Beyond the walled garden silo – true ROAS across platforms

Google says your campaign generated 150 sales.

Amazon claims 200. Meta swears it drove 180.

Add them up and you get 530 conversions. Check your actual revenue and you’ll find you sold 250 units total.​

This is the walled garden nightmare every e-commerce marketer lives with. Three platforms, three separate attribution systems, three versions of reality, and zero idea which one reflects actual performance.​

Welcome to the single biggest problem in modern digital advertising: data silos that make accurate ROAS calculation impossible.​

The walled garden definition nobody explains properly

A walled garden in advertising means a closed ecosystem where one platform controls everything: ad inventory, user data, targeting capabilities, and performance measurement.​

Google. Meta. Amazon. Apple. TikTok.

These companies built massive user bases, then locked advertisers inside their systems. You can run campaigns. You can see some data. But you can’t export granular information that works outside their walls.​

The platforms designed it this way deliberately. Not out of malice, though it certainly benefits them, but because controlling the entire stack creates better user experiences and more accurate targeting.​

But here’s the problem that nobody talks about enough.

Each walled garden uses last-click attribution by default. Google takes credit if someone clicked your Google ad last. Amazon claims the sale if they bought after seeing your Amazon ad. Meta counts it as a conversion if their platform was the final touchpoint.​

So when someone sees your Facebook ad, searches on Google, reads reviews on Reddit, then buys on Amazon three days later, who gets credit?​

Everyone. And nobody. Depending on which dashboard you’re looking at.​

This wouldn’t matter if you only ran campaigns on one platform. But you don’t. Most e-commerce brands spend across Google, Amazon, Meta, and increasingly retail media networks like Walmart Connect or Target’s Roundel.​

That means you’re getting 4-6 different “truths” about performance, all claiming credit for the same conversions.​

Why this matters more now than ever

83% of digital ad spend flows through walled gardens as of 2025. By 2027, that number hits 90%.​

You can’t avoid them. You need their reach. Google controls over 63% of web traffic. Meta has 3 billion daily active users. Amazon owns 40% of U.S. e-commerce.​

But as walled gardens consolidate power, the attribution problem gets worse, not better.​

Privacy changes accelerated this. iOS 14+ killed Facebook’s pixel tracking. Chrome’s cookie deprecation removed cross-site tracking. GDPR and CCPA limited data collection.​

The result? Platforms retreated further into their walled gardens, using first-party data exclusively and providing even less visibility to advertisers.​

This creates a brutal problem for marketers managing budgets.

You’re spending $100,000 monthly across Google, Amazon, and Meta. Each platform reports amazing ROAS. But when you look at actual revenue growth, the numbers don’t add up.​

Are you overspending? Underspending? Should you shift budget from Google to Amazon? From Meta to retail media networks?​

You literally can’t tell, because the data lives in silos that don’t communicate.​

The Amazon attribution breakthrough

Amazon built something interesting that most advertisers ignore – Amazon Attribution.​

It’s a free tool available to brand-registered sellers that tracks how off-Amazon marketing drives Amazon sales. You create special tracking links for your Google Ads, Facebook campaigns, or email marketing. When someone clicks that link and buys on Amazon, you see exactly which external channel drove the sale.​

This sounds simple. It’s revolutionary for walled garden attribution.​

Before Amazon Attribution, running Google Ads to send traffic to Amazon product pages was blind guessing. You spent money, hoped for sales, but couldn’t prove connection.​

Now you can see:

• Which Google keywords drive Amazon purchases​
• How Facebook traffic converts on Amazon​
• Whether influencer promotions lead to sales​
• What your actual cross-platform ROAS looks like​

The bonus is financial too. Amazon’s Brand Referral Bonus program gives you back up to 10% of sales driven by external traffic. So not only do you get attribution data, you get paid for bringing customers to Amazon from other channels.​

But here’s where it gets complicated.

Amazon Attribution only tracks Amazon sales. If someone clicks your Google ad, browses Amazon, then buys on your Shopify store three days later, Amazon Attribution doesn’t see that conversion.​

You still have the fundamental walled garden problem, just shifted slightly.​

How retail media networks changed everything

Retail media networks are the fastest-growing advertising channel in 2025.​

Walmart, Target, Kroger, CVS, Best Buy—basically every major retailer built ad platforms that let brands promote products using the retailer’s first-party purchase data.​

The spend is insane. Retail media hit $166 billion globally in 2025. It’s growing faster than Google or Meta advertising.​

Why?

Because retail media networks have what walled gardens like Google and Facebook lost: actual purchase data.​

When you run ads on Walmart Connect, you’re targeting people based on what they actually bought, not what they clicked or browsed. The attribution is clean because it happens within a closed loop. ad impression, click, purchase, all tracked within Walmart’s ecosystem.​

But. and this is critical. retail media networks created NEW walled gardens.​

Now you’re not just managing Google, Meta, and Amazon attribution. You’ve added Walmart, Target, Kroger, Instacart, and whoever else. Each one has its own dashboard, reporting standards, and attribution methodology.​

The silo problem didn’t get solved. It multiplied.​

What true cross-platform attribution actually requires

Let me be blunt about what solving this problem takes.

First, you need unified data infrastructure. Not just exporting CSV reports from each platform and putting them in a spreadsheet. I mean building a data warehouse that ingests event-level data from every advertising platform in real-time.​

This is hard. Like, enterprise-level data engineering hard.​

Google Ads API. Facebook Marketing API. Amazon Advertising API. Walmart Connect. Target Roundel. Instacart. Each platform has different data structures, update frequencies, and access limitations.​

You need:

• Technical resources to build and maintain integrations​
• Storage infrastructure to handle billions of events​
• Processing capabilities to join data across platforms​
• Identity resolution to match the same user across systems​

That last one is the killer.​

Identity resolution means figuring out that the person who clicked your Google ad, watched your YouTube video, saw your Instagram Story, and bought on Amazon is the same human.​

Without cookies or cross-platform tracking pixels, this is nearly impossible at scale. The platforms don’t share user identifiers. They can’t, due to privacy regulations.​

So you’re left with probabilistic matching, using signals like device type, timestamp, location, and behavioral patterns to guess when events belong to the same person.​

It works. Sort of. Accuracy ranges from 60-85% depending on methodology.​

Second, you need a unified attribution model.​

Last-click attribution is bullshit for cross-platform campaigns. But what replaces it?​

Multi-touch attribution tries to credit every touchpoint in the customer journey proportionally. Someone sees a Facebook ad (20% credit), searches on Google (30% credit), reads reviews (10% credit), clicks an Amazon ad (40% credit).​

Sounds fair. Impossible to implement accurately across walled gardens that don’t share conversion data.​

Data-driven attribution uses machine learning to weight touchpoints based on their contribution to conversions. Google’s version only works for Google campaigns. Amazon’s only covers Amazon. Meta’s only tracks Meta.​

Marketing mix modeling (MMM) is making a comeback. Instead of tracking individual users, MMM uses statistical analysis to correlate ad spend with sales at an aggregate level.​

Run $50K in Google Ads one week, $30K the next. Did sales patterns shift proportionally? That’s MMM.​

It works for high-level budget allocation but misses channel-level optimization opportunities.​

The technical implementation nobody wants to discuss

Okay, here’s what actually building cross-platform attribution looks like.

Step one: Set up server-side tracking.​

Client-side pixels (JavaScript tags on your website) are dead or dying. Browsers block them. Users opt out. iOS kills them.​

Server-side tracking means your server sends event data directly to advertising platforms via their APIs. This bypasses browser restrictions and improves data accuracy.​

But it requires:

• Custom development work​
• Server infrastructure to handle tracking requests​
• Consent management to stay compliant​
• Maintenance as APIs change​

Step two: Build a customer data platform.​

Your CDP becomes the source of truth for customer behavior. It ingests:​

• Website events (pageviews, clicks, purchases)​
• Email engagement data​
• CRM records​
• Advertising platform callbacks​
• Retail media network conversions​

The CDP stitches these together into unified customer profiles using identity resolution.​

Step three: Implement conversion APIs for each platform.​

Google Conversion API. Meta Conversions API. Amazon Attribution. TikTok Events API.

These let you send conversion data back to advertising platforms server-to-server, bypassing tracking limitations.​

Critical for maintaining campaign optimization when client-side tracking fails.​

Step four: Build unified reporting dashboards.​

Pull data from your CDP, apply your chosen attribution model, and visualize actual cross-platform performance.​

This is where you finally answer: What’s my TRUE ROAS when accounting for cross-platform customer journeys?​

Spoiler: It’s almost always lower than what any single platform reports. But at least it’s real.​

The tools that claim to solve this

Improvado focuses on marketing data integration, connecting 500+ data sources to unified dashboards.​

Northbeam specializes in cross-platform attribution for DTC brands, using sophisticated modeling to track customer journeys.​

Triple Whale aggregates e-commerce data from Shopify, Amazon, Google, and Meta into one interface.​

Rockerbox builds multi-touch attribution across walled gardens using proprietary tracking and modeling.​

MetaRouter handles server-side event streaming with privacy-first infrastructure for retail media networks.​

Do they actually work?

Kind of. They make the problem manageable, not solved.​

You still hit walled garden limitations. You still deal with identity resolution gaps. You still have attribution model debates.​

But these tools get you 80% of the way there, which is infinitely better than managing seven separate dashboards and guessing.​

What this means practically for budget allocation

Here’s the honest advice.

Stop optimizing for platform-reported ROAS. It’s a lie. Not intentionally, the platforms believe their own attribution, but it’s not reality.​

Build incrementality tests instead. Turn off one channel completely for 2-4 weeks. Did total revenue drop proportionally? That’s the channel’s real contribution.​

Painful. Scary for performance marketers. But it’s the only way to know true impact when attribution systems lie to you.​

Use marketing mix modeling for strategic decisions. Where should you allocate the next $100K, Google, Amazon, retail media, or Meta? MMM answers this better than last-click attribution.​

Deploy Amazon Attribution aggressively. Even if it only tracks Amazon sales, that’s huge for most e-commerce brands. Knowing which external channels drive Amazon conversions is worth the setup effort.​

Invest in first-party data infrastructure. Build email lists. Launch loyalty programs. Create accounts that give you direct relationships with customers.​

As walled gardens tighten, your first-party data becomes your only cross-platform view.​

Accept imperfect attribution. You will never have perfect visibility. Ever. The technology and privacy landscape won’t allow it.​

The goal isn’t perfection. It’s “better than guessing”.​

The future is more walled gardens, not fewer

Every platform is building walls higher, not tearing them down.​

Netflix launched an ad tier with its own measurement system. Disney+, Paramount+, and Max each have separate attribution. TikTok barely shares data outside its ecosystem.​

Retail media networks will consolidate but remain fragmented. You’ll have three giant retail media platforms instead of twenty smaller ones, but they’ll still be walled gardens.​

The trade-off gets worse. More control for platforms, less transparency for advertisers.​

But the scale and targeting they offer make them impossible to avoid. You need Google’s reach. You need Amazon’s purchase intent. You need retail media’s closed-loop attribution.​

So you build systems to manage the mess. You use tools that bridge silos imperfectly. You run incrementality tests to validate what attribution systems claim.​

And you accept that measuring true cross-platform ROAS is less science, more educated guess backed by expensive data infrastructure.​

Welcome to modern marketing attribution. It’s broken. It’s getting worse, and there’s no way out.​