The attribution problem no one wants to name
Ad platforms compete for credit. Meta wants you to believe Meta drove the conversion; Google wants Google credited; TikTok claims the same purchase. Last-click gives credit to whichever was touched last. First-click gives credit to whichever was touched first. Multi-touch attribution uses each platform's own definition of contribution — which inevitably favours the platform doing the reporting.
The result: aggregate platform-reported ROAS routinely overstates real performance. A campaign showing 6× on dashboards might have incremental ROAS — what you wouldn't have earned without that spend — closer to 2×. The difference between 6× and 2× is the difference between scaling spend and cutting it.
Quick test: Add up reported conversions across all your ad platforms for last month. Compare to actual orders from your back-end. If the platform total exceeds back-end conversions by more than 15%, you have attribution double-counting and your ROAS numbers are unreliable.
How we measure incrementality
1. Geo holdout tests
Pick two comparable markets. Run the full paid stack in one; zero paid spend in the other. Run for 30 days. Compare total revenue (not platform-attributed revenue). The difference is your real incremental contribution.
Example: if the unpaid market generates 80% of the revenue of the paid market, real incrementality is roughly 20% of total revenue. Platform-reported ROAS of 6× against total revenue becomes incremental ROAS of roughly 1.2× — a very different conversation about whether to scale spend.
2. Branded search pause tests
Most established brands spend significant amounts bidding on their own brand name in Google Ads. Pause it for 2 weeks. If brand revenue doesn't drop materially, that spend was non-incremental — paying Google to intercept traffic already coming to you. For most established brands, 60–80% of branded search spend turns out to be non-incremental.
3. Dashboard reconciliation
Sum reported conversions across Meta, Google, TikTok, email. Compare to actual orders from back-end CRM. The gap reveals double-counting. A 15–30% gap is common and indicates platform-level ROAS optimisation is happening against inflated numbers.
What replaces broken ROAS
Media mix modelling (MMM)
For larger accounts (typically AED 100K+ monthly media spend), we build statistical models estimating each channel's contribution to total revenue using time-series regression. MMM doesn't require user-level tracking and isn't affected by iOS privacy changes or cookie deprecation. It produces a single calibrated view that survives platform reporting inflation.
Server-side conversion tracking
Where user-level tracking is needed for optimisation, we deploy server-side tracking via Meta's Conversion API, Google's Enhanced Conversions, and similar server-to-server integrations. These bypass browser-side ad blockers and iOS privacy restrictions and let us deduplicate conversions across platforms.
First-party data foundations
Everything depends on clean first-party data: properly configured GA4, CRM tying marketing leads to revenue outcomes, consistent customer identifiers across platforms. Most performance marketing problems we encounter trace back to broken first-party data infrastructure.
The 30-60-90 audit framework
Days 1–30: Audit and baseline
- Tracking validation — verify every conversion event fires correctly
- Dashboard reconciliation against back-end
- Spend audit across audiences, placements, creatives
- Branded search isolation — pause and measure
- Documented baseline of real incremental performance
Days 30–60: Restructure
- Cut non-incremental spend (typically 30–60% of branded search)
- Rebuild audience structures based on what's actually converting
- Implement server-side tracking where missing
- Restructure creative testing for proper statistical significance
Days 60–90: Scale validated
- Scale spend on validated incremental channels
- First geo holdout test to validate channel contribution
- Establish reporting cadence and KPI framework
- Quarterly review with documented decision log
What this looks like for different account sizes
Small accounts (under AED 50K/month)
Focus on tracking validation, channel concentration, tight feedback loops between acquisition and product iteration. Geo holdouts difficult at this spend; we rely more on dashboard reconciliation and branded search pause tests.
Mid-market accounts (AED 50K–500K/month)
Geo holdouts become viable. Server-side tracking infrastructure pays for itself. First MMM exercises typically possible by month six. Quarterly recalibration based on actual incremental performance.
Large accounts (AED 500K+/month)
Full MMM with regular recalibration. Multiple parallel holdout tests. Custom incrementality measurement infrastructure. Direct integration with CRM and revenue systems. Decision-grade reporting on real customer acquisition economics.
Frequently asked questions
What's wrong with how most agencies report ROAS?
Most agencies report platform-attributed ROAS — Meta says 6×, Google says 8× — and present the sum as overall performance. Platforms double-count conversions: a purchase that touched Meta, Google search, and email gets claimed by all three. Aggregate dashboard ROAS can show 7× when real incremental contribution is closer to 2×.
How do you actually measure incrementality?
Three primary methods: (1) geo holdout tests — comparable markets where we run full paid stack in one, zero spend in the other, compare total revenue; (2) branded search pause tests — pause brand-name Google Ads for 2 weeks, measure revenue change; (3) dashboard reconciliation — sum platform conversions versus actual back-end orders.
Do you work with Meta, Google, TikTok, all of them?
Yes — we run across Meta, Google Ads, TikTok, LinkedIn (for B2B), and Snapchat where it fits. No religious preference for any platform; we go where the audience converts. But we always start with incrementality testing to understand which channels are genuinely contributing.
What's a realistic timeline for results?
First 30 days: audit, tracking validation, baseline incrementality measurement. Days 30–60: restructuring. Days 60–90: scaling validated approaches. Honest expectation: first 30 days might show worse platform-reported numbers as we cut non-incremental spend; real business impact typically starts moving in month two.
Do you guarantee specific ROAS numbers?
No, and be skeptical of anyone who does. Real ROAS depends on your offer, audience, competition, and a dozen other variables. What we guarantee is methodology rigour: honest measurement, documented decisions, regular reporting, quarterly recalibration.
What stack do you use for analytics?
GA4 as the foundation. Server-side tracking, server-to-server conversion APIs (Meta CAPI, Google Enhanced Conversions). For larger accounts we build media mix models (MMM) using regression on first-party data. We integrate directly with client CRM systems to reconcile platform-reported leads against actual back-end conversions.