All posts
Performance Marketing
3 min read• 11/22/2026

Beyond Conversions API: Graph-Based Attribution Engines

Client-side pixels and probabilistic fingerprinting are dead. Here is how modern growth engineering builds first-party deterministic event graphs directly in the data warehouse.

Share X LinkedIn

Tip: use ← / → to browse posts.

Beyond Conversions API: Graph-Based Attribution Engines
## The Complete Breakdown of Traditional Attribution Performance marketing attribution has spent five years in a defensive retreat. Apple’s App Tracking Transparency (ATT) stripped IDFA access, Safari’s ITP decimated third-party cookies, and modern browser engines actively truncate first-party cookies set via JavaScript to a 24-hour lifespan. Google Chrome's ongoing deprecations and regulatory mandates (GDPR, CCPA) have effectively closed the book on the old playbook. The industry's reactionary bandage was the "Conversions API" (CAPI) model: sending server-side webhooks directly to Meta, Google, and TikTok. But if your backend relies on IP addresses, User-Agent strings, and probabilistic device fingerprinting to stitch identities together, your conversion signals are deteriorating daily. Consumer IPs shift constantly across cellular towers, VPNs, and iCloud Private Relay. If you are scaling ad spend beyond seven figures annually, relying on the ad platforms' black-box algorithmic attribution is a losing battle. The platforms systematically double-count conversions, claiming credit for the same incrementality. Modern growth engineering requires building your own deterministic Identity Resolution Graph directly inside an analytical columnar data warehouse. ## The Event Graph Architecture Deterministic identity resolution works by treating every touchpoint as a node, and every explicit identifier connection as an edge. Instead of attempting to "guess" that an anonymous visitor on a mobile Safari session is the same user who checked out on a desktop machine, the graph engine waits for explicit deterministic anchors: - An authenticated email address (hashed via SHA-256). - A telephone number formatted in E.164 standard. - A cryptographically signed token embedded in an email click. - A payment token returned by a processor like Stripe. ```sql -- Materialized View in ClickHouse to generate real-time identity edges CREATE TABLE identity_edges ( source_type LowCardinality(String), source_id String, target_type LowCardinality(String), target_id String, observed_at DateTime64(3, 'UTC') ) ENGINE = ReplacingMergeTree(observed_at) ORDER BY (source_type, source_id, target_type, target_id); -- Example ingestion from an authenticated event INSERT INTO identity_edges VALUES ('anonymous_cookie', 'anon_uuid_992184ad', 'hashed_email', 'e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855', now()), ('hashed_email', 'e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855', 'customer_id', 'cust_849204', now()); ``` Once edges are recorded in your warehouse (e.g., ClickHouse or DuckDB for low-latency queries), an entity resolution job computes connected components. This merges fragmented journey timelines into a unified `canonical_person_id`. ## Implementing Server-Side Deterministic Enrichment To power this engine, the tracking layer must be moved entirely off the client browser runtime and into an edge-routing layer (Cloudflare Workers, Fastly Compute, or custom Rust gateways). ### 1. HTTP-Only, Secure, First-Party Cookies When an edge proxy intercepts incoming traffic, it inspects incoming cookies. If an identity cookie is missing, the edge proxy generates a high-entropy UUID and sets it via the `Set-Cookie` HTTP header with flags: `HttpOnly; Secure; SameSite=Lax`. Because the cookie is set directly by the application server at the network level rather than via `document.cookie` in JavaScript, it is completely immune to browser script lifetime truncation algorithms. ### 2. Upstream Link Preserving Ad click parameters (`gclid`, `fbclid`, `ttclid`, `msclkid`) must be captured at the very first edge proxy hop. - Extract the parameters before any redirect chains or application routing logic can strip them. - Persist these parameters in both the transactional session store and write them immediately to your analytical stream alongside the user's `session_id` and the newly minted HTTP-only identity cookie. - Strip these query parameters from the downstream URL before passing the request to the upstream application servers to prevent cache fragmentation on the edge. ## Running Deterministic Multi-Touch Models With a fully connected identity graph, you can run multi-touch attribution that accurately reflects real unit economics. ``` [Touchpoint 1: Meta Ad] | (Anon ID: anon_01) v [Touchpoint 2: Organic Search] | (Anon ID: anon_02) v [Touchpoint 3: Email Newsletter Click] | (Anon ID: anon_02 + Hashed Email) <--- [IDENTITY RESOLUTION RESOLVED] v [Touchpoint 4: Purchase Completed] | (Customer ID: cust_999 + Stripe Token) ``` When the transaction occurs, the engine traverses the graph edges backwards, linking the Meta Ad (`anon_01`), the Organic Search session (`anon_02`), and the Email click directly to the final paying customer. Instead of blindly crediting Meta with a 100% last-click conversion, you can apply custom Shapley value calculations or algorithmic markov-chain weights to understand the true incrementality of top-of-funnel spend. ## The Strategic Takeaway Performance teams wasting capital on off-the-shelf SaaS attribution dashboards are looking at synthetic estimations of reality. By constructing a first-party event graph in your data warehouse, you regain complete ownership over your acquisition truth. You stop paying ad networks for conversions they did not create and unlock the ability to scale your media spend based on mathematical reality.
martech
analytics
attribution
clickhouse
growth
Share X LinkedIn

What clients say

Real reviews from founders and teams we've shipped with.

5.0 · 6 reviews
"Ad marketplace with clean analytics and pixel-perfect design. A joy to work with Hashim."
Sana I.
CEO, AdsOnMarket
"Resend + React Email templates look premium on every client. Deliverability is 99%+."
Camille W.
Growth, Loopwise
"Design, engineering, growth — Hashim's team owned every layer. We finally look enterprise."
Arjun P.
COO, Bangalore Stays
"Fast, compliant, conversion-focused. Hashim brings both craft and commercial thinking."
Giulia R.
Growth Lead, Olymp Trade IT
"Hashim rebuilt our checkout in a weekend and conversions jumped 34% the following week. Unreal."
Ethan R.
Head of Growth, Shopstack
"Copilot, Azure AI and Copilot Studio deployed across the enterprise — enterprise auth, private data, zero drama."
Microsoft AI Stack
Microsoft Copilot · Azure AI · Copilot Studio