Frameworks & methodology for integrated media planning

Next quarter's budget arrives as one number in one cell. Eleven lines want it: branded search, non-brand search, paid social, programmatic video, retail media, TV, out of home, podcasts, sponsorship, email, events. Someone wants the split justified by Friday, and last quarter's CPAs will not justify it, because a CPACPACost Per Acquisition: the total cost to generate one customer or conversion, computed by dividing total spend by the number of acquisitions.View full definition → describes money you already spent, not money you are about to move. Three things settle the argument: the shape of each line's response curve, how long its effect lasts, and a defensible ratio between spend that builds future demand and spend that harvests today's. This lesson is that method. It assumes the vocabulary the foundations lesson earns (the plan as one coordinated object, and the job each channel does inside it) and goes straight to the arithmetic.

Sub-Concept 1: Start from the response curve, not the channel list

Media response is non-linear. Every line has a region where extra money does almost nothing, a region where it works hard, and a region where it saturates. Econometric or marketing mix modelling (MMMMMMA statistical approach that estimates how each marketing channel and other factors drive sales, guiding budget allocation.View full definition →) estimates that curve by regressing weekly sales against weekly media alongside price, distribution, seasonality, competitor activity and anything else that moves the business. It needs history: two to three years of weekly data, so roughly 104 to 156 observations, and genuine variation in each media line.

Two outputs drive allocation. Adstock, the decay rate, says how long an exposure keeps working: direct response search behaves with a half-life measured in days, while brand video and out of home commonly model with half-lives of several weeks and tails running past a year. Marginal ROI says what the next pound does at your current spend level, which is the only figure a reallocation decision depends on.

The optimisation rule is unglamorous. Move money until marginal returns equalise across lines. If non-brand search returns £2.10 on the next pound and paid social returns £1.40, you shift from social into search until the two converge, then stop. Average ROIROIReturn on Investment: the ratio of net profit to the cost of an investment. A 300% ROI means each dollar invested returns $3.View full definition → plays no part: a line with an average ROI of 8 can have a marginal ROI below 1 once you have pushed it into saturation.

Sub-Concept 2: What the model cannot see

Three limits decide whether you can trust the split you just built.

• A channel spending under about 5 percent of total budget usually cannot be estimated separately from noise. Neither can a channel running always-on at flat weekly spend, because a variable with no variance has no coefficient. If you want a read on it, you have to create variation deliberately.
• Channels flighted together are collinear. If TV and OOH always launch in the same week, the model splits their credit arbitrarily and both confidence intervals go wide.
• MMM typically assigns most of the sales to base rather than media. Long-run brand effect sits inside that base, so a model fitted on a 13-week window systematically under-credits brand building and hands the performance team a win by construction. Fit over 52 weeks minimum, and expect confidence intervals of plus or minus 20 to 30 percent on channel ROI. Differences smaller than that are not findings.

Sub-Concept 3: The brand versus performance ratio

Binet and Field's analysis of roughly 1,400 cases in the IPA databank landed on about 60 percent brand building and 40 percent activation as the long-run efficient split. That average moves by category: higher brand weight for low-involvement, high-frequency purchases, and closer to 50:50, or the 46:54 their later B2B work suggested, for considered purchases with long cycles and small buying groups.

The companion calculation is excess share of voiceshare of voiceYour brand's share of total advertising or conversation volume in your category, measured against competitors over a defined period.View full definition →. Take your share of category media spend, subtract your market sharemarket shareThe percentage of total industry sales your company captures in a given period. It measures competitive position relative to rivals in a defined market.View full definition →, and the gap predicts growth: roughly 10 points of ESOV buys about half a point of market share per year. Run that in reverse and it becomes a budget floor. A brand at 12 percent share wanting a full point of growth needs share of voice near 32 percent, which in a category where the leader outspends you three to one is a plan you either fund or abandon. It is also why challenger brands that hold spend flat and blame creative are usually misdiagnosing arithmetic.

60:40 is a long-run average, not a quarterly quota. In a launch quarter, or when supply is constrained, forcing the ratio is misuse of the finding.

Sub-Concept 4: Triangulating so the split survives the CFO

How you measure decides what you buy, so the measurement stack needs a division of labour. MMM sets the strategic split between brand and performance and between large channels. GeoGeoThe practice of making your brand and content visible and citable inside AI-generated answers from tools like ChatGPT, Gemini and Perplexity.View full definition → holdout experiments calibrate it: split comparable markets, dark one channel for six to eight weeks, read the incremental difference, then force the model's coefficient to match. Platform and multi-touch numbers are for in-flight optimisation only; Apple's ATT change in April 2021 and the erosion of third-party cookies removed the user-level data that made MTAMTAA method that distributes conversion credit across all marketing touchpoints in the customer journey, rather than crediting only the first or last interaction.View full definition → coherent in the first place. Data-drivenData-drivenAn approach where decisions are systematically informed by data analysis rather than intuition alone.View full definition → attribution in GA4 is a better default than last click, which defunds every channel that created the demand branded search then books.

Google's Meridian and Meta's Robyn are both free, open-source MMM implementations, and both come from companies that sell the media the models are asked to evaluate, which is a reason to bring your own priors rather than accept theirs. Nielsen and the other measurement vendors sell modelling as a service; the same caution applies to anyone whose fee scales with the complexity of the answer.

Real-World Cases

Case 1: Adidas rebalancing after the attribution trap

In 2019 Adidas' then global media director, Simon Peel, said publicly that the company had over-invested in performance marketing: around 77 percent of budget to performance and 23 percent to brand, when its own econometric work indicated brand activity was doing most of the sales work. The interesting part is the mechanism. Nobody decided on 77:23. Attribution reporting made performance spend legible and brand spend illegible, so every quarterly budget meeting arrived with evidence on one side only, and the ratio drifted through repeated small in-quarter raids. Adidas' 2021 "Own the Game" plan committed to stepping brand investment back up, which is a multi-year repair, not a reallocation.

Case 2: Diageo and modelling as an internal capability

Diageo invests on the order of £2bn a year in marketing, roughly a sixth of net sales, across a couple of hundred brands and around 180 markets. At that spread the useful question is not "TV or digital" but "does the next £5m go behind Guinness in Britain or Johnnie Walker in India", which only an in-house econometric capability with a common method can answer, since outsourced one-off studies are not comparable to each other. Spirits also show the confound cleanly: price promotion and distribution gains move volume hard, so a promo calendar left out of the model loads onto whichever media line happened to run alongside it. Add low purchase frequency and equity built over decades, and any 12-week read on a Diageo brand is close to meaningless.

Case 3: When 60:40 is the wrong answer

If your addressable market is 300 named accounts, ESOV logic does not apply, because you are not buying category-wide coverage; you are buying depth against a list you can enumerate. Brand money there goes into fewer, heavier, higher-quality impressions and into category-defining content, not broad reach. The other edge case is capacity. If you cannot fulfil demand, the marginal ROI on activation is negative while the marginal value of brand building is unchanged, so the correct move during a supply shortage is the opposite of the instinctive one.

CMO Action Items

• Ask your analytics team or vendor for marginal ROI with confidence intervals at current spend for your five largest lines, not a single average ROI per channel. If they can only supply averages, they cannot advise on reallocation.
• Fix the read window before the debate starts. Agree on 52 weeks rather than 13, in writing, or performance wins the argument by measurement design.
• Put a brand floor in the budget as a protected line with its own owner. Drift to 23 percent happens through small raids, never through a decision anyone signed.
• Book one geo holdout a year on your largest channel, six to eight weeks minimum, and pre-register which result would change the split. A test with no pre-committed consequence is a report.

Common Mistakes That Kill Results

Mistake 1: Optimising average ROI. Branded search commonly reports the highest ROI in the deck and the lowest incrementality in the test, because it books conversions that other spend created. Brand-search pause tests routinely show a large share of those clicks arriving organically anyway. Judge every reallocation on the next pound.

Mistake 2: Treating model output as measurement. MMM is an estimate with error bars, and omitted variables (price, distribution, weather, a competitor's launch, a major sporting event) attach themselves to whichever media line moved with them. Any coefficient that has not survived an experiment is a hypothesis.

Mistake 3: Reading a flat quarter after a brand cut as proof the cut was free. Decay is slow, so the damage surfaces two to four quarters later as rising CPAs, falling branded search volume and weaker conversion rates, by which point nobody connects it to the cut. The second-order cost is worse than the saving: you end up rebuilding awareness from a lower base while paying auction prices for demand you are no longer creating.