Attribution in 2026 is messy. iOS App Tracking Transparency, Safari ITP, ad blockers, DPDP consent gates, and increasingly aggressive browser privacy defaults have all eroded deterministic tracking. The operators winning are running a three-layer stack — not relying on any single attribution model.
This is the Frameleads operator reference. Anchored to the Attribution & Measurement pillar.
The 3-layer attribution stack
Layer 1 — Deterministic (CAPI + server-side)
Server-side event firing via GTM Server-Side container routing to Meta CAPI + Google Ads Enhanced + LinkedIn CAPI + GA4 Measurement Protocol. Recovers signal from iOS / Safari / ad-blocker users (where their browser allowed any tracking at all). Match-rate typically 70-85% for warm audiences, 40-60% for cold audiences.
Layer 2 — Probabilistic (modelled attribution)
Platform-side modelled attribution fills the remaining gap with statistical estimation. Meta's Advantage+ Attribution, Google's data-driven attribution, GA4's modelled conversions. These work — but tend to over-attribute in ways that flatter the platform's own performance. Use as one input among several, not as truth.
Layer 3 — Survey (post-purchase 'how did you hear about us')
Single-question prompt at checkout: 'How did you hear about us?' with a fixed list (Google, Meta, friend referral, podcast, article, other). Typical response rate 30-60% for Indian D2C. Survey data is unbiased by platform modelling — gives the cleanest cohort-level signal available for self-reported attribution. Reconcile against deterministic + probabilistic data monthly.
What each layer is good for
- Deterministic — best for daily optimisation decisions. Real-time enough to feed ad-platform optimisation algorithms.
- Probabilistic — best for ad-platform self-optimisation (let Meta + Google ML use their own modelled data; reporting it to you separately).
- Survey — best for monthly + quarterly strategic decisions (channel-mix re-allocation, budget shifts between platforms). Highest signal quality, lowest velocity.
- Reconciliation — monthly comparison of all three layers + Shopify/CRM truth surfaces measurement gaps + attribution-model bias.
When to add MMM
Marketing Mix Modelling (MMM) — top-down statistical analysis of media spend vs revenue — becomes accessible at ₹2-3Cr+ annual media spend. Open-source tools (Robyn from Meta, LightweightMMM from Google) made MMM possible without commercial vendor relationships. Below ₹3Cr media, deterministic + survey is sufficient. Above ₹3Cr multi-channel, MMM adds incremental insight on cross-channel halo effects + diminishing returns curves.
DPDP compliance
India's DPDP Act 2023 became enforceable in 2025. Every attribution layer must respect DPDP requirements: explicit consent capture, named purpose for data collection, audit trails, right-to-deletion. Frameleads' standard consent layer uses Google Consent Mode v2 + custom DPDP audit logging — sits at the GTM Server-Side container, blocks downstream events when consent is denied.
The monthly reconciliation report
Frameleads' standard monthly attribution report compares: (1) Platform-reported revenue (Meta + Google + LinkedIn) vs (2) GA4 modelled attribution vs (3) Server-side actual events vs (4) Survey self-reported attribution vs (5) Shopify/CRM truth. Spread analysis surfaces measurement gaps + attribution model bias. Recommendations: budget re-allocation, channel-mix decisions, consent layer improvements.
Read the Attribution & Measurement pillar for the deeper operator framework. Or book a free audit — we'll score your current stack on the call.