Equal impressions do not mean equal delivery
One million impressions could reach 500,000 people twice or 100,000 people ten times. Those delivery patterns need not produce the same response. For video planning, a model that can distinguish reach and frequency may answer a different question from a model that sees only spend or impressions.
- Choose a reporting window
- Reconcile reach and frequency
- Model the response
- Translate exposure to cost
Start by deciding the reporting unit and time window. Reach is deduplicated within a defined population and period; summing weekly reach does not generally produce monthly unique reach. The Meridian guide introduces a workflow that can use reach and frequency inputs.
A reproducible workflow makes data checks, modelling assumptions and decision sensitivity visible. The mmm tools, code and platforms learning path connects this idea to the assumptions, practical checks and next steps.
To work through the connected decisions, read Paid social MMM: distinguish saturation from creative fatigue and MMM saturation curves: know when you are extrapolating.
Build a reconciled delivery table
For each modelled geography and week, document spend, impressions, reach and average frequency, including platform definitions and missing values. The reach and frequency guide explains the underlying model; this audit checks whether your inputs actually share those definitions. As a basic consistency check, reach multiplied by average frequency should be compatible with impressions under the provider's definitions. Differences may reveal rounding, coverage or reporting-window mismatches.
Do not fill missing reach with zero when ads ran. That asserts no audience was reached. Preserve missingness, investigate the source and choose a documented treatment. If reporting definitions change halfway through the history, treat it as a data break rather than a marketing effect.
Work through a planning comparison
Imagine two hypothetical plans with the same one million impressions. Plan A reaches 500,000 people at frequency two; plan B reaches 250,000 at frequency four. Neither is universally better. The response depends on the audience, creative, objective and how additional exposures contribute.
A fitted frequency curve should be interpreted within the supported range. If the historical frequency was almost always between two and three, an apparent optimum at ten is largely an extrapolation. Show the observed range beside the recommended plan and compare uncertainty across plausible curves.
Translate the model back into an executable budget
An exposure recommendation is not yet a budget. Auction prices and audience availability determine what the proposed reach and frequency cost. State the cost assumptions and any caps before comparing plans. A change in targeting can alter both price and response, so the old curve may no longer apply.
Use the budget allocation guide to constrain the proposal and the saturation guide to interpret diminishing response. Then validate a small change using a suitable experiment or monitored rollout.
For review, include the provider definitions, reconciliation checks, supported frequency range, cost conversion and the uncertainty around the suggested plan. This keeps a sophisticated input format from creating a false impression of precision.
Sources and further practice
This article’s worked example is illustrative, not a measured client result. Use the linked methodology to check the assumptions before applying it.