Reach Planning and Cross-Platform Frequency Control

The frequency number that matters is the one counted across every platform at once, and no single DSP can produce it. A cap of three per week in DV360 sitting next to a cap of three in The Trade Desk is not a cap of three; the same household sees both, plus whatever social and video are adding on top. Rebuilt from impression logs, the real distribution in campaigns we inherit is almost always more skewed than the plan assumed, with a small share of reached devices absorbing a large share of delivered impressions while the tail sees one and forgets it. How we count: We pull impression-level logs, DV360 Data Transfer files and The Trade Desk's raw event stream, into BigQuery and rebuild reach and frequency on the identifier each platform actually writes into the log, then model the household view rather than asserting one. That produces a reach curve for the campaign as bought, the marginal cost of each additional point of reach, and the point at which more money starts buying repetition instead of people. That point is the planning decision. Everything downstream of it is arithmetic. What we will not claim: Deterministic deduplication across walled gardens. Meta and YouTube do not export identifiers that would make it possible, and anyone showing you one deduplicated reach number spanning open web and social has modelled it. Where that question is worth paying to answer, we say so and use panel-based measurement rather than pretending log data reaches further than it does. A number with its method attached is more useful in a board meeting than a cleaner one you cannot defend when asked. Capping where it binds: Caps set only at insertion-order level leak whenever the same deal sits under more than one insertion order or more than one platform. We set caps at the level the buy is actually structured, split them between prospecting and returning audiences, and set them per creative where a single execution wears out faster than the campaign around it. Where the product has a decision window worth respecting we cap on time between exposures rather than on count alone, because nobody signs on an apartment because they saw the ninth banner. Seasonality in our categories: Sport and fitness reach is bought against a calendar rather than a month: a season opening, a tournament window, the first fortnight of January in gym membership. Property launches concentrate reach into the two weeks around an opening and should sit close to flat afterwards. Industrial buys run the opposite way, since a buying committee of a few thousand people needs presence across a long evaluation, and a heavy cap there spends the budget on people who were never going to be in market that quarter. Who this is not for: Campaigns with no route to log-level data on at least one platform, where this becomes guesswork with a spreadsheet attached. Direct response programmes where reach is not an objective and frequency is only a cost control. Buyers who want a single effective-frequency number to plan against; that number does not survive contact with a reach curve, and we will not supply one because the previous agency did. What you receive: A reach and frequency model built on your own logs, the curve with the marginal-reach point marked and the assumptions written beside it, a capping structure implemented across your seats, and the query set left in your own BigQuery project so your analysts rerun it next quarter without us. Where a claim is modelled rather than observed, it is labelled as such in the document.