How a Nobel-Winning Stock Investing Framework Optimizes Ad Spend

The practical implication of applying stock-portfolio logic to ad spend optimization is not that marketers should pretend campaigns are stocks. It is that the same allocation problem keeps showing up in a different costume: one asset has higher expected return and higher volatility, another has lower immediate return and different movement, and the person holding the budget has to decide whether maximizing the next-period return is worth the concentration risk.

That is the part of the brand-versus-performance debate worth keeping. If brand and performance returns do not move together, a mixed portfolio can beat a performance-heavy portfolio on risk-adjusted return even when performance has the higher raw ROAS. The argument does not need softer language about balance. It needs expected return, variance, covariance, and a correlation matrix that everyone in the room can inspect.

Product Philosophy’s external MMM benchmark work puts the brand-performance return correlation at about 0.14, with an illustrative worked example showing brand at 4.8x ROAS with 1.9x standard deviation, and performance at 8.8x ROAS with 3.9x standard deviation.[1] Those are not Signal & Convert campaign results. They are useful because they show the shape of the decision: performance looks stronger if the only column is ROAS, but it also moves around more.

The Useful Translation From Stocks To Media

Modern Portfolio Theory is useful here for one reason: it gives budget owners a way to separate return from the reliability of that return. A media channel can be treated like an asset if the team can estimate three things: what it tends to return, how much that return varies, and how it moves relative to other channels.

The correlation number is doing more work than the finance vocabulary. A 0.14 correlation means brand and performance returns are only weakly related in the benchmark set Product Philosophy analyzed.[1] They are not opposites. They are not interchangeable. They are different enough that combining them can reduce portfolio volatility without giving up all of performance’s higher expected return.

That is the diversification mechanism. If paid search, retargeting, conversion-optimized social, reach media, video, and other brand activity all rise and fall together, a portfolio model adds little. It just gives a fancier name to concentration. If they move differently, the budget mix can become more stable than any single channel would suggest.

Why The Highest ROAS Portfolio Can Still Be The Weaker Portfolio

A 100% performance allocation can have the highest absolute ROAS in the Product Philosophy example and still produce the worst risk-adjusted result once volatility is included.[1] That is the budget meeting problem in one sentence. Finance sees a higher return column. The channel owner sees a defensible near-term number. The CMO sees the platform risk, auction risk, and demand-harvesting ceiling that are not visible in a raw ROAS ranking.

The portfolio version of the discussion asks a sharper question: for each extra point of expected return, how much instability does the team accept? In finance, the Sharpe Ratio answers that by comparing return against risk. In media, a marketing Sharpe Ratio can do the same job if the return estimates come from credible incrementality or MMM work rather than platform-reported attribution alone.

This matters because performance media often looks cleanest exactly where it is least diversified. Last-click or platform-reported ROAS may reward the channel closest to conversion. A portfolio view does not deny that the channel produces return. It asks whether the whole account becomes too dependent on conditions the marketer does not control: bidding rules, auction density, tracking changes, ranking systems, creative fatigue, and the amount of existing demand available to capture.

The cleanest use of the framework is not to punish performance. It is to stop treating a volatile 8.8x return and a steadier 4.8x return as if the only rational act is to move every dollar toward the larger number. If the lower-return asset has different movement and lower volatility, it may improve the portfolio even while reducing the arithmetic average ROAS shown in a simple channel table.

Where The Efficient Frontier Lands

When Product Philosophy plots the marketing efficient frontier using its benchmark assumptions, the maximum-Sharpe allocation converges around a 55–65% brand and 35–45% performance range.[1] That is the part worth taking seriously, and also the part worth not over-selling. The point is not that every advertiser has been handed a universal split. The point is that, under a weak brand-performance correlation and materially different volatility profiles, the math stops favoring a performance-only budget.

The more interesting fact is that this lands near Binet and Field’s widely cited 60/40 recommendation from 996 IPA effectiveness case studies, a separate empirical route summarized in Product Philosophy’s analysis.[1] One path starts from portfolio mathematics. The other starts from advertising effectiveness history. They meet in roughly the same zone.

That convergence is cross-validation, not a law of nature. The IPA evidence base is heavily associated with large advertisers and classic effectiveness cases, including many packaged-goods contexts. A B2B software company, a cash-constrained DTC brand, a marketplace with strong network effects, and a local services advertiser should not assume the same optimum before testing their own data.

Still, the agreement is hard to dismiss as coincidence. The portfolio model does not begin with a reverence for brand. It begins with return, volatility, and correlation. If that route arrives near the empirical 60/40 finding, the burden shifts. The performance-heavy plan now has to prove that its extra raw ROAS compensates for the concentration risk it creates.

What AI Changes, And What It Does Not

AI makes the allocation problem more frequent, not automatically more disciplined. Madison and Wall reported that AI-powered ad spend reached $57 billion with 63% year-over-year growth in March 2026.[2] That is a large pool of money being pushed through systems that can reweight bids, audiences, creative variants, placements, and campaign types faster than most weekly planning meetings can explain.

The uncomfortable part is that automation can optimize inside a bad portfolio. A bidding system may become excellent at finding marginal conversions within a narrow performance pool while the account as a whole becomes more exposed to the same demand signals, same attribution windows, and same platform shocks. Portfolio optimization belongs above the bidding layer. It decides the risk budget before the machine starts fighting for impressions.

There is promising technical work in automated budget allocation, but it should not be made to carry the 60/40 argument. Sony’s AAMAS 2025 combinatorial-bandit study reported 19% more clicks than human operators in experiments on Japanese ad platforms.[3] That is useful evidence that algorithmic allocation is advancing. It is not evidence that clicks are the right business objective for every advertiser, or that the result generalizes across regions, verticals, and platform mixes.

The better connection is architectural. Bandits and reinforcement-style systems can help adjust budgets dynamically once the objective and constraints are defined. Modern Portfolio Theory helps define the portfolio-level tradeoff those systems should respect. Without that layer, the optimizer can become very fast at moving dollars toward the measurement system’s blind spots.

The Calculation A Media Team Actually Needs

The model is not complicated in concept. The hard part is feeding it data that deserves the confidence implied by the output. A team needs a return estimate for each channel or campaign class, a volatility estimate around that return, and a cross-channel correlation matrix. With those, it can simulate different budget weights and find the mixes that sit on the efficient frontier.

• Estimate expected incremental return by channel or campaign class, preferably from MMM or incrementality work rather than platform attribution alone.
• Calculate return variance or standard deviation over comparable periods, so high-return channels are not treated as equally reliable by default.
• Build a correlation matrix showing how channel returns move together, especially across brand and performance groupings.
• Test budget weights across the feasible range, including current allocation, 100% performance, 100% brand, and blended mixes.
• Plot expected return against volatility and identify the efficient frontier rather than debating every possible budget mix.
• Compare candidate allocations using a marketing Sharpe Ratio or equivalent risk-adjusted measure before moving money.

For a team already running MMM, the incremental work is mostly standardization. The team has to decide whether the asset classes are individual channels, campaign objectives, funnel stages, geographies, or some combination that still produces stable estimates. Too much granularity creates noisy assets. Too little hides the correlation structure that makes the model useful.

The threshold in the Product Philosophy framework is 24 months of reliable MMM data.[1] That is a real barrier. Twelve months can confuse seasonality with channel behavior. Sparse spend shifts can make the model mistake budget policy for market response. A major product launch, pricing change, supply issue, or measurement migration can contaminate the return history if the model treats the period as normal.

This is where the finance metaphor earns or loses its keep. If the team cannot explain the return estimate, the variance estimate, and the correlation estimate, it should not present the frontier as a decision rule. It can still use 60/40 as a benchmark. It should not call the benchmark an optimized budget.

When Not To Use The 60/40 Output As A Command

There are cases where the framework is directionally useful but operationally premature. A new brand with limited history may not have enough variation in spend and outcomes to estimate channel-level variance. A company entering a new geography may inherit no reliable correlation structure. A business with extreme short-term cash constraints may accept higher portfolio risk because survival depends on nearer-term payback.

There are also measurement-specific reasons to slow down. If brand activity is under-measured, its expected return will be understated. If performance channels receive too much credit through attribution leakage, their expected return will be overstated. If the same demand shock lifts every channel at once, correlations can look more favorable or less favorable depending on how well the MMM controls for external conditions.

That does not make the model fragile. It makes it auditable. A useful allocation framework should expose which assumptions are carrying the decision. If a finance lead wants to cut brand because its raw ROAS is lower, the model shows what happens to volatility and portfolio concentration. If a brand lead wants a larger long-term allocation, the model requires a return estimate and a covariance argument, not a mood.

Teams building this layer should first tighten the data requirements behind AI-driven performance measurement. Signal & Convert’s guide to data prerequisites for AI-driven ad performance is a useful companion for deciding whether the inputs are ready. The AI marketing analytics reference guide is the more practical place to connect MMM, platform reporting, and decision logs before turning the math into a budget recommendation.

The Budget Decision This Framework Improves

The strongest version of this approach is modest and useful: with 24 months of reliable MMM data, channel-level return and variance estimates, and a credible correlation matrix, a media team can reproduce the calculation and test whether its current budget sits on or below the efficient frontier. If the maximum-Sharpe region lands near 55–65% brand and 35–45% performance, the team has a mathematically grounded reason to challenge a performance-heavy allocation.[1]

Without those inputs, 60/40 is a serious benchmark, not a budget command. It is a better starting point than the usual argument by anecdote, but it still has to survive the advertiser’s own category, margin structure, sales cycle, measurement quality, and platform exposure.

The immediate work is verification. Pull the internal analytics, inspect the benchmark records, and compare platform-specific claims against independent measurement. Signal & Convert’s benchmarks on Google AI advertising and its work on the transparency premium in AI ad platforms are inputs to that posture, not substitutes for the advertiser’s own model. The external ROAS and variance figures cited here belong to Product Philosophy’s benchmark example, and the useful question is whether your own data produces the same frontier.

References

1. Brand vs. Performance: A Portfolio Optimization Framework Using Markowitz Theory for Marketing Budget Allocation — Product Philosophy.
2. AI-powered ads driving US ad growth — Business Insider, March 2026.
3. Adaptive Budget Optimization for Multichannel Advertising Using Combinatorial Bandits — arXiv 2502.02920.