media-planning

Media Planning (budget allocation across channels)

Allocate on marginal return, not average

The only defensible split equalizes marginal return across channels: keep moving the next dollar to whichever channel returns most at its current spend level, until the returns equalize or the payback bar is hit. Average CPI per channel is the wrong input — it always flatters the small channel.

marginal CPI ≈ Δspend / Δinstalls   (measured between adjacent spend levels of the same channel)
scale ceiling = the spend beyond which marginal CPI exceeds the breakeven CPI (see unit-economics)

Every channel saturates. A plan that scales a channel linearly from its current spend is the single most common planning error — the CPI curve bends upward well before the budget runs out.

The plan itself

A media plan is a table, not a paragraph. One row per channel × month:

| channel | month | spend | assumed CPI | expected installs | basis of the CPI assumption | confidence |

• Every CPI assumption must name its basis: last quarter's actual at comparable spend, a test result, or a vendor estimate (mark vendor estimates as unverified).
• Sum to the actual budget and reconcile: plans that don't add up get discovered in month two.
• Show a downside scenario (CPI +20–30%, or the channel not scaling past X) with what gets cut first. A plan without a downside branch is a forecast, not a plan.

Reserve, pacing, seasonality

• Test reserve: hold back roughly 10–20% for new channels/creative/geos. Without it the plan can only ever re-buy last quarter's mix, and the mix decays.
• Pacing: front-load learning (new campaigns need volume to exit the learning phase), then flatten. Month-end spend dumps buy the worst inventory of the month.
• Seasonality: Q4 and local holiday peaks inflate CPM materially; a plan built on Q2 CPMs under-delivers in Q4 at the same budget. Plan the same installs at a higher CPI, or shift timing.

Channel roles differ — don't compare them on CPI alone