School Me MarketingSMM
FoundationsAdvanced11 min deep read · 30-sec skim available

Attribution & Incrementality

Add up what every platform claims it 'drove' and you'll find you sold your revenue three times over. Attribution assigns credit; incrementality proves cause — and the gap between them is where marketing budgets go to be wasted with confidence. The modern stack triangulates: experiments for truth, econometrics for strategy, attribution demoted to tactical steering.

⚡ Understand it in 30 seconds

  • Every attribution model is an opinion about credit, not a measurement of cause. Last-click hands the medal to whoever harvested the decided buyer, which is why branded search looks like a genius and the campaign that created the intent looks like a cost.
  • Data-driven models (Shapley, Markov) are better bookkeeping of tracked paths — and still just bookkeeping: blind to TV, word of mouth, and the counterfactual question that actually matters. Would she have bought anyway?
  • The tracking substrate is collapsing anyway — iOS privacy prompts, cookie deprecation, walled gardens grading their own homework — which is why marketing-mix modelling came back from the 1960s and experiments became the gold standard.
  • The canonical demonstrations: eBay switched off brand-search ads and sales barely noticed — the clicks were harvest, not cause. P&G cut heavily into digital spend with no reported sales impact. The triangulated stack — MMM for allocation, geo-experiments for truth, attribution for in-channel tactics — exists so you find this out on purpose, cheaply, instead of by accident, expensively.

Go deeper

The core idea

Break the intuition first: the dashboard doesn't know which ad worked. Platform-reported conversions answer 'was our ad somewhere on the path?' — and paths are crowded, so every platform finds itself present and claims the sale. Summed platform claims routinely exceed actual revenue several times over. They can't all be right, and structurally none of them is, because presence isn't influence. The question a budget actually needs answered is counterfactual: how many of these sales would have happened without the spend? That question has a name — incrementality — and no attribution model, however sophisticated, answers it, because models allocate observed conversions. They never observe the world where the ad didn't run.

Understand the models as opinions with known biases. Last-click credits the final touch: it systematically flatters harvest channels (branded search, retargeting, cashback affiliates) that intercept buyers who've already decided, and starves the creation channels (video, content, upper-funnel social) that made them decide — optimising to it slowly reallocates budget from farming to gleaning. First-touch inverts that bias. Linear, time-decay, and position-based split the credit by arbitrary formula — different opinions, same epistemic status. Data-driven attribution (Shapley values, Markov removal effects computed over observed paths) is genuinely better bookkeeping: it at least asks how conversion probability changes when a touchpoint appears. But it stays bookkeeping of the tracked world — offline influence, dark social, and TV don't emit clicks — and it still can't see the counterfactual. Changing attribution models reallocates credit, and therefore budget, without a single fact about causation changing. That sentence should be laminated above every dashboard.

Incrementality is the counterfactual discipline, and it's older than the internet: hold something back, compare, believe the difference. The instruments, in rising order of ambition: conversion-lift studies (platform-run randomised holdouts — users eligible for your ad but shown none — measuring the true delta); geo-experiments (matched cities or states dark versus exposed — the workhorse, because it needs no user tracking at all and survives every privacy change); and the bravest, cheapest instrument in marketing, which is simply turning it off — regionally, briefly, on purpose. The classics earned their fame: eBay's economists found paid clicks on brand terms were overwhelmingly harvest (the buyers arrived anyway via the organic link one centimetre below), and P&G's public digital cuts — around two hundred million dollars — reportedly changed sales not at all. Neither result proves your brand search or your digital is waste. Both prove the only way to know is to test the counterfactual rather than admire the click.

The working stack triangulates three instruments at three altitudes. Marketing-mix modelling — econometrics regressing sales on spend, seasonality, pricing, and competition — is back from the 1960s because it needs no user-level tracking: it sees TV, print, and trade alongside digital, estimates diminishing-returns curves per channel, and answers the quarterly allocation question. Its price is assumptions, data hunger, and lag (and the majors now open-source their tooling, which says a lot about where measurement is heading). Experiments are the truth layer: run a handful of geo-tests a year on your biggest, most contested lines, and use the readouts to calibrate the MMM — an MMM anchored to experimental lift is twice the instrument. Attribution survives at the tactical altitude, where its bias is roughly constant and its speed is the point: comparing creatives, audiences, and bids within one channel, daily. One instrument for strategy, one for truth, one for steering — the failure mode is using any of them at another's altitude.

The business case

Why marketers care

The direct stakes are the media budget. Over-credited harvest channels compound quietly — each quarter's 'efficient' reallocation shifts money from cause to coincidence until growth stalls with excellent ROAS. Incrementality testing is how the ratchet gets caught and reversed.

The politics need the vocabulary: every platform, agency, and internal team prefers the model that flatters its line. Knowing that models are opinions — and that experiments outrank them — converts measurement meetings from theology to engineering.

And privacy made this mandatory: the click-trail era is ending by law and by platform policy. Teams whose measurement rests on user-level tracking are building on melting ice. MMM plus experiments is the stack that survives, and the migration rewards whoever starts early.

See it

The visual model

Four instruments, four altitudes

Each instrument answers a different question — the failure mode is asking any of them a question from another column.

Read this diagram as text

Four pillars comparing measurement instruments. Rules-based attribution: fast and free, but an arbitrary opinion that flatters harvest channels. Data-driven attribution: honest bookkeeping of tracked paths, blind to offline influence and the counterfactual. Experiments — holdouts and geo-tests: the ground truth on incrementality, run occasionally at scale. Marketing-mix modelling: strategic allocation across all media including offline, privacy-proof but slow and assumption-heavy.

The receipts

Where it comes from

Last-click inherited from 1990s ad-server logging; Shapley's cooperative game theory (1953) later applied to path data; Blake, Nosko & Tadelis's eBay search experiments (Econometrica, 2015); marketing-mix modelling from 1960s econometrics, revived industry-wide after Apple's App Tracking Transparency (2021).

Attribution wasn't designed — it congealed. Early ad servers logged the last click before purchase because it was easy, and a generation of marketing organised itself around an engineering convenience. Most measurement débâcles since are that original sin compounding.

The eBay study became the field's Copernican moment: rigorous experiments showing paid brand-search clicks were overwhelmingly non-incremental — the customers were coming anyway. Its deeper lesson wasn't 'brand search is waste' but that observational credit and causal contribution can diverge totally, and only experiments reveal the gap.

The MMM revival is history rhyming: the technique that measured TV in the 1960s returned because it never needed cookies. The modern twist is calibration — mix models tuned to experimental lift readouts — and the open-sourcing of MMM tooling by the largest ad sellers, a tacit admission that user-level attribution's decade is over.

Brands you know

Seen in India

Educational readings of familiar brands — how the concept helps you see what they do, not claims about their current campaigns.

Meesho

The value-commerce platform whose growth ran on measured efficiency.

Its measurement culture is commonly described in industry discussion as incrementality-first: performance spend at that scale makes last-click flattery ruinously expensive, and the reported practice — holdout-style testing and mix modelling guiding allocation, rather than platform-reported ROAS — reads as the modern stack adopted early because the arithmetic of thin-margin commerce forgives nothing else.

What to steal: The thinner your margins, the less you can afford flattering measurement — efficiency businesses adopt incrementality first because they feel the waste fastest.

boAt

The D2C audio brand that added cricket sponsorships to a performance engine.

Its trajectory poses the attribution question in its hardest form: a business raised on trackable performance media, now adding IPL-adjacent sponsorships and celebrity association — spend that emits no clicks and books no platform conversions. The brand's continued investment in it reads as a verdict that the untracked layer builds demand the tracked layer then harvests — precisely the effect last-click structurally miscredits.

What to steal: When you buy unclickable media, your click-based dashboard will call it waste by construction — judge it with MMM and geo-tests, or you'll cancel what was working.

Beyond India

The global lens

eBay

The experiment that embarrassed an industry.

Its economists paused paid search across markets and found brand-keyword ads almost entirely non-incremental — buyers simply took the organic link below — while effects on genuinely undecided audiences were modest. Published in a top economics journal, it remains the canonical proof that a channel can show spectacular attributed ROAS while causing almost nothing.

What to steal: Run the turn-it-off test on your most 'obviously efficient' channel first — obvious efficiency is exactly what harvest looks like.

P&G

The world's largest advertiser, auditing digital in public.

Its leadership's reported cuts of roughly two hundred million dollars in digital spend — targeting fraud-prone and over-frequent placements — with no discernible sales impact became the CFO-friendly incrementality parable: money removed, revenue indifferent, waste located. The sequel mattered too: the funds were reallocated toward reach and quality context, not returned — measurement in service of better spending, not less.

What to steal: Incrementality findings are reallocation orders, not budget cuts — the point is moving money from coincidence to cause.

From theory to Monday morning

How to use it

  1. Publish the over-claim audit

    Sum every platform's claimed conversions for one quarter and divide by actual orders. That single ratio — routinely 2–4x — is the internal argument for the incrementality programme, comprehensible to any CFO in one slide.

  2. Rank channels by harvest suspicion

    Branded search, retargeting, and cashback affiliates first — anywhere the customer had plainly decided before the click. These get the first geo-holdouts or lift tests, and expect the attributed number and the incremental number to part company dramatically.

  3. Run a standing experiment calendar

    Three to six geo-tests a year on the biggest or most contested budget lines, each with pre-registered metrics and minimum detectable effects. One honest readout per quarter beats a decade of model debates.

  4. Stand up MMM and calibrate it

    Start with two-plus years of weekly sales, spend, pricing, and seasonality data. Fit response curves per channel, then anchor the model's lift estimates to your experimental readouts. Re-fit quarterly — an uncalibrated MMM is just attribution with a statistics degree.

  5. Demote attribution formally

    Write the measurement charter: attribution steers creative, audience, and bid decisions within channels; MMM allocates across channels; experiments adjudicate disputes. Any budget move above a set threshold requires experimental or MMM evidence — dashboards alone don't move money.

Watch out

Common mistakes

✕ Optimising the business to platform-reported ROAS.

Fix: Platforms grade their own homework and structurally favour harvest. Treat reported ROAS as a within-channel steering metric, and let experiments and MMM decide what each channel truly adds before money moves.

✕ Switching attribution models and calling it insight.

Fix: A model change reallocates opinion, not causation — the sales didn't move, only the credit did. If a decision hinges on which model you use, that's the signal an experiment is required, not a longer model debate.

✕ Testing incrementality once and framing the answer forever.

Fix: Incrementality decays and drifts — creative wears out, auctions reprice, competitors move. Date-stamp every readout and re-test your biggest lines on a calendar, exactly like the re-scoring discipline in strategy pages.

✕ Judging brand and unclickable media on click instruments.

Fix: Sponsorships, TV, and audio book no conversions by construction, so last-click will convict them of waste every time. Assign them to MMM and geo-experiments from day one, with horizons long enough for demand effects to surface.

Don't just read it

Practice task — 10 minutes

Take one channel you (or your employer) believe is 'obviously working' and design its turn-it-off test on paper: the geography or audience held out, the duration, the primary metric, the minimum effect that would justify current spend, and what you'll reallocate toward if the effect misses. Then answer honestly: what's stopped this test from running already — the statistics, or the politics?

If you remember five things

  • Attribution models are opinions about credit. Incrementality is evidence about cause — the budget question is always the second one.
  • Last-click flatters harvest and starves creation. Data-driven models are better bookkeeping that still can't see the counterfactual.
  • Experiments — lift studies, geo-holdouts, turning it off — are the ground truth, and eBay and P&G made the genre famous for good reason.
  • MMM returned because it needs no cookies: strategic allocation across all media, calibrated against experimental lift.
  • Triangulate by altitude — MMM for strategy, experiments for truth, attribution for tactics — and never let an instrument answer another altitude's question.