How Much to Bet on Marketing: A Capital Allocation Framework

Executive Summary

Most companies set marketing budgets using rules of thumb -- a percentage of revenue, last year's number plus 10%, or whatever the loudest voice in the room demands. We analyzed 1,221 Marketing Mix Models across $1.85B in ad spend and found that companies using systematic capital allocation frameworks outperform rule-of-thumb allocators by 2-3x in marketing efficiency. This article introduces a four-step framework borrowed from quantitative finance -- using Kelly Criterion logic, risk-adjusted returns, and portfolio theory -- to size marketing bets based on measured edge rather than intuition.

Table of Contents

• The Problem: Marketing Budget as Guesswork

• From Cost Center to Capital Allocation

• The Framework: 4 Steps to Optimal Marketing Capital Allocation

• Real Example: From Rules of Thumb to Data-Driven Allocation

• Why This Outperforms

• The VaR Check: What If You're Wrong?

• How to Apply This Framework

• Conclusion

The Problem: Marketing Budget as Guesswork

Here is how most marketing budgets get set. The CFO picks a percentage of projected revenue -- typically 5-15% depending on the industry -- and hands it to the CMO. The CMO then distributes that budget across channels using some combination of historical precedent, platform rep recommendations, and gut feel.

This is not an exaggeration. In our work across 123 brands, we have seen the internal budget documents. The majority use one of three methods:

Consider what this means. Ninety-five percent of companies are deploying millions of dollars in capital without measuring the expected return or risk of that deployment.

No hedge fund sizes positions this way. No venture capital firm allocates across portfolio companies this way. No trader bets a percentage of their book on an asset because "that is what we bet last quarter." These disciplines survived by developing rigorous frameworks for capital allocation under uncertainty. Marketing has not.

The cost of this gap is not theoretical. Across our dataset, the difference between a brand's current allocation and their measured optimal allocation averages 31% of total spend. For a company deploying $2M per month, that is $620K per month being deployed suboptimally -- not wasted entirely, but generating materially less return than it could.

From Cost Center to Capital Allocation

The mental model shift is straightforward but consequential: marketing spend is capital deployed under uncertainty, not a cost to be managed.

When you treat marketing as a cost, the goal is to minimize it. You negotiate CPMs down. You cut the channel that "looks expensive." You celebrate efficiency gains that are really just measurement artifacts. The CFO's instinct is to reduce the line item.

When you treat marketing as capital allocation, the goal changes entirely. Now you are asking: where does my next dollar generate the highest risk-adjusted return? You might increase spend in a channel that looks expensive on a CPM basis because its incremental return per dollar is high and its variance is low. You might cut spend in a channel with great-looking ROAS because the uncertainty around that estimate is enormous.

Each channel becomes an asset class with three measurable properties:

This is not a metaphor. The mathematics are identical. A portfolio of marketing channels has an expected return, a variance, and a correlation structure -- just like a portfolio of financial assets. The tools that optimize financial portfolios apply directly.

The difference is that most marketing teams have never measured these inputs. They know their ROAS (or think they do). They have never quantified the variance of that ROAS. They have never measured channel correlations. Without these inputs, optimization is impossible. You are flying a plane with an altimeter but no airspeed indicator or compass.

The Framework: 4 Steps to Optimal Marketing Capital Allocation

Step 1: Measure True Returns (MMM, Not Attribution)

You cannot allocate capital to an asset if you do not know its return. In marketing, the first requirement is measuring incremental returns -- not attributed returns.

The distinction matters enormously. Attribution tells you which channels touched a conversion. MMM tells you which channels caused incremental conversions that would not have happened otherwise. The gap between these numbers is routinely 5-50x.

We have documented this across our dataset: channels that show a $5.50 CPA in attribution regularly show $452 in MMM -- a 7,218% difference. Attribution inflates the apparent return of channels that sit at the bottom of the funnel (branded search, retargeting) and suppresses the apparent return of channels that create demand (video, prospecting, social).

If you allocate capital based on attributed returns, you are systematically over-investing in channels that capture existing demand and under-investing in channels that create it. This is like a fund manager evaluating stocks based on last month's price movement rather than fundamental value.

The foundation of marketing capital allocation is a Marketing Mix Model that produces incremental return estimates for each channel. At Cassandra, we run Bayesian MMMs that output full posterior distributions -- not just point estimates, but probability distributions of each channel's true incremental effect. This is critical for every subsequent step.

Step 2: Quantify Uncertainty (Confidence Intervals, Not Point Estimates)

A return estimate without a confidence interval is not actionable for capital allocation. Knowing that Meta's iROAS is "3.2x" tells you nothing about how much to bet on it. Knowing that Meta's iROAS is "3.2x with a 90% credible interval of 2.4x to 4.1x" tells you a great deal.

This is the concept behind Risk-Adjusted ROAS. A channel with a 4x iROAS and a credible interval width of 3.0 is a fundamentally different investment than a channel with a 3x iROAS and a credible interval width of 0.8. The second channel is more predictable, more plannable, and in most cases a better allocation of marginal capital.

From our analysis of 1,221 MMMs, the average confidence interval width across channels is 1.6x -- meaning the typical channel's true iROAS could reasonably be 1.6 points higher or lower than the median estimate. Channels with interval widths above 2.0 are essentially speculative bets. Channels with widths below 0.8 are relatively dependable.

You need both numbers -- the return and the uncertainty -- before you can size a position.

Step 3: Map the Efficient Frontier (Optimal Channel Mix)

With return estimates and uncertainty quantified for each channel, you can now construct the marketing efficient frontier -- the curve that shows every possible channel mix plotted by expected return against risk.

This is direct Modern Portfolio Theory applied to marketing. The efficient frontier reveals the set of allocations where no reallocation can improve expected return without increasing risk, or reduce risk without decreasing expected return.

Most brands sit well below their efficient frontier. Across our dataset, the median brand's current allocation is 23% below their frontier -- meaning they could achieve the same expected return with 23% less risk, or 23% more expected return at the same risk level.

The frontier also reveals something that single-channel ROAS analysis misses entirely: diversification value. Channels with moderate individual returns but low correlation to other channels improve the portfolio. Cutting them to concentrate in "higher ROAS" channels often increases portfolio risk without proportionally increasing return.

Step 4: Size Positions Using Kelly-Inspired Logic

This is where the framework diverges from standard marketing optimization. Most optimizers, given a set of returns and constraints, will recommend the allocation that maximizes expected return. This sounds rational. It is not.

Maximizing expected return ignores a critical variable: the probability that your return estimates are wrong. If you have a 60% confidence that Meta's iROAS is 3.2x and a 40% chance it is actually 1.5x, going all-in on the 3.2x estimate is reckless.

The Kelly Criterion, developed by John Kelly at Bell Labs in 1956, solves this problem. Originally designed for optimal bet sizing in gambling, it answers: "Given my edge and the odds, what fraction of my bankroll should I wager?"

The formula, adapted for marketing capital allocation:

Optimal Channel Share = (Edge / Variance) * Scaling Factor

Where:

In practice, the formula produces allocations that are more conservative than pure expected-return optimization. It naturally sizes positions smaller when uncertainty is high, even if the point estimate looks attractive. It sizes positions larger when you have high confidence in a strong return.

This is precisely the behavior you want. When you know something works reliably, bet more. When you think something might work but you are not sure, bet less. No one follows this intuitively because point-estimate ROAS makes every channel look like a sure thing.

We use a fractional Kelly (typically 0.25x to 0.5x full Kelly) because full Kelly is known to be optimal only under perfect information -- which no one has. Half-Kelly sacrifices roughly 25% of theoretical growth rate but reduces variance by 50%. In marketing, where model estimates carry substantial uncertainty, fractional Kelly is the right posture.

Real Example: From Rules of Thumb to Data-Driven Allocation

Consider a mid-market ecommerce brand spending $2M per month across five channels. Here is their current allocation -- set historically with minor annual adjustments -- alongside the MMM outputs from their Bayesian model:

Current Allocation (Rule of Thumb)

Total expected incremental revenue at this allocation: $6.12M/month.

Now applying the Kelly-inspired capital allocation framework (Step 4 formula, using 0.4x fractional Kelly with diminishing return curves and correlation adjustments):

Framework Allocation

Total expected incremental revenue at framework allocation: $7.34M/month.

The same $2M budget. A 20% increase in expected incremental revenue. The allocation shift is not dramatic in absolute terms -- no channel is zeroed out, no channel doubles. The edge comes from right-sizing positions to their measured edge and uncertainty, not from making bold bets.

The most counterintuitive move: YouTube gets a raise despite having the lowest raw iROAS. Its low correlation with Meta and Google means it reduces portfolio variance. Under a capital allocation framework, diversification has quantifiable value.

The most painful move: TikTok gets cut nearly in half. Its median iROAS of 2.4x looks acceptable in isolation. But a CI width of 1.8 means the true iROAS could plausibly be anywhere from 0.6x to 4.2x. The Kelly formula refuses to size a large position on an edge that uncertain.

Why This Outperforms

A 20% efficiency gain in one month is meaningful. Over 12 months, it compounds.

The rule-of-thumb allocator generates $6.12M/month in incremental revenue and makes no systematic adjustments. Over 12 months: $73.4M.

The capital allocation framework generates $7.34M/month initially and recalibrates quarterly as new data narrows confidence intervals and shifts the frontier. Each recalibration captures incremental edge. By month 12, the compounding effect of quarterly reoptimization typically produces a cumulative efficiency advantage of 2-3x over the static allocation.

This is not a one-time arbitrage. It is a structural advantage that compounds because:

1. Uncertainty shrinks with data. Each quarter of new data narrows your confidence intervals. Narrower intervals allow larger positions in validated channels, which increases expected return.

2. Diminishing returns are dynamic. The point at which a channel hits diminishing returns changes with competitive dynamics, seasonality, and creative quality. Quarterly recalibration catches these shifts.

3. Correlation structures change. Channel co-movements shift as platforms evolve. Reoptimizing the portfolio captures diversification opportunities that a static allocation misses.

The 2-3x efficiency gap we observe across our dataset is not aspirational. It is the measured difference between brands that reallocate quarterly based on updated models and brands that set-and-forget annual budgets.

The VaR Check: What If You're Wrong?

Every allocation should come with a downside scenario. This is where Marketing Value at Risk completes the framework.

For the example above, the framework allocation produces a Marketing VaR(95%, quarterly) of 18% -- meaning in the worst 1-in-20 quarter, the portfolio will underdeliver expected revenue by 18% or more. That translates to a worst realistic quarterly revenue of $18.0M instead of the expected $22.0M.

Compare this to the rule-of-thumb allocation, which has a Marketing VaR of 26% due to its overweight in high-variance channels. The framework allocation not only delivers higher expected return -- it also carries less downside risk.

This is the direct analogue of what happens in financial portfolio optimization: properly diversified portfolios based on measured risk deliver better risk-adjusted returns than concentrated bets. The efficient frontier is not just a theoretical construct. It describes a real boundary that separates disciplined allocation from guesswork.

Before deploying any allocation, run the VaR check. If the downside scenario exceeds what your business can absorb in a bad quarter, reduce the scaling factor. Move from 0.4x Kelly to 0.25x Kelly. Accept a lower expected return in exchange for a tighter loss distribution. This is risk management, not timidity.

How to Apply This Framework

Here is the step-by-step implementation:

Quarter 1: Build the measurement foundation.

Run a Bayesian Marketing Mix Model on your historical data. You need at least 2 years of weekly spend and outcome data across all channels. The model must produce posterior distributions, not just point estimates. Cassandra's MMM platform generates these outputs natively -- including channel-level credible intervals, correlation matrices, and diminishing return curves -- which feed directly into the allocation framework.

Quarter 2: Compute your current position.

Map your current allocation against the efficient frontier. Quantify the gap. Calculate your Marketing VaR. This gives you the baseline: how far below the frontier are you, and how much downside risk are you carrying unknowingly?

Quarter 3: Deploy the optimized allocation.

Apply the Kelly-inspired sizing formula to generate target allocations. Implement gradually -- shift 50% of the reallocation in month 1, the remaining 50% in month 2. Monitor actual returns against model predictions.

Quarter 4: Recalibrate.

Update the MMM with new data. Recompute the frontier. Resize positions. This is not a one-time exercise. The edge compounds only with continuous recalibration.

Ongoing: institutionalize the process.

The goal is to make quarterly reallocation as standard as quarterly financial reporting. The CMO presents: expected return, risk profile, VaR, and position sizing rationale. The CFO reviews it in the same language they use for every other capital allocation decision.

Conclusion

Marketing budgets are capital deployed under uncertainty. The tools to allocate capital under uncertainty have existed in finance for 70 years. The gap is not technical -- it is conceptual. Most marketing teams have never been shown that their budget allocation problem is isomorphic to portfolio construction.

The framework is four steps: measure true incremental returns with MMM, quantify the uncertainty around those returns, map the efficient frontier of channel mixes, and size positions using Kelly-inspired logic that scales bets to your measured edge. Companies that execute this framework outperform rule-of-thumb allocators by 2-3x in marketing efficiency -- not because they spend more, but because every dollar is deployed where the risk-adjusted return justifies it.

The question is not "how much should we spend on marketing?" The question is "given what we know about each channel's return and risk, what is the optimal allocation of our next dollar?" Answer that question systematically, and the budget number takes care of itself.

---

Want to run a capital allocation analysis on your actual data? Book a call with our team and we will map your current allocation against your efficient frontier, compute your Marketing VaR, and show you exactly where reallocation unlocks incremental return.