The measurement problem is not a lack of metrics
Paid media teams have more reporting than ever.
Google Ads, Meta, analytics platforms, ecommerce systems, CRMs and attribution tools can all report revenue, conversions and return on ad spend. The problem is that those systems can disagree — sometimes materially — while each appears internally consistent.
That does not necessarily mean one is broken.
They may be answering different questions.
Attribution asks which touchpoints receive credit for observed conversions. Incrementality asks how many conversions would not have happened without the advertising. Marketing mix modeling estimates how changes in media and other factors are associated with business outcomes over time and can support higher-level allocation decisions.
A professional measurement stack does not ask which one should replace the others.
It asks which method is appropriate for which decision.
Attribution is fast, granular and not causal by default
Attribution is the operational workhorse of digital advertising.
It can report campaign, keyword, ad, audience or creative performance quickly enough to influence daily decisions. Platform attribution also feeds the optimization systems that decide where ads should be delivered.
That speed is extremely valuable.
But attribution usually observes relationships between exposure, clicks and conversions. It does not automatically establish that the advertising caused the conversion.
A branded search campaign is the classic example. Someone already intending to buy may search the brand name, click an ad and purchase. The platform can correctly attribute the sale to the ad interaction under its chosen model while still overstating how much incremental demand the ad created.
This does not make attribution useless.
It makes attribution operational rather than universally causal.
Incrementality creates the counterfactual
Incrementality testing attempts to answer the question attribution cannot answer directly:
What would have happened if the advertising had not run?
A test can withhold advertising from a control group while maintaining exposure in a treatment group, then compare business outcomes.
Depending on the channel and platform, that can involve randomized holdouts, geographic experiments or other controlled designs.
The strength is causality.
The weakness is that experiments cost time, require statistical power and cannot practically answer every campaign-level question every day.
That makes incrementality especially valuable for strategic uncertainties: whether a channel is creating new demand, whether a retargeting program is taking excessive credit, whether increased spend is still incremental, or whether upper-funnel media drives business outcomes that click-based attribution misses.
Google has been moving incrementality deeper into its measurement stack, including geo-experiment and MMM tooling.
MMM sees the portfolio
Marketing mix modeling works at a broader level.
Rather than reconstructing individual customer journeys, an MMM typically uses aggregated historical data to estimate the relationship between media investment and outcomes while accounting for other factors.
That makes it particularly useful for cross-channel allocation, channels with weak click paths, offline media and strategic budget planning.
It also introduces important limitations.
Models depend on sufficient variation and high-quality data. Correlated channels can be difficult to separate. Business variables such as price, promotions and seasonality can distort interpretation if they are omitted or poorly specified.
Modern MMM development is increasingly trying to address those weaknesses.
Google’s Meridian, for example, supports calibration from incrementality experiments and has added features for non-media variables, longer-term effects and marginal ROI.
The important lesson is not that every advertiser should adopt Meridian. Google is itself a media platform and its measurement products should not be treated as neutral proof of Google media effectiveness.
The lesson is that MMM is becoming more accessible and increasingly connected to experimental calibration rather than being treated as an isolated annual econometric exercise.
The three methods should reconcile, not compete
A useful measurement stack creates a hierarchy of questions.
Attribution can help decide which search term, creative or campaign deserves attention today.
Incrementality can determine whether a channel or tactic is actually creating additional business.
MMM can help decide whether the portfolio-level allocation between Meta, Search, YouTube, retail media, television and other investments is economically rational.
The methods should also challenge one another.
If platform attribution says a channel produces 6x ROAS but repeated incrementality tests show substantially weaker causal return, daily campaign decisions should not ignore that evidence.
If an MMM suggests a channel is highly productive while experiments repeatedly fail to detect lift, the model assumptions may need investigation.
If an incrementality result from one short test conflicts with years of broader evidence, the test itself may have been underpowered or contaminated.
Measurement quality comes from triangulation.
The stack needs a financial layer
Marketing measurement often stops too early.
Even perfectly incremental revenue is not the same as profit.
A channel can create new sales and still destroy contribution margin if acquisition cost, discounts, fulfillment, payment fees, returns or servicing costs consume the economics.
The measurement stack should therefore connect media outcomes to finance.
At minimum, operators should be able to move from attributed revenue to incremental revenue and then to incremental contribution.
That creates a much stronger budgeting question than “Which channel has the highest ROAS?”
The better question becomes:
Where can the next dollar of spend create the greatest acceptable incremental economic value?
Use each method at the cadence it deserves
The stack becomes practical when each method is assigned a decision cadence.
Attribution and platform reporting operate continuously.
Incrementality can be run periodically around material uncertainties, new channels, budget expansions, promotions or suspected attribution bias.
MMM can operate on a slower strategic cadence and inform quarterly or annual allocation, scenario planning and diminishing-return curves.
There is no universal source of truth
The phrase “single source of truth” is useful for transaction data. It is more dangerous for marketing effectiveness.
Finance can have a source of truth for realized sales.
But no observational reporting system can perfectly answer every causal question about media.
Professional measurement therefore requires a different standard: not one number, but a coherent evidence system.
Attribution provides speed. Experiments provide causal validation. MMM provides portfolio perspective. Finance determines whether the resulting growth is economically valuable.
When those layers disagree, the disagreement is not noise to eliminate.
It is a signal to investigate.
