The ROI Illusion: How Pure Efficiency Worship Can Cripple Growth

The ROI Obsession: A Decade‑Long Drift

Scroll back ten years and hardly anyone in marketing talked about return on investment. Marketers cared about building brands, winning mind‑share and creating long‑term demand. The language of finance belonged to the CFO. Today the opposite is true. Marketing conferences and boardroom decks are cluttered with ratios and efficiency metrics. The CMO is expected to serve as chief investment officer. CEOs and CFOs consistently tell researchers they want outcomes and ROI from marketing, not soft metrics. In 2025, the CMO Survey found that 63 % of marketing leaders were under increased pressure from finance chiefs, 61 % from CEOs and 50 % from boards to demonstrate financial outcomes (cmosurvey.org). Another survey by Tom Roach and Marketing Week showed that 48% of marketers say ROI is the metric their CEO or CFO cares about most and 37% reported that the emphasis on ROI has increased (thetomroach.com). Yet at the same time, more than one‑third (34%) of marketers rarely or never measure marketing ROI (marketingweek.com) and only 38% evaluate ROI across channels holistically (nielsen.com). Marketers are being told to chase a metric they often can’t measure properly.

This obsession doesn’t end at the marketing team. Finance is also hooked on efficiency. IBM’s 2024 CFO study found that nearly two‑thirds (65%) of finance leaders feel pressure to accelerate ROI from technology investments (cfo.com). 57% prioritise short‑term targets over long‑term innovation, sacrificing growth for near‑term ratios (cfo.com). Only 30% of CFOs hit the balance between efficiency and growth, and just 9% are recognised as “leading CFOs” who manage to balance cost reduction with growth (cfo.com). The entire C‑suite is chasing ROI, often at the expense of strategic growth. The result is a business culture that worships efficiency and punishes long‑horizon investment.

ROI vs. Growth: A False Equivalence

Return on investment is seductive because it offers the illusion of precision. An ROI of 150% seems objectively better than 50%. But ROI is a ratio, not a measure of absolute value. A high percentage can come from a tiny base. Tom Roach illustrates this with a simple example: a 150% ROI on a $10 m campaign produces $5 m profit, while a 500% ROI on a $100 k test yields just $400 k profit (thetomroach.com). A marketer chasing the higher ratio would choose the smaller campaign and leave $4.6 m in profit on the table. This distortion is why ROI is sometimes called a marmite metric – loved by some, hated by others. Roach warns that focusing on ROI favours small, tactical spends and may signal underinvestment; high ROI can even “destroy value” because it leads to chronic under‑spending (thetomroach.com).

Growth, by contrast, is about absolute impact. It asks how many dollars of incremental revenue or profit a marketing program delivers, not just the percentage return per dollar. Growth acknowledges that businesses need scale to survive. Yet the ROI mindset pushes marketers to reallocate budgets from channels that drive growth but look less efficient to those that produce impressive ratios at small scale. This efficiency obsession compresses budgets into bottom‑funnel tactics, starving brand building and long‑term demand creation. When Nielsen surveyed global marketers in 2024, it found that 72% expected bigger budgets, yet two‑thirds of spend now goes to digital channels, while only 38% measure ROI across traditional and digital media together (nielsen.com). The shift to performance marketing channels may deliver high ROIs but can reduce reach, limiting top‑of‑funnel growth (nielsen.com). Efficiency has become a tyrant.

The Marketer’s New Toy: Mixed‑Modelled ROI

The industry’s infatuation with ROI coincided with the rise of marketing mix modelling (MMM). Modern MMMs claim to quantify the incremental impact of each channel and deliver an ROI figure for every dollar spent. Open‑source tools such as Meta’s Robyn and Google’s Meridian have made MMM accessible to teams with limited budgets. According to documentation, Robyn uses ridge regression (L2 regularisation) to manage multicollinearity and shrink coefficients towards zero (facebookexperimental.github.io). Meridian employs a Bayesian linear regression with parametric transformations for lag and saturation (developers.google.com). These models provide static weighted coefficients representing an assumption that channel effectiveness remains constant over time. A channel that delivered a given return last year is assumed to deliver the same return this year.

This assumption is convenient for modellers but often false. Channel effectiveness changes with consumer behaviour, competition, seasonality and macro‑economic shocks. PyMC’s marketing team calls the constant‑coefficient assumption “unrealistic” because it assumes the effect of a channel never changes (pymc-labs.com). They propose using Gaussian processes to model time‑varying channel effectiveness, capturing how the response to advertising evolves (pymc-labs.com). Their implementations allow both the intercept and media contributions to vary over time (pymc-marketing.io). The difference is profound: when coefficients vary, the model can detect when a previously effective channel declines or a laggard improves. It can reflect the real world where a Facebook campaign that thrived during lockdown may lose potency post‑pandemic. Models with static coefficients, like Robyn’s ridge regression or Meridian’s non‑varying betas (developers.google.com), obscure these dynamics and mislead decision‑makers.

Academic research echoes this warning. A 2024 working paper from the Marketing Science Institute argues that non‑linear and time‑varying effects are often indistinguishable; patterns that look like saturation could actually be time‑varying effectiveness (thearf-org-unified-admin.s3.amazonaws.com). Estimating models on rolling windows – a common workaround – implicitly assumes time‑varying coefficients (thearf-org-unified-admin.s3.amazonaws.com). In plain terms: if your model doesn’t allow coefficients to drift, you will misattribute changes in channel performance to noise or saturation and get the wrong answer. That means the ROI figures you rely on could be systematically biased, and open source compounds this issue in a major way.

Open‑Source Explosion, Open‑Source Problems

Open‑source MMMs exploded because they promised free, transparent measurement. Yet the reality is messy. Robyn and Meridian both produce ROI figures using weighted coefficients that implicitly assume constant channel performance. Robyn uses L2 shrinkage which biases coefficients toward zero (facebookexperimental.github.io). Meridian includes a time‑varying intercept to capture trend and seasonality, but media coefficients remain static (developers.google.com). When you feed these models multi‑year data, they generate ROI numbers that appear precise but ignore whether the channel’s effectiveness has deteriorated or improved. Worse, many implementations are slow and deliver results weeks after a campaign, long after decisions were made.

The rise of open source also means anyone can spin up a model without understanding its assumptions. As Ipsos MMM experts warn, Bayesian models like Meridian require careful specification of priors; assuming media effectiveness is always positive can lead to unrealistic results if data are poor or priors are mis‑specified (mma.com). They emphasise that good MMM requires comprehensive data – not just marketing spend but economic variables, competitor actions and macro events – and continuous measurement rather than one‑off optimisations (mma.com). Most open‑source setups ignore these complexities. As a result, marketers end up with ROI numbers that look scientific but are built on shaky foundations.

The Trap of High‑ROI Channels: A Worked Example

To see how an ROI‑first over Growth-first mindset can distort budgets, imagine a brand with two channels:

• Channel A: Retargeting ads that reach existing visitors. Each dollar spent yields $5 in incremental revenue, producing an ROI of 400%. However, the audience pool is small; the channel saturates after $50 k in spend. Beyond that threshold, additional impressions go to the same customers and incremental sales collapse. The true incremental revenue curve flattens sharply.

• Channel B: Brand advertising on premium video. Each dollar yields $2 in incremental revenue, an ROI of 100%. The channel’s audience is broad and unsaturated; you can invest millions before hitting diminishing returns. It also generates long‑run brand equity that drives repeat purchases not immediately captured in short‑run sales.
  
  

If you optimise purely on ROI, you would pour your entire budget into Channel A because 400% beats 100%. After spending $50 k, you will have earned $200 k incremental revenue and exhausted the audience. Investing the next $950 k in retargeting will produce close to zero additional revenue. Meanwhile, ignoring Channel B means you forego the $2 m in incremental revenue ($2 for every $1 on a $1 m budget) and the long‑term halo. The ROI‑maximising allocation yields $200 k profit; the growth‑maximising allocation yields ten times more. This is not a hypothetical: digital advertising often delivers high ROI at small scale and saturates quickly, while mass reach media looks less efficient but scales up. Tom Roach notes that high ROI can be a sign of underinvestment (thetomroach.com). Chasing the percentage rather than the absolute value means settling for the smallest possible impact.

The opposite trap also exists. A channel can show a low ROI because of bad creative or poor execution. Suppose Channel C is a video campaign with an ROI of 30%, well below your hurdle rate. If the creative is sloppy or the frequency is mis‑managed, the low ROI reflects execution failure, not intrinsic channel potential. Cutting budget based on that number may permanently damage brand awareness. A growth‑first view would diagnose why the channel underperformed, fix the creative and evaluate again, rather than abandon it. The ROI metric alone cannot tell you whether the problem is the channel or the execution.

The Illusion of Stability: When ROI Numbers Lie

Another hidden danger is the confidence interval. ROI figures are point estimates derived from noisy data. Some channels, such as small test programs or emerging platforms, have few observations and wide variance. The topline ROI might look attractive but the uncertainty bands may span negative and positive values. For example, your MMM might estimate that TikTok drives $4 in revenue per dollar spent with an ROI of 300%, but the 95 % confidence interval ranges from –100 % to +700 %. A risk‑averse marketer would not bet the farm on a channel whose true ROI could be negative. Without confidence intervals (for example Mutinex’s approach in Scenario Planning is built around ranges, not point estimates), the numbers look definitive. With them, you see just how unreliable some estimates are. This is why Bayesian models matter: they provide probabilistic estimates and credible intervals (mutinex.co).

Moreover, ROI figures can be unstable over time. A channel might deliver 200 % ROI one quarter and 20 % the next because of seasonality or competitive noise. The underlying coefficient in the model may be fluctuating, but if the model assumes it’s constant, the changes manifest as noise. This creates a dangerous cycle: marketers see ROI swinging wildly, lose trust and cut budgets, which further skews the data and degrades the model. The Forrester marketing survey cited in Mumbrella reports that 64% of B2B marketing leaders do not trust their organisation’s marketing measurement when making decisions (mumbrella.com.au). Lack of trust leads to short‑termism and reactive cuts.

I liken giving marketers ROI numbers without rigorous validation to handing a pilot uncalibrated instruments. Without validation, those numbers can send you crashing into the mountain. Validation means examining predictive accuracy, checking multicollinearity, inspecting uncertainty and ensuring the model performs out‑of‑time. Most open‑source tools deliver ROI numbers but leave validation to the user, who may not have the technical expertise. Confidence intervals, time‑varying coefficients and out‑of‑sample tests are often absent.

A Response: The Open MMM Validation Framework

We confronted these issues head‑on. After building MMMs for years, the team realised that providing ROI figures without robust validation created dangerous blind spots. We made mistakes; models over‑fit historical data and under‑predicted future performance. To fix this, they built an internal validation framework. The framework tests any model – open or proprietary – and measures predictive accuracy, ROI stability and cross‑validation performance It allows procurement teams to demand independent validation in RFPs.

Growth as the North Star

So what should marketers do in a world awash with ROI metrics? The answer is to stop defining themselves by purely ROI and start defining themselves by growth. Growth is not the opposite of efficiency; it simply recognises that efficiency without scale is meaningless. A 50 % ROI on a $10 m spend ($5 m profit) is better than a 500 % ROI on a $100 k spend ($400 k profit) (thetomroach.com). Growth demands that you look at the absolute incremental revenue generated by a channel or mix, not just the percentage. It requires models that detect how channel effectiveness changes over time—which rules out static‑coefficient tools like Meridian and Robyn. It requires platforms with an optimiser that can simulate different budgets, predict growth under varying scenarios and provide saturation curves so you know when a channel will saturate. Without these features, you will default to ROI and under‑invest.

Returning to First Principles

The marketing industry’s fixation on ROI is understandable. It offers the promise of objectivity and a way to speak the language of finance. But like any ratio, it can mislead when divorced from scale and context. Efficiency is not a strategy; it is one dimension of a broader growth equation. As open‑source models proliferate, marketers must become more skeptical about what the numbers really mean. They must demand transparency, validation and time‑varying insights. They must resist the temptation to optimise purely on ROI and instead allocate budgets to deliver the greatest incremental growth.

The stakes are enormous. Global advertising spend exceeds $150 billion in some markets; misallocating even a fraction of that has material consequences. The ROI illusion is costing businesses tens of billions in lost growth. The only remedy is a return to first principles: invest for growth, validate your models and treat ROI as a tool for growth, not a master.