What a Media Plan Should Commit To in Writing

Most media plans arrive as a channel list and a calendar. Both are necessary and neither commits anyone to anything, which is how a flight runs exactly to plan and still leaves both sides arguing about whether it worked. A plan worth signing contains decisions, written down while they can still be changed.

Audiences, described rather than selected

"Interest targeting on Meta, keywords on Google Search" is not an audience definition. A plan should say how each audience is constructed and why that construction matches the person you are trying to reach: first-party lists and how they were assembled, on-platform behaviour, a lookalike and the seed that taught it, search intent split by query type, and broad delivery on the surfaces where the platform's own model genuinely outperforms a hand-built segment — which, on some of them, it now does.

It should also name the exclusions. Exclusions are where plans quietly go wrong: three overlapping audiences reach the same person from three directions, so the plan reads as breadth while the delivery is repetition.

Channels, with a job each

Every channel in the plan should have a stated job, phrased so a client could repeat it. Google Search captures intent that already exists. Paid social manufactures the intent that search later harvests. Programmatic holds a brand present against defined inventory between campaigns. TikTok and Snapchat reach people who behave differently enough from a Meta audience that the same creative cannot simply be reposted across all three.

A channel without a stated job cannot be evaluated. It can only be defended.

Creative volume, and the cadence that replaces it

Paid social does not fatigue because the idea got worse. It fatigues because the same people have now seen it often enough that it stops registering, and the platform — reading the same weakening response — shows it less. That is a production problem wearing a media problem's clothes, so the plan has to answer it in production terms: how many variants go live at launch, which single variable each one changes (the hook, the first frame, the length, the ratio, the caption, the call to action), when the next batch is due from the edit, and the signal that triggers the swap.

The signal belongs in writing rather than in somebody's judgement on the day. A falling click-through rate while frequency climbs is fatigue. A falling click-through rate at flat frequency is usually the creative or the audience, and those are different repairs.

Perspective slot — awaiting Ashraf's own words on how a flight is actually run week to week, and what a client should expect to read every Monday.

An optimisation cadence, and what may change inside it

Weekly beats monthly for a structural reason rather than an enthusiastic one: a weekly loop answers the question while the flight is still running and something can still be done about the answer. A report written after the flight answers it once nothing can be changed. So the plan should commit to a weekly written record — what was paused, what was widened, which creative entered rotation, which audience was cut, and what will be watched next week as a result.

Measurement, agreed before launch

• A naming convention and UTM structure, decided once, so reports can be assembled at all.
• The events that count as the outcome, defined in the same words on both sides.
• One attribution window everybody reads from, and a commitment not to move it mid-flight.
• Who holds account access, and where the data lives when the relationship ends.
• One report both sides argue from, rather than two that disagree.

Brand safety, decided in advance

Exclusion lists, placement opt-outs, the categories a brand refuses to sit beside, and who may approve an exception. Programmatic reach is only as good as the inventory it lands on, and that is a decision to make in the plan rather than in an incident.

None of this makes a plan longer to read. It makes disagreement specific, which is the only kind of disagreement that ever gets resolved.