Stop Stressing About Attribution – Here’s What Actually Works
Step-by-step guide to improving your attribution
Perfect attribution doesn’t exist, but a good enough system will make your Return on Investment (ROI) better than your competitors and your Cost Per Acquisition (CPA) cheaper—because you’ll stop scaling the wrong things.
1. Build an aggregated view without platform data
Meta says it drove £1m in revenue. Google says it drove £1m in revenue. You made… £1.5m revenue total. Each platform takes full credit for multi-touch conversions, so comparing them directly is useless.
What to do today
1️⃣ Continue to use platform data to compare performance within a single marketing channel.
2️⃣ Ignore platform data when comparing different marketing channels—they’re all inflating their impact.
3️⃣ If spend on a platform increases but blended ROAS doesn’t, that spend isn’t incremental. Be volatile with spend, lose in the short term, learn fast, and win the long game.
2. Layer Multiple Attribution Methods
No single model is perfect. Combining multiple helps you understand channels. Here’s a good scope of different attribution you can implement (source):
What to do today
1️⃣ Stop relying on a single attribution model—implement at least three (one from each category is ideal).
2️⃣ Treat surveys and tracked attribution as a directional signal, lift experiments as truth, and Marketing Mix Modelling (MMM) for ongoing strategic decisions.
3️⃣ If two models disagree wildly, investigate before shifting spend.
💡 Is this going over your head, hire a Marketing Data Analyst ASAP.
3. Run Lift Tests to Find the Truth
Attribution models predict impact—lift tests prove it. If you don’t know what’s actually working, you’re probably burning budget somewhere.
What to do today
1️⃣ Run a geo-lift test: Pause spend in a region for a period and track revenue changes.
2️⃣ Use conversion lift tests to see what’s actually driving sales.
3️⃣ Want the fast-track to truth? If you can afford it, turn off each channel for a month on rotation. It’s painful, but your decision-making after this cycle will be better than your competitors’.
4. Use Marketing Mix Modelling (MMM) for Long-Term Decision Making
MMM isn’t perfect, but it’s a great option for a full-picture view—especially as tracked attribution continually gets harder.
What to do today
1️⃣ If you’re spending enough to run MMM, hire someone to build one - they’ll deliver more value than another marketing manager, I promise.
2️⃣ Start small: Lightweight Bayesian MMM can give you a rough idea of what’s working, then build up to a full featured, powerful modelling library such as PyMC.
3️⃣ Validate predictions with incrementality tests—if the results don’t align, tweak the model.
The Bottom Line
❌ No one has perfect attribution, and platform-reported ROAS is a lie.
✅ If you’re past Step 2 above, you’re already ahead of most marketers.
🚀 Layer multiple views, test incrementality, and use MMM for long-term insights.
📉 Stop chasing perfection—build a system that’s good enough, make better decisions, and then focus on optimising your actual marketing strategy.