Attribution is broken. Here is what we do instead.

Ask three attribution tools which channel drove a sale and you will get three answers. Google Analytics credits the last non-direct click. Facebook claims the conversion under a seven-day view-through window. Your CRM says it was the sales call. All three are looking at the same customer, and all three are confident.

This is not a tooling problem you can fix by buying a fourth tool. Attribution, the idea that you can assign fractional credit to each touchpoint and sum it to the truth, is built on assumptions that stopped being true. Cookies expire or get blocked. iOS strips tracking parameters. People research on their phone and buy on a laptop. Someone hears about you from a friend, searches your brand name, clicks a Google ad, and now paid search takes the credit for a referral it had nothing to do with.

Why the model fails, mechanically

The core promise of attribution is a clean chain: impression, click, visit, conversion, all tied to one identifiable person across devices and time. Every link in that chain has broken.

• Identity is fragmented. A single buyer shows up as three or four anonymous users across a phone, a work laptop, and a home computer. Without a login, no tool stitches them reliably.
• The tracking layer leaks. Ad blockers, privacy browsers, and platform restrictions drop a meaningful share of events before they are ever recorded. You do not know which share, which is the problem.
• Windows are arbitrary. A seven-day click and one-day view window is a business decision made by the ad platform that profits from the attribution. Change the window, change the "results."
• Brand search launders credit. Paid search on your own brand name catches demand created everywhere else and reports it as its own.
• Offline is invisible. The phone call, the referral, the billboard someone remembered: none of it produces a click, so the models pretend it did not happen.

Each platform's model is also self-serving. Facebook's attribution makes Facebook look good. Google's makes Google look good. That is not a conspiracy, it is just what happens when the scorekeeper is also a player.

Attribution tries to answer "who gets credit." The better question is "what happens to the business when we change the input."

The trap of precise wrong numbers

The seductive thing about attribution dashboards is that they produce decimals. A 3.7x return on ad spend feels more trustworthy than "we think this is working." But precision is not accuracy. A number can be specific to two decimal places and still be off by half, because the error is in the model, not the arithmetic.

We have watched teams cut a channel because its last-click number looked weak, then watch total revenue fall while the dashboard for the remaining channels looked better than ever. The demand did not disappear. It got reassigned to whatever was left standing closest to the sale. The report improved and the business got worse. If you optimize toward a broken map, you will confidently walk off a cliff.

What we do instead

We stopped trying to reconstruct the perfect customer journey. Instead we measure the thing we can actually verify: does the whole business move when we change what we spend. Three practices carry most of the weight.

1. Watch the blended number first

Total spend across all channels, divided into total new revenue, over a consistent period. No per-channel credit games. If we increase spend by 20 percent and blended new revenue does not respond, the channel-level dashboards claiming success are lying, and it does not matter which one. The blended view is coarse, but it is honest. It is the only number no platform gets to grade for itself.

2. Run holdouts and geo tests

The cleanest way to know if a channel works is to turn it off somewhere and watch what happens. Pause paid search in three comparable markets, keep it running in three others, and compare revenue over four to six weeks. This is real evidence, not modeled credit. It is slower and it costs some short-term revenue to run, but it answers the question attribution only pretends to answer. Incrementality is the whole game: not "did this touchpoint appear before a sale," but "would the sale have happened anyway."

3. Ask buyers directly

A one-question post-purchase survey ("How did you first hear about us?") is unglamorous and shockingly useful. It captures the referral, the podcast, the word of mouth that no pixel ever saw. It is self-reported and imperfect, so we treat it as a directional cross-check against the blended and test data, not as gospel. When the survey, the holdout, and the blended trend all point the same way, we act. When they disagree, we dig before we spend.

Put together, these give us a triangulated picture: one coarse and honest metric, one causal test, and one human signal. None is precise. Together they are accurate enough to make real decisions, which is more than any attribution model can claim.

What this means for how we get paid

We are a performance partner, so our compensation is tied to outcomes, not to a dashboard we control. That constraint is clarifying. If we could pick our own attribution model, we could make any campaign look like a winner and get paid on fiction. We cannot, and we do not want to. We would rather be measured on the blended business result, because that is the number a client actually feels in their bank account.

This also changes what we build. Our reporting is designed to surface the blended trend and the test results up front, and to treat platform-reported ROAS as a diagnostic input rather than the headline. A channel's own scorecard tells us how that channel is pacing internally. It does not tell us whether the business is growing. We keep those two questions separate on purpose.

The honest position is that you cannot fully know which touch caused which sale, and you should stop pretending you can. What you can know is whether your spending moves the business. Measure that, test into it, and ask your customers. It is less tidy than a waterfall chart with clean percentages, and it is far more likely to be true.

If you want a partner who reports the number you can feel instead of the number that flatters the channel, apply for partnership. If you would rather start smaller, run your site through the free grader and see what it turns up.