Skip to content
English
Contact us
  • There are no suggestions because the search field is empty.

The Model Waterfall (Full Walk-Through)

Stage: How It Works

Reading time: ~20 minutes

The Model Waterfall

Fospha builds full-funnel measurement in five verified steps. Each step addresses a blind spot that the previous level of measurement cannot see.

Data Pipeline

Fospha ingests daily from: GA4, Ad platforms (Meta, Google, TikTok, Snapchat, Pinterest), Ecommerce platforms (Shopify etc.), Amazon Seller Central + Ads.

Step 00: GA4 Conversions (Starting Point)

The measurement pipeline begins with raw conversion data from GA4 — the last-click view of the customer journey, showing final touchpoints before purchase. GA4's default model (LNDC) excludes direct visits and credits the last marketing touchpoint before conversion.

The problem: last-click massively overcredits demand capture channels (Google Search, retargeting) and gives zero credit to brand-building activity (TikTok, YouTube, Meta awareness) that actually created the demand.

This is where last-click measurement stops. For Fospha, it is just the beginning.

Step 01: Click Measurement

Fospha measures sales driven by paid and organic clicks using a multi-model approach that combines two categories of methodology:

Touchpoint models (using Google's own touchpoint data):

  • LNDC (Last Non-Direct Click) — GA4's default. Credits 100% of the conversion to the last ad or link clicked before purchase.
  • DDA (Data-Driven Attribution) — Distributes credit across the last four touchpoints in the customer journey, weighted by their measured influence.

Probabilistic models (using Fospha's ML to find hidden patterns):

  • Linear Regression — Identifies linear relationships between marketing activity (clicks, sessions, cost) and conversions.
  • Tree-Based Models — Captures more complex, non-linear patterns. Builds many smaller models sequentially, each one learning from the errors of the previous, to produce a stronger overall prediction.

By merging outputs from all four models, the multi-model approach uncovers hidden relationships between clicks, cost, and sales — providing a fairer, more balanced view of how each channel contributes to conversions. No single model's blind spots dominate the output.

Step 02: Capture Every Sale (Reconciliation)

GA4 undercounts conversions by 10-30% or more due to cookie consent limitations and tracking gaps. Fospha reconciles against your ecommerce platform as the source of truth to ensure 100% accurate sales data.

The reconciliation process has three stages:

  1. Isolate unmatched transactions — Fospha matches Order IDs from your ecommerce platform (Shopify, Magento, WooCommerce) with GA4 transactions. Conversions that GA4 missed entirely are isolated in a reconciliation table before ML processing.
  2. Calculate channel mix coefficients — After modelling visits, Fospha determines what percentage of GA4 conversions each channel group represents. For example: if 30 of 100 conversions came from email, email's coefficient = 30% (0.3).
  3. Redistribute proportionally — Unmatched transactions are redistributed across channels based on each channel's coefficient. If there are 10 unmatched conversions and email's coefficient is 0.3, email receives 3 of those transactions.

By default, Fospha reconciles to "Online Store" data for Shopify, ensuring total conversions in Fospha match the ecommerce platform's actual completed orders.

Step 03: Daily MMM (Impression Measurement)

Fospha's key differentiator. This is where brand-building channels finally get fair recognition.

Using an XGBoost model and Shapley values (derived from game theory), Fospha uncovers the true impact of impressions — so upper-funnel channels like TikTok, YouTube, and Meta awareness get the recognition they deserve.

How it works at a technical level:

  • Data is modelled with the Daily MMM down to Campaign Type/Objective level first
  • Granular platform signals are then applied to attribute results down to individual ads
  • Statistical significance testing validates results before granular attribution is applied

This captures the impact of channels that drive demand but rarely receive credit when sales later convert through Direct, Organic, or Brand PPC.

Step 04: Halo (Unified Measurement)

Extends measurement beyond DTC to create integrated analysis across multiple sales channels.

Amazon: Pulls Amazon sales and Amazon Ads data via Seller Central and Ads APIs. ML models (XGBoost and ElasticNet) evaluate Amazon ad attribution, separating ads-driven from organic conversions. Unclaimed organic conversions are run through the Daily MMM to identify how DTC ad impressions influenced Amazon purchases.

TikTok Shop: Ingests TikTok Shop order data via Shopify or custom file upload. Integrates TikTok Shop ad impressions into the Daily MMM to reveal TikTok's full impact across DTC, Amazon, and TikTok Shop.

Output: Unified ROAS, CPP, Revenue, and Conversions across .com, Amazon, and TikTok Shop — accessible down to individual ad level.

The Output

A fully reconciled, cross-channel, daily, ad-level view trusted by marketing, finance, and leadership. The full-funnel measurement reveals the true value of every channel — from the first impression to the last click.

Key Analogies

  • Multi-model approach: "Four people independently checking each other's maths" — touchpoint models (LNDC, DDA) and probabilistic models (Linear Regression, Tree-Based) averaged together
  • Reconciliation: "GA4 is a bucket with holes. Fospha fills it back up" — matching ecommerce Order IDs, calculating channel mix coefficients, redistributing unmatched transactions
  • Impression measurement: "See a TikTok ad on Monday, Google the brand on Friday and buy. Last-click credits Google. Impression measurement credits TikTok — the channel that actually planted the seed."