Most attribution models are built to give you an answer. The problem is they're often giving you a confident wrong answer. And confident wrong answers are more dangerous than no answer at all, because they shape decisions.
I've seen this play out dozens of times. A team leans into last-touch attribution, paid search gets all the credit, organic and brand get cut, and six months later pipeline is down and nobody understands why. The model worked great. The business didn't.
Why Last-Touch Fails (Even When the Numbers Look Good)
Last-touch attribution is intuitive: whoever closed the deal gets credit. It's clean, it's easy to report on, and it maps to how salespeople think. That's exactly why it's seductive and exactly why it leads you astray.
Buyers don't move in straight lines. A prospect might read three blog posts over two months, attend a webinar, talk to a friend who uses your product, and then click a retargeting ad the week they're ready to buy. Last-touch gives 100% of the credit to that final ad. The blog, the webinar, the word-of-mouth: invisible.
The channels that look worst in attribution are often the ones doing the most upstream work.
Cut those channels, and you cut the thing feeding everything else. Conversions don't disappear right away, so you don't notice immediately. By the time you do, the damage is already done.
What I Use Instead
I don't think there's a perfect attribution model, so I've stopped looking for one. What I do instead is run multiple lenses in parallel and use them to triangulate, not to get a definitive answer.
- First-touch and last-touch together. Compare them. Where they diverge, there's a story worth understanding.
- Pipeline surveys. Ask prospects directly: "How did you first hear about us?" Self-reported data is imperfect, but it catches things tracking can't (dark social, word-of-mouth, podcasts).
- Holdout experiments. For any significant channel investment, run a geo or audience holdout. Measure the actual lift. It's the only way to get close to causality.
- Cohort analysis. Look at how cohorts that came through different channels perform over time. Cost per acquired customer is often more telling than cost per lead.
The Honest Version
Most of us are operating with imperfect data and making judgment calls. That's fine. The mistake is pretending otherwise. Attribution models should inform your instincts, not replace them.
If your paid search rep is telling you the numbers look great and your gut says something's off, trust the discomfort. Run a holdout. Pull the cohort data. Look for the thing the model can't see.
The goal isn't perfect attribution. It's good enough attribution, paired with healthy skepticism about what the numbers are actually telling you.