Paid Media Attribution: Knowing What Actually Works

Paid Media Attribution: Knowing What Actually Works

Every ad platform will tell you it drove your results. Meta, Google, TikTok, and Pinterest are each claiming credit for the same conversion. The problem is not that they are lying — it is that they are all using attribution models that maximize reported return on ad spend (ROAS) within their own ecosystem. Your job is to see past that.

The Attribution Illusion

Last-click attribution, the default model most small businesses use, awards 100% of credit to the final touchpoint before conversion. This systematically over-credits bottom-of-funnel channels (Google Branded Search, retargeting) and under-credits the channels that created awareness and desire.

The result: brands gut their prospecting budgets because they "don't perform" and then wonder why top-line growth stalls. What they killed was the engine that filled the retargeting pools their numbers depended on.

The Three Layers of Attribution Reality

Layer 1 — Platform attribution: What the platforms report. Inflated, self-serving, and useful only for optimization within a single channel.

Layer 2 — Blended analytics attribution: What tools like GA4, Northbeam, or Triple Whale show after deduplication. Better, but still modeled. These use statistical heuristics, not causal measurement.

Layer 3 — Incrementality: What actually happened versus what would have happened without the spend. This is truth. It requires experiments.

How to Run an Incrementality Test

1. Define your test metric (purchases, signups, revenue per user — one metric)
2. Create a holdout group of 10-20% of your audience, excluded from the campaign
3. Run the campaign for at least one full purchase cycle (minimum 2 weeks for most e-commerce)
4. Compare conversion rates: holdout vs. exposed
5. Calculate lift: (Exposed CVR - Holdout CVR) / Holdout CVR × 100

A channel showing 8× ROAS in platform but 1.2× incremental lift is mostly claiming credit for organic demand — not creating it.

Media Mix Modeling: When to Use It

MMM is the right tool when:

• You have 18+ months of weekly spend and revenue data
• Your business has sufficient scale (typically $1M+ annual media spend)
• Privacy changes have degraded pixel-level measurement
• You need channel allocation decisions at the strategic level, not the tactical

MMM is not the right tool for:

• Testing a new creative
• Optimizing bids week-to-week
• Understanding individual user journeys

Channel-Specific Attribution Traps

Meta: Default 7-day click / 1-day view windows dramatically inflate conversions. Set to 7-day click / 0-day view for most accuracy. Compare Meta-reported purchases against actual Shopify/Stripe data weekly.

Google: Smart Bidding uses Google's own attribution model to set bids. If your attribution model is wrong, your bidding model optimizes toward the wrong outcomes. Audit conversion actions quarterly.

TikTok: View-through attribution is baked into default reporting. TikTok's median purchase window is much shorter than its default 7-day view window. Test 1-day view or suppress view-through entirely.

Building a Measurement Framework That Works

1. Set a north star metric that maps directly to revenue — not impressions, CPMs, or even CTR
2. Run baseline incrementality tests on your top two channels before scaling
3. Use blended analytics as directional, not authoritative — compare it to actual sales data every week
4. Test creative, not just spend — 60% of campaign performance variance comes from creative quality, not bid strategy
5. Review your attribution windows on every platform every quarter — platforms change defaults and it affects your numbers silently

Attribution is not an analytics exercise. It is a capital allocation discipline. Wrong attribution means wrong budget decisions compounded across 12 months.