Marketing attribution is theatre. Use incrementality instead.

Most marketing attribution is theatre: a convincing performance of certainty staged on data that cannot support it. The dashboards are real. The credit they assign is mostly fiction.

Multi-touch attribution made a promise it could keep in 2015 and cannot keep now. It claims to follow each customer across every touch and assign fractional credit to the channels that moved them. After Apple's App Tracking Transparency arrived in April 2021, with shrinking cookie coverage and word of mouth that never shows up in any pixel, that promise is a coin flip dressed as a spreadsheet. We still run paid ads aggressively. We just stopped pretending the attribution column is the truth.

Here is the replacement stack: incrementality instead of attribution, asking customers directly instead of pixel-stitching, and blended cost-per-job pacing instead of channel-level ROAS worship. It is less satisfying than a clean waterfall chart, and it is right far more often.

Biased by design

Why the dashboard lies

Attribution platforms report what they can see, and after privacy changes they see a shrinking, biased slice of reality. Branded search and retargeting hoover up credit for demand that other channels created, because they sit closest to the conversion. Meanwhile the neighbour's recommendation, the truck parked on the street for 3 days, or the sign on a job site around the corner drove the actual decision and shows up nowhere. The dashboard is not neutral, it systematically over-credits the channels that intercept customers who were already coming.

Those figures come from an EMARKETER and TransUnion survey of 196 US brand and agency marketers in July 2025. The tell is the click window. Meta's standard attribution setting credits an ad with a conversion inside a fixed window after the click, 7 days by default, and in March 2026 Meta narrowed what counts as a click to link clicks alone. Any model like that can only assign credit to a touch it observed inside its window, so anything that creates demand without an immediate trackable click, which is most of what actually fills a service calendar, is invisible to it by design. Treating that model as ground truth is how teams defund the channels building their brand and overfund the ones harvesting it.

There is exactly one question worth answering: if we turned this channel off, would we lose jobs we would otherwise have booked? Attribution guesses at it. Incrementality measures it, by running a controlled experiment and reading the lift. We aim to put every paid channel we run through a test on a regular rotation.

1. 01
   Geo-blackout test

   Turn a channel completely off in a matched set of the cities you serve while leaving it on in the rest. Compare booked jobs in the dark cities to the lit ones. The gap is the channel's true contribution, with no pixel required.

2. 02
   Percent-holdout test

   Withhold ads from a random percentage of your addressable audience and measure the difference in conversion between the held-out group and the exposed group. The lift is incremental value, full stop.

3. 03
   Read lift, not last click

   The output is a single honest number: how many conversions this channel caused. If turning a channel off changes nothing, it was harvesting demand other channels created, and the budget belongs elsewhere.

Incrementality is mainstream now precisely because attribution broke. In the same EMARKETER and TransUnion survey, 52% of US brand and agency marketers said they already use incrementality tests and experiments, and EMARKETER describes experiments as having moved from niche practice to mainstream, pushed by privacy limits and pressure to prove ad budgets do real work. You do not need a data-science team to start. You need the discipline to turn one channel off and watch what happens.

Ugly, honest, decision-grade

Self-reported attribution, coded monthly

The cheapest measurement you can run is a free-text field on your quote form, and the same question asked on every call that comes in: how did you hear about us? It is qualitative, messy, and biased toward the last memorable touch, and it still beats pixel-stitching for the word-of-mouth channels no tracker can see. Customers will tell you about the referral and the yard sign that your dashboard erased.

The discipline is in the coding. Once a month, read every free-text answer and bucket it into a consistent taxonomy. Patterns emerge fast: a channel your dashboard rates near zero will show up again and again in the words of people who actually booked. Combined with incrementality tests, what customers tell you triangulates the truth from 2 independent directions, which is the closest thing to certainty this domain allows.

If you cannot trust channel-level attribution, stop making channel-level promises. We pace budget to a blended cost-per-job target, total marketing spend over total jobs booked, and let the channels fight inside that envelope. Branded search does not get to claim the calls it merely intercepted. The portfolio gets judged as a portfolio.

1. 01Set a blended cost-per-job target the margins can actually support, tied to your average job value and how much repeat work a customer brings, not vanity.
2. 02Pace total spend to hold that blended number, scaling up when it improves and pulling back when it drifts.
3. 03Use quarterly incrementality tests to decide which channels deserve more of the envelope, not the in-platform ROAS.
4. 04Let what customers tell you on the phone break ties on the word-of-mouth channels no experiment cleanly isolates.

Audience quality sits upstream of all of it. A paid custom audience built from a scored account list, the approach in the lead intelligence playbook, concentrates spend on accounts that already passed fit, which makes the blended number easier to hold. The hard part of blended cost per job is holding the nerve when a single channel looks like it is dragging the average. The instinct is to cut it on its in-platform numbers, but those are the numbers you already decided not to trust. Discipline means letting the quarterly experiments, not the dashboard's panic, decide what gets cut. A channel that builds demand will look inefficient on last-click forever, and killing it is how teams accidentally starve the top of their own funnel.

The objection to incrementality is always the same: it sounds like a data-science project we do not have the team for. It is not. A 1 person marketing team, or an owner with a spreadsheet, can run a useful testing rhythm with a calendar and the willingness to turn a channel off for a couple of weeks. Here is a cadence that works for a typical paid portfolio.

1. 01
   Pick one channel per quarter

   Do not test everything at once. Rotate, so each major paid channel gets one clean experiment per year, plus a re-test on anything that surprised you. 4 channels, 4 quarters, one calendar.

2. 02
   Choose the test type

   Use a geo-blackout where your service areas are clean and matched, and a percent-holdout where the cities blur into each other but you control the audience. Pick the one that fits the channel, not the one that is easier to chart.

3. 03
   Run for 2 to 4 weeks

   Long enough to clear the lag between the first search and the booked job, short enough that the opportunity cost of going dark stays bounded. Pre-commit to the window so you do not bail the moment call volume dips.

4. 04
   Read lift, then reallocate

   Compare the held-out or dark group against the exposed one, write down the incremental number, and move budget toward whatever proved causal. One honest number per quarter compounds into a portfolio you actually understand.

4 experiments a year is enough to catch the biggest lies your dashboard is telling you. It will not make measurement perfect, because nothing will, but it replaces confident fiction with rough truth, and rough truth is what good budget decisions are made of.

Mostly subtraction

What to stop doing on Monday

Most of the upgrade is subtraction. Stop reading the paid ads platform dashboard as if it reports reality, stop letting a short click window set budget, and stop giving the team attribution tools that are precise and wrong. Precision without accuracy is the most expensive kind of confidence.

Then redirect that energy to the things a customer actually decides on. Incrementality tells you a channel works, but the landing pages, the creative design, and the video pipeline behind the click are what turn proven reach into booked jobs. The landing page system post shows how we instrument a page section by section, and the video pipeline post covers how one shoot feeds the many ad variants a test needs. Measurement that is honest about channels frees you to obsess over the creative, which is where the real advantage in paid actually lives. That is the principle running through everything we build: measure what is causal, build what converts, and ignore the theatre in between.

Attribution dashboards are precise and wrong. I will take a holdout test that is roughly right every single time.

Shivam Bindal