Marketing communication: fund causal roles, not a list of channels.
Answer:
A communication mix must not rank channels on one poorly defined metric. Video, influence, search, CRM, retail media and public relations can create demand, encode a proposition, capture intent, reassure, convert or retain. Their value depends on role, audience, horizon and other activities. The design unit is the causal task assigned to each activity, with an early signal, economic outcome and proportionate evidence.
Method and decision:
Define the behaviour that must change, the population and comparator. Give every activity one primary and at most one secondary role, documenting dependencies such as video → branded search → conversion. Estimate incremental audience, effective frequency, cost per additional person and expected contribution, correcting duplicated reach and demand that would have been captured without the activity. Use creative tests, holdouts, geo-experiments and calibrated models; decide on both own effect and synergy with capture activity. Fund each activity for its contribution to the system, ensuring it covers creation, capture and conversion without excessive duplication.
Worked example:
A €600k plan contains €240k video, €180k search, €90k CRM and €90k retail media. Attributed contributions are €180k, €540k, €260k and €240k, but incremental estimates are €210k, €145k, €160k and €105k. An experiment finds that removing video reduces incremental search volume by 18%. System contribution with video is 210 + 145 + 160 + 105 = €620k. Video-search interaction is 145 × 18% = €26.1k, already included in €620k but isolated for the decision. Reducing video by €80k saves €80k but loses €70k own effect and about €8.7k interaction, a €1.3k net gain. Cutting €60k branded search loses €22k incremental effect and releases €38k net. Attributed ROAS would cut video; the scorecard retains a video base, cuts low-incrementality branded search first and invests €38k in a creative test.
Limits:
Cross-platform audience deduplication is often modelled. Delayed effects depend on window and distribution controls. A synergy observed at one pressure may disappear at another.