimization decisions.
The consequence is not academic. Budgets get shifted toward whichever channel is
best at observing conversions — typically branded search and retargeting,
which mostly intercept demand that already exists — and away from the channels
that created the demand in the first place. The dashboard improves. The business
does not.
What I do instead
Establish causality before allocating. Holdouts and geo tests are cheap
relative to the budgets they govern. A test that costs a fraction of a quarter’s
spend and reveals that a channel is half as productive as reported pays for
itself immediately.
Allocate on contribution, not revenue. Revenue-based ROAS targets quietly
push spend toward high-revenue, low-margin products. Allocating on contribution
margin changes which campaigns look like winners, sometimes dramatically.
Model the whole mix, including what cannot be clicked. CTV, video, audio,
and out-of-home (OOH) do real work that click-based systems cannot see. Modeling them
alongside the trackable channels is the only way to stop underfunding the top of
the funnel by default.
Keep the plan executable. Measurement that does not change a buying decision
is a research project. Every read should terminate in a specific action: shift
this budget, kill this line item, extend this test, raise this bid floor.
Where the work usually starts
Most engagements begin with the audit, because it is the fastest way to separate
the three failure modes that look identical from the outside: an account
executed badly, an account measured badly, and an account doing fine inside a
strategy that was wrong to begin with. Those need entirely different responses,
and guessing between them is how quarters get lost.
From there the sequence is typically: fix the obvious execution problems, stand
up a measurement design you can defend, run the first tests, and then rebuild
the allocation on what the tests actually said.
Player-coach, not slide deck
I still build the media plans, sit in the accounts, and produce creative. That
is deliberate. A media strategy written by someone who has not looked at the
account in a year is a hypothesis, and the difference between a hypothesis and a
plan is whether the person writing it knows what will break.
Frequently asked questions
Why is platform-reported ROAS misleading?
Because it is a measure of correlation presented as a measure of cause. A platform counts a conversion it can associate with an impression or click it served, which means it takes credit for demand that already existed, for buyers who would have purchased anyway, and frequently for the same conversion a second platform is also claiming. Add up the platform-reported revenue across a mature account and it routinely exceeds what the company actually booked. That is the tell.
What is incremental ROAS and how do you measure it?
Incremental ROAS is the return on the revenue that would not have occurred without the spend. You measure it by creating a comparison — a geographic holdout, a matched-market test, an audience-level suppression, or a platform lift study — and reading the difference between exposed and unexposed groups. It is more work than opening a dashboard, and it is the only number that survives a conversation with a CFO.
Do I need media mix modeling, incrementality testing, or attribution?
They answer different questions and work best layered. Attribution is a fast, directional read for in-flight optimization, and it should never be the basis of a budget decision on its own. Incrementality testing gives you causal truth about a specific channel over a specific window. Media mix modeling gives you a top-down allocation view across everything, including channels with no click at all, like OOH and CTV. The right structure uses tests to calibrate the model and the model to allocate.
Can you work across upper-funnel channels like CTV and out-of-home?
Yes, and they are usually where the measurement conversation gets honest. Those channels have no click to hide behind, so they force you into geo-based testing and modeling rather than last-touch storytelling. Brands that can measure upper funnel properly tend to find they were underfunding it, because the channels that generate demand rarely get credit from systems designed to observe the moment demand converts.
How do you use AI in media analysis without letting it run the budget?
I build agents to do the parts that are labor: pulling and normalizing reporting, flagging anomalies, running forecast scenarios, and preparing the modeling inputs. Each one gets guardrails and defined parameters. What the agents do not do is decide. There is always a human on top with the context to catch drift before a dashboard reports it, and to use the agents to validate a decision chain rather than outsource it.
What does a paid media audit actually look at?
Account structure, audience overlap and cannibalization, creative volume and fatigue curves, bidding and budget pacing, landing experience, and then the layer most audits skip: whether the measurement being used to justify the spend can support the claims made on it. A large share of “underperforming” accounts are performing fine and being measured badly, and a share of “winning” accounts are harvesting demand they did not create.