The uncomfortable state of measurement
Marketing measurement runs on a confidence gap. Nielsen's Marketing ROI Blueprint, published October 2025, found 85% of marketers express confidence in their ability to measure ROI while only 32% actually measure it holistically across channels. The stakes of that gap keep rising: Gartner's 2025 CMO Spend Survey shows marketing budgets flat at 7.7% of company revenue for a second straight year, with half of CMOs working with 6% or less.
Flat budgets mean every dollar needs a defensible story, and the most common story is built on the weakest foundation available: a survey by EMARKETER and Snap of 282 senior US marketers found 78% still lean on last click attribution, while only 22% believe it reflects real business impact. Believing one thing and reporting another is not a measurement strategy.
Why attribution keeps lying to you
Attribution's problems are structural, not fixable with a better dashboard. Last click systematically hands credit to the channel closest to checkout, usually branded search, and starves the channels that created the demand. Analytics veteran Avinash Kaushik put it bluntly: the only remaining use of last click attribution is to get you fired.
Multi touch models promised better and collided with reality: privacy rules, Apple's tracking prompts, walled gardens that refuse to share user paths, and a growing share of buying journeys that happen in dark social and AI assistants where no tracker follows. Even asking customers directly is noisier than it looks: a Dreamdata analysis of demo signups found a third skipped the how did you hear about us field and only about half of the answers were usable. Every attribution number you have is a partial reconstruction. Useful for channel tuning, dangerous for budget truth.
32%
of marketers measure ROI holistically across channels, while 85% are confident they can (Nielsen, 2025)
78%
of senior marketers still rely on last click attribution; only 22% trust it (EMARKETER and Snap)
0.70x
median incremental return on branded search spend across Haus geo experiments
19%
average incremental lift found across 640 Meta experiments (Haus, 2025)
Incrementality: The question that actually matters
The only question a CFO ultimately cares about is counterfactual: what would have happened without the spend? Incrementality testing answers it directly by running geographic experiments, ads on in some markets, off in others, and measuring the difference. The canonical cautionary tale is eBay's experiments published in Econometrica: when eBay paused brand keyword search ads, organic search recaptured almost all the traffic, and measured returns on that spend were a fraction of what attribution reported.
Modern platforms have industrialized the method, and the results keep humbling dashboards: Haus, analyzing 640 Meta experiments for advertisers averaging 14 million dollars in annual spend, found a real average lift of 19%, that Meta's own attribution understated true incrementality by about 15% on average, and that branded search delivered a median incremental return of 0.70, meaning the median advertiser loses money on the spend their dashboard celebrates most.
You do not need a data science team to start: one geo holdout on your biggest spend line, run for three to four weeks, will teach you more than a year of attribution reports.
Why marketing mix modeling came back
MMM, the statistical workhorse of the TV era, returned because it needs no cookies or user tracking: it models aggregate spend against aggregate outcomes. Meta open sourced Robyn, and Google released Meridian to all advertisers in January 2025. A healthy skepticism note from industry coverage: the biggest ad platforms now also author the measurement models, so calibrate any MMM with your own incrementality experiments rather than accepting defaults.
A measurement stack that matches company size
- Weekly, at any size: blended metrics. Total revenue over total marketing spend, fully loaded CAC, and pipeline by month
- Monthly: channel dashboards used for optimization decisions only, never for budget allocation claims
- Quarterly, once spend is meaningful: one incrementality test on the largest or most questioned channel
- Annually, at roughly a million dollars of spend across four or more channels: a marketing mix model calibrated with your experiment results
- Always: a written definition of CAC, LTV and payback that finance has signed, so the numbers cannot move between meetings
The metrics worth defining properly
Blended metrics resist manipulation because they include everything. Customer acquisition cost only means something fully loaded: all sales and marketing spend, including salaries and tools, divided by new customers. Lifetime value should be gross margin adjusted, not revenue, or it flatters everyone. The ratio between them has usable benchmarks: SaaS benchmark data from Benchmarkit puts the median around 3.2 to 1 for growth stage B2B companies, with top performers above 4.
CAC payback, the months of gross margin needed to recover acquisition cost, has published targets near 12 months for SMB focused companies, stretching toward 24 for enterprise, per First Page Sage's 2025 benchmarks. And one ratio deserves rehabilitation: blended marketing efficiency, total revenue over total marketing spend, is crude, but it is the one number no channel dashboard can inflate, which is precisely why boards trust it.
What honest measurement admits it cannot see
A credible framework names its blind spots. Brand effects run on lags your dashboards cannot hold: Nielsen's long term effect research finds roughly half of media's total impact lands beyond the measurement window of standard reporting, and rebuilding a brand that went dark takes years, not quarters. Dark social, private shares, podcast mentions and AI assistant recommendations move real buyers and leave no referrer. B2B sales cycles mean this quarter's revenue partly reflects last year's marketing.
And small budgets face a statistical wall: incrementality tests need enough conversions to detect a lift, which is why the practical sequence for smaller teams is blended metrics first, experiments when volume allows, models when scale demands. The goal is not measurement perfection. It is making resource decisions on evidence strong enough to survive a skeptical CFO, and being honest about the rest.
Attribution answers who gets credit. Incrementality answers what would have happened anyway. Budgets should follow the second question.
Written by
Matias Pavez
Managing Partner
MBA from UC Berkeley Haas. Executive with experience in venture building, operations and growth strategy. He has led teams at BairesDev, SONDA and technology startups, and connects the studio to the innovation ecosystems of Chile and Silicon Valley.
