A customer sees your Meta ad on Monday. Searches your brand on Google on Wednesday. Clicks an Amazon Sponsored Products ad on Friday and buys. Which ad gets credit for the sale? In last-click attribution, Amazon gets 100%. In first-click, Meta gets 100%. Both answers are wrong, because the real answer is that all three touchpoints contributed. Multi-touch attribution attempts to distribute credit more accurately across the full journey.
The stakes are real. If you attribute all your sales to Amazon (last-click), you'll shift budget from Meta and Google to Amazon. Then Amazon sales drop because you stopped feeding the top of the funnel. This reallocation death spiral has cost brands millions. MTA is how you avoid it.
Attribution Models Explained
| Model | How Credit Is Assigned | Pros | Cons |
|---|---|---|---|
| Last Click | 100% to the last touchpoint before conversion | Simple, easy to implement, available in every platform | Over-credits bottom-funnel channels, ignores awareness and consideration |
| First Click | 100% to the first touchpoint | Values awareness and discovery | Ignores the closing touchpoints, rarely used as primary model |
| Linear | Equal credit to all touchpoints | Simple, acknowledges every touchpoint | Treats all touchpoints as equally important (they're not) |
| Time Decay | More credit to touchpoints closer to conversion | Reasonable assumption that recent interactions matter more | Still undervalues awareness touchpoints that started the journey |
| Position-Based (U-shaped) | 40% first touch, 40% last touch, 20% split across middle | Values both discovery and closing | Arbitrary weighting, doesn't adapt to actual data |
| Data-Driven (Algorithmic) | Machine learning assigns credit based on observed conversion patterns | Most accurate, adapts to your actual data | Requires high conversion volume, black-box results, needs advanced tools |
Why MTA Is Hard in 2026
Multi-touch attribution has always been difficult. It's gotten harder because of three converging trends:
Privacy restrictions have killed cross-device tracking
Apple's ATT (App Tracking Transparency) lets iOS users opt out of tracking. Over 75% do (Flurry Analytics, 2024). GDPR, CCPA, and the gradual erosion of third-party cookies mean you can no longer follow a single user across devices and platforms the way you could in 2019. MTA models that depend on user-level tracking are increasingly incomplete.
Walled gardens don't share data
Amazon, Google, and Meta each have their own attribution systems. They report their own results using their own attribution windows and models. None of them share user-level data with each other. Amazon Marketing Cloud (AMC), Google's Attribution tools, and Meta's Conversions API each tell you what happened within their ecosystem but not across ecosystems.
Customer journeys are getting longer and more fragmented
The average B2B buyer interacts with 6-8 touchpoints before purchasing (Salesforce, 2024). For CPG, the journey from awareness to Amazon purchase might span TikTok, Instagram, Google Search, a blog article, an email, and finally an Amazon search. Tracking and attributing all of those accurately is an unsolved problem at scale.
MTA Approaches That Work Today
Data-driven attribution within platforms
Google's data-driven attribution model (available in GA4 and Google Ads) uses machine learning to assign credit across Google touchpoints. Meta's attribution tools do the same within Meta. These are solid for intra-platform decisions but don't help with cross-platform allocation.
Unified measurement platforms
Tools like Measured, Northbeam, Rockerbox, and Triple Whale attempt to unify data across platforms by ingesting data from all your ad accounts, your website analytics, and your transaction data. They build cross-platform attribution models that approximate the full customer journey. Accuracy varies, but they're better than platform-reported metrics for allocation decisions.
Incrementality testing (the gold standard)
Rather than trying to attribute individual conversions, incrementality testing measures the causal impact of each channel. Geo-holdout tests (turn off Meta ads in Dallas, keep them running in Houston, compare sales lift) give you a direct measure of what each platform actually contributes. This is the most reliable input for retail media mix decisions, though it's slower and more expensive to run than model-based MTA.
Media mix modeling (MMM)
Statistical models that correlate advertising spend with business outcomes at an aggregate level (no user-level tracking needed). MMM works in the post-privacy world because it uses spend and revenue data, not individual user data. Google's Meridian and Meta's Robyn are open-source MMM tools. The downside is that MMM requires 1-2 years of historical data and is better at long-term allocation than day-to-day optimization.
MTA for CPG Brands
CPG brands face unique attribution challenges because conversions often happen off-platform (in a retail store or on Amazon) rather than on your website:
- Amazon Marketing Cloud (AMC) connects Amazon ad exposure to Amazon purchases, including cross-campaign attribution within Amazon's ecosystem. It's the closest thing to MTA for Amazon advertising.
- Retailer data clean rooms (Walmart Luminate, Kroger 84.51) match ad exposure to in-store purchases in a privacy-compliant way. These are emerging but available for brands with retailer relationships.
- Coupon and promo tracking assigns unique promo codes or landing pages to each channel for direct attribution. Low-tech but effective for measuring specific campaigns.
- Brand lift studies measure whether advertising exposure increased brand awareness, consideration, and purchase intent. Meta, Google, and Amazon all offer brand lift measurement for campaigns above minimum spend thresholds.
Example: Discovering the Meta-to-Amazon Pipeline
A CPG brand selling protein bars was spending $200K/month across Meta ($80K), Google ($40K), and Amazon ($80K). Amazon's last-click ROAS was 5.8x. Meta's was 1.9x. The natural conclusion: shift budget from Meta to Amazon. They tested this by cutting Meta spend 30% in three markets while keeping control markets unchanged. Four weeks later, Amazon sales in the test markets dropped 14%. Meta campaigns weren't just driving DTC sales — they were creating brand awareness that converted on Amazon when shoppers searched the brand name. The brand installed Northbeam to unify cross-platform measurement and discovered Meta's true incremental ROAS (including Amazon conversions it influenced) was 3.4x, not 1.9x. They restored Meta budget and rebalanced toward a cross-platform view. This is why retail media ad spending in the US reached $62 billion in 2025 (eMarketer): brands are investing more in retail media, but smart ones use MTA to understand how all channels feed each other.
Frequently Asked Questions
What's the best attribution model for CPG brands?
No single model is "best." Use data-driven attribution within each platform for tactical optimization (bid adjustments, creative decisions). Use incrementality testing for strategic decisions (budget allocation across platforms). Use MMM for long-term planning. The combination of all three gives you the most complete picture.
Is last-click attribution ever appropriate?
Last-click is fine for bottom-funnel, single-platform analysis (e.g., optimizing Amazon Sponsored Products bids based on conversion data). It's dangerous for cross-platform budget allocation because it systematically overvalues bottom-funnel channels and undervalues awareness channels.
How much does a proper MTA setup cost?
Platform-native attribution tools (GA4, Amazon Attribution, Meta Attribution) are free. Unified measurement platforms (Measured, Northbeam, Rockerbox) range from $1,000 to $10,000+/month depending on ad spend volume. Incrementality testing costs vary, but a basic geo-holdout test requires redirecting 10-20% of budget for 4-8 weeks as a "control" market.
Should I build custom attribution or buy a tool?
Buy. Unless you have a dedicated data science team and 6+ months to build and validate models, off-the-shelf tools from Measured, Northbeam, or similar platforms will give you faster, more reliable results. The attribution problem is complex enough that specialized vendors with cross-industry benchmarks outperform most custom builds.
Getting attribution right is the difference between optimizing your ad spend based on reality and optimizing based on each platform's self-reported fiction. Our Paid Media team sets up unified measurement frameworks for CPG brands advertising across Amazon, Meta, and Google. Let's build your attribution stack.