Attribution: MTA vs MMM in the age of signal loss

Multi-touch attribution (MTA) used to be the gold standard. Then signal loss happened: iOS privacy settings, GA4 cookie decay, platform conversion APIs that disagree with each other, and every major ad network quietly modelling the missing data in its own direction.

Marketing mix modeling (MMM) — which pre-dates MTA by 40 years — is quietly back as the more reliable answer for strategic budget decisions.

Here's how to reason about which one to use when.

What MTA is good at

Tactical optimization inside a channel. Which ad set, audience, or keyword converted? MTA via platform conversion APIs is fine for this, because the comparison is within the same attribution system. You're not asking "is Google better than Meta?" You're asking "within Google, which campaign?"

What MTA is bad at now

Cross-channel comparison. "Meta reports $4.2M attributed, Google reports $3.8M, LinkedIn reports $1.1M — total = $9.1M. Actual revenue = $6.7M. Where did the $2.4M go?"

The answer is: into double- and triple-counting. MTA systems in 2026 routinely over-report by 20–45% when you sum them.

What MMM does differently

Top-down regression on aggregated data. Given monthly spend per channel + monthly revenue + external factors (seasonality, pricing changes, macro events), MMM fits a model that estimates incremental contribution per channel. No tracking. No cookies. No ad platform cooperation required.

Strengths: Unbiased across channels. Immune to signal loss. Measures brand/offline spend alongside digital.

Weaknesses: Slow (quarterly cadence, typically). Needs 2+ years of clean data. Can't optimize individual campaigns.

The 2026 stack that works

Run both. MMM as the strategic budget-allocation tool (refreshed quarterly), MTA as the tactical within-channel optimizer (real-time).

Practical setup:

1. MMM quarterly. Use tooling like Meridian (Google open-source) or Robyn (Meta open-source). Outputs: incremental contribution per channel, saturation curves, and an "optimal budget split" recommendation.

2. MTA within each channel. Use the platform's own conversion API + server-side events. Optimize bid strategies and creative based on these events.

3. Incrementality tests quarterly. Run a geo-holdout or user-holdout test on one channel per quarter. Validates MMM outputs against ground truth.

4. Roll it up for the CFO. Monthly revenue report uses MMM-adjusted ROAS per channel. MTA numbers are never surfaced to finance directly — they're a diagnostic internal tool.

A worked example

A SaaS client, Q4 2025:

• MTA (platform-reported): Meta 4.1×, Google 3.8×, LinkedIn 2.2×. Total attributed revenue $8.4M.
• MMM: Meta 2.9×, Google 3.4×, LinkedIn 1.3×. Incremental revenue $6.1M.
• Actual logged-in revenue: $6.3M.

MMM was 97% accurate to ground truth. MTA was 33% overcounted.

The decision that fell out: shift 40% of LinkedIn spend to Google non-brand. The holdout test the following quarter confirmed the incrementality gain.

What to stop doing

• Stacking platform-reported attribution and treating the sum as revenue. You're over-reporting by 20–45%.
• Running a single-vendor "attribution tool" that black-boxes the math. If you can't audit the inputs, you're in a worse position than the platforms.
• Making cross-channel budget decisions from MTA alone. That's what MMM is for.

The right conversation with your CFO

"Our incremental (MMM-adjusted) ROAS is 3.1×. Platform-reported sums to 4.4× — which we don't trust and never use for budget decisions." Two sentences. Every CFO nods.