Marketing attribution's central limitation is this: it observes sequence, not cause. An attribution model records that a touchpoint occurred before a purchase. It cannot establish that the touchpoint created the purchase, and the research comparing attribution against randomised experiments shows the gap between those two things is often large.

If your budget decisions rest on an attribution dashboard read literally, this is worth fifteen minutes of your attention. Our pillar essay argues that the digital era's deepest error was substituting the measurable for the valuable, and attribution is where that substitution did its most expensive work. Here is what the evidence shows, where attribution still genuinely helps, and how to build a measurement approach you can defend.

Why the Last Click Gets Credit It Did Not Earn

A customer who searches your brand by name, clicks an ad and buys may have been influenced by years of product experience, recommendations, brand advertising and reputation. The final click is visible because it happened close to the transaction, and visibility is not the same thing as causation.

Attribution models hand credit to observable touchpoints because those are the only ones they can see. The memory that put you on the shortlist two years ago produces no signal at all. That asymmetry systematically flatters demand capture (search, retargeting, outreach) and systematically undervalues demand creation (brand, content, reputation), which is how budgets drift to the bottom of the funnel one reallocation at a time.

What the Randomised Experiments Found

Two research programmes matter here.

First: Brett Gordon, Florian Zettelmeyer, Neha Bhargava and Dan Chapsky published A Comparison of Approaches to Advertising Measurement in Marketing Science, using large-scale randomised experiments to test the observational attribution methods most organisations rely on. The observational methods often failed to recover the true causal effects of the advertising. Sometimes they overstated it substantially; the direction and size of the error were hard to predict in advance.

Second: Randall Lewis and Justin Rao analysed 25 large field experiments with major advertisers in The Unfavorable Economics of Measuring the Returns to Advertising. Their finding is sobering: advertising's commercial effect is often so small relative to the natural variation in customer purchasing that even experiments involving millions of customers produced wide confidence intervals around return on investment.

Together the studies say two things. Attribution's answers can differ materially from causal reality, and even the gold-standard methods struggle for precision. Anyone selling you certainty about advertising ROI is selling past what the evidence supports.

What Attribution Is Still Good For

The conclusion is not "stop measuring", and it is not "attribution is useless".

Attribution remains genuinely useful for tactical optimisation within a channel: comparing creative, spotting broken journeys, managing bids, identifying where interested buyers drop out. Les Binet's work on marketing effectiveness in the digital age makes the distinction cleanly: direct attribution supports continuous tactical decisions, but it misses earlier influences in the journey, so it should never be the sole evidence for strategic budget allocation.

The failure mode is elevation: taking a tool built for tactical comparison and treating its output as the definition of marketing effectiveness.

How to Measure When No Single Method Tells the Truth

Different questions need different methods, each with known limits:

1. Attribution for tactical optimisation. Fast, granular, always on. Blind to incrementality and to anything that happens far from the conversion window.
2. Experiments for causal questions. Geo splits, holdouts and incrementality tests establish whether an intervention actually changed behaviour. Slower and costlier, but the only honest answer to "did this work?"
3. Marketing mix modelling for the portfolio view. Models how investment across channels contributes to outcomes over time, including effects attribution cannot see. Requires history and statistical care.
4. Brand and customer research for the invisible effects. Awareness, consideration and mental availability shifts appear here years before they appear in sales data.

The practical pattern for a B2B team: run attribution for weekly channel management, at least one genuine incrementality test per year on your biggest line item, and a simple brand tracker so demand creation has evidence of its own. Then make the budget call with judgement across all three, documented, so the reasoning can be defended and revisited. If you have not yet named the growth constraint your measurement is meant to serve, start with our guide to diagnosing marketing problems before you optimise.

Better measurement is combined evidence plus judgement, not a dashboard read literally.

For the full story of how the digital era came to mistake measurement for marketing, read the pillar essay: Digital Marketing vs Marketing Strategy.

Sources

• Brett Gordon, Florian Zettelmeyer, Neha Bhargava and Dan Chapsky, A Comparison of Approaches to Advertising Measurement (Marketing Science), on observational attribution vs randomised experiments.
• Randall Lewis and Justin Rao, The Unfavorable Economics of Measuring the Returns to Advertising (Quarterly Journal of Economics), on the statistical difficulty of measuring ad returns.
• Les Binet, Unlocking Marketing Effectiveness in the Digital Age, on misattribution, incrementality and combining attribution, experiments and mix modelling.
• Les Binet and Peter Field, The Long and the Short of It (IPA), on short-term signals vs long-term brand effects.