The art of analytics is about how to ask the right questions. The science of advertising is about how to get evidence-based, honest answers. The work is applying both to a client’s campaigns to achieve A-level outcomes.
The scorecard most agencies are graded on
The ANA’s 2023 study of the programmatic supply chain put the agency’s job as a matrix: pay low prices, get high ad quality, and avoid the corner where you pay high prices for low quality, the “winner’s curse.”
After the ANA, Programmatic Media Supply Chain Transparency Study: Complete Report, December 5, 2023, page 120.
It is true as far as it goes, and it is more complicated than the matrix suggests. Price and quality are inputs. The matrix doesn’t ask the question a CMO is paid to ask: how do you maximize the outcome?
The work is painstaking, and it is not the point
Buying well is hard, unglamorous work: chasing inventory across open, guaranteed, preferred and direct, blocking supply paths and unblocking them when they change, re-trafficking creative, calling publishers. We told one story about it in 2024, about buying connected TV from a publisher that turned out not to fit the plan, as a board game with snakes and ladders. It is on the blog as A Tale of Gouda.
That sort of work is what makes an A-level media agency. But working hard only matters when you have prioritized the right goals. The first decision on any campaign is its north star: a controlled brand study for awareness or purchase intent, or cost per acquisition where an acquisition can be measured. Everything else on this page follows from that choice. A handful of examples of what the evidence means.
Find the frequency that moves the needle, then count the people moved
Lumen, Brand Metrics and Havas lined up 9,000 brand lift studies against viewable frequency and total attentive seconds. One exposure moves almost nothing. Impressions with under half a second of attention are negative at every frequency. Lift starts at the third exposure once the ads have held a person for more than a second, and peaks around four or five.
| Total attentive seconds across all impressions | |||||||
|---|---|---|---|---|---|---|---|
| Viewable frequency | 0 to 0.5 | 0.5 to 1 | 1 to 1.5 | 1.5 to 2 | 2 to 3 | 3 to 5 | 5+ |
| 1 | -1.6% | -1.3% | -1.8% | -1.4% | -1.5% | -0.5% | |
| 2 | -1.4% | 0.5% | 0.3% | -0.1% | -0.1% | -0.1% | 0.3% |
| 3 | -1.6% | 0.6% | 1.5% | 1.1% | 0.9% | 0.8% | 1.3% |
| 4 | -3.0% | 0.7% | 2.0% | 3.1% | 2.2% | 2.4% | 1.7% |
| 5 | -1.2% | -0.3% | 2.6% | 2.2% | 2.9% | 2.8% | 2.7% |
What it means. The frequency that moves awareness is a number you find, campaign by campaign, not a number you assume. And once you have the lift, the figure to report is the one a CFO can picture: people moved, which is the people reached times the absolute lift. A brand study at high confidence with a 20% relative lift in aided awareness among 100,000 people reached, from 25% aware to 30%, moved 5,000 people. That is what the campaign bought.
If cost per acquisition is high, invest in awareness. If it is low, go down the funnel.
Google's 2024 work with WARC and Nielsen on media returns found that a 1% rise in brand awareness goes with a 0.4% rise in short-term sales and a 0.6% rise in long-term sales, that returns in the first four months of a media investment equal the returns over the following twenty, and that the growth-maximizing split lands around 50 to 60% brand and 40 to 50% performance. Google sells awareness media, so read it with that in mind. The direction agrees with Binet and Field's independent work on the IPA effectiveness cases, and with what we see in client accounts: when the brand work stops, cost per acquisition drifts up a quarter or two later.
Where the return on a media investment lands
What one point of brand awareness is worth
Google, WARC and Nielsen, 2024. A 1% rise in awareness, and the return on a media investment over 24 months.
What it means. A high cost per acquisition is usually an awareness problem, and the fix is upstream. A low one says the demand is there, and more budget can go to harvesting it. The split is set from your real economics, not a ratio, and the job of the agency is to help the CMO build the investment plan and the time horizon that the whole leadership team signs up to, then make it perform.
Source: Google, WARC and Nielsen, Beyond the Horizon: the holistic path to measuring media investments, October 2024; Binet and Field, The Long and the Short of It, IPA.
Optimize to the right distance from the money
A platform optimizes toward whatever event you give it, and it needs on the order of fifty a month to learn from: Google asks for about thirty conversions a month before its target-CPA bidding is worth turning on, and Meta for about fifty optimization events a week per ad set before the learning phase ends. Give it page views and it finds people who view pages. Give it a sale that happens twelve times a quarter and it learns nothing. Conversions built on page URLs fire on the wrong pages, and one order reported by every platform that touched it is several orders.
Events a month, for one brand, and which ones a platform can learn from
Illustrative counts for one brand, on a log scale. Thresholds: Google Ads, About Target CPA bidding; Meta, About the learning phase.
What it means. Don't optimize to things far from the money, and don't optimize to things so close to the money that you can't measure them. Between the two sits a proxy: a qualified lead, a funded application, a scheduled consult that usually becomes a patient. Know which proxies are good, which are bad, and check which page actually fires each one before money runs against it.
Decisioning power: get more of it, and cut inventory only for a reason
An algorithm is only as good as the choices it has. Two DSPs each see half the data and bid against each other. Twenty narrow line items each starve. A health brand cleared on three publishers can't reach the people the fourth would have found. The number of places the algorithm can choose among, and the signal it gets from each, is one of the strongest drivers of results we see.
The same 240 conversions a month, divided among decision-makers
Each algorithm learns only from the conversions it sees. Split the same signal six ways and every piece falls under the fifty a month that learning needs, and the two DSPs bid against each other for the same impressions on top.
What it means. Do what it takes to give the buy more decisioning power: one platform unless there is a reason for two, fewer and larger campaigns, creative approvals cleared with every publisher in advance, targeting as wide as the objective allows. Cut inventory out of a buy only when there is a real reason, and validate the targeting decision with the measurement afterward rather than with an assumption beforehand. Where we buy says which platform usually carries which objective.
Branded keywords are defensive, and a limited part of a buy
The cleanest evidence is eBay's field experiment, published in Econometrica in 2015: when eBay paused its paid search ads on its own brand terms, the organic result absorbed almost all of the clicks, and total brand traffic barely moved. Paid brand clicks mostly substituted for free ones. The exception is when a competitor is bidding on your name or your organic result is not first; then the paid result defends traffic you would otherwise lose.
What it means. Branded keywords are defensive in nature and should be a limited part of a buy, sized by a test of your own account rather than by habit. Brand search volume is more useful as a reading: a connected-TV flight that works shows up as people typing your name.
None of this is proprietary. What is ours is holding to it when a dashboard, a vendor or a quarter’s pressure says otherwise. How we measure is the other half of the sentence.