Which attribution model should we ship: last-touch, data-driven, or MMM?

They answer different questions, so we usually ship more than one. Last- and multi-touch models are good for operational, journey-level credit; MMM is better for budget allocation and privacy-safe, aggregate measurement; incrementality tests are the closest thing to ground truth for causation. We build the layer that lets you compare them side by side instead of betting the product on a single method.

How do you handle the death of third-party cookies?

We architect for it directly. The system is built on first-party events and server-side conversion APIs, with a deterministic-first identity graph and probabilistic stitching only where it earns its place. As deterministic signal degrades, MMM and incrementality carry more of the measurement weight, so your attribution doesn’t quietly break when a browser changes its rules.

Ad platforms each claim the same conversions. How do you reconcile that?

Platform-reported conversions double-count by design; every network claims the sale. We normalize spend, impressions, and conversions into one schema, de-duplicate against your own first-party conversion events, and let a single attribution or MMM model arbitrate credit, rather than summing four dashboards that each already took full credit.

Can you build marketing mix modeling (MMM), not just click-path attribution?

Yes. We build regression and Bayesian MMM, including adstock and saturation curves and budget-response simulation, commonly on PyMC-based stacks or open frameworks like Meta Robyn and Google Meridian. We report confidence intervals alongside every estimate, because an MMM that hides its uncertainty is worse than no MMM at all.

How do you prove a channel actually caused conversions, not just correlated?

Attribution alone can’t prove causation. It allocates credit under a set of assumptions. For causal claims we build incrementality testing: geo experiments, holdout and PSA control groups, and conversion-lift study pipelines. We’ll be honest about what your data can and can’t support before we quote the build.

AI Marketing Attribution Development

We build multi-touch attribution models, incrementality and lift testing, and marketing mix modeling (MMM) engines for marketing-AI products. Identity-resolved, ad-platform-normalized, and honest about what the data can actually prove.

Discuss my project See what we build

What we build

Multi-touch attribution models

Rule-based and algorithmic (data-driven, Markov, Shapley) models over a unified event graph, with every assumption visible rather than buried inside a black box.

Incrementality & lift testing

Geo experiments, holdout and PSA control groups, and conversion-lift study pipelines so a channel earns credit for causation, not just correlation.

Marketing mix modeling (MMM)

Regression and Bayesian MMM with adstock and saturation curves, budget-response simulation, and confidence intervals reported alongside every point estimate.

Ad-platform data normalization

Spend, impression, and conversion feeds from Meta, Google, TikTok, and LinkedIn reconciled into one schema, de-duplicated across each platform’s own claimed conversions.

Identity resolution & event stitching

Cross-device and cross-session journey stitching over first-party events, with a deterministic-plus-probabilistic identity graph and clear match-confidence tiers.

ROI dashboards & reporting

ROAS, CAC, and payback dashboards for your end users, with cohort views, model-comparison toggles, and warehouse-native queries behind them.

Marketing data that resists wishful thinking

Reference architecture

Event & Ad-Spend Ingestion
Identity Resolution
Unified Journey Graph
Attribution / MMM / Lift Engine
Model Store
ROI Dashboards & API

A typical attribution system: first-party events and ad-platform spend land in the warehouse, identity resolution stitches touchpoints into journeys, and the modeling layer runs multi-touch, MMM, and lift methods over the same graph. Every dashboard number traces back to a specific model and its assumptions.

Integrations shipped across 26+ partners.

Ad platforms

Meta Marketing API
Google Ads API
LinkedIn Ads API
TikTok Ads API
X Ads API

Analytics & event pipelines

Google Analytics 4
Segment
Mixpanel
Amplitude
RudderStack

Warehouses

Snowflake
BigQuery
Redshift
Databricks
Postgres

Transformation & modeling

dbt
Python / PyMC
Meta Robyn
Google Meridian

CRM & conversion sources

HubSpot
Salesforce
Stripe
Shopify

BI & visualization

Looker
Metabase
Custom React dashboards

Which engagement fits

Project Build

For a new attribution product: a focused 8–14 week build covering ingestion, the identity graph, the modeling layer, and the end-user dashboards.

Explore Project Build

Embedded Squad

For evolving the models as your data grows: a dedicated team that ships new attribution methods, MMM refreshes, and dashboards alongside yours.

Explore Embedded Squad

Tech Audit

For pressure-testing an attribution model you already have: a 5-day diagnostic of data quality, modeling assumptions, and where the numbers overstate credit.

Explore Tech Audit

Compliance considerations

Compliance posture for AI attribution builds. Status reflects how we engineer the controls into your platform. Data-processing ownership and privacy policy stay with your team.
Standard	Status	What we ship
GDPR / UK GDPR	Compliant	Lawful-basis handling, data subject rights, and retention controls across the event pipeline and identity graph. See /compliance/gdpr for how we engineer it.
CCPA / CPRA	Compliant	"Do not sell / share" signals and consumer opt-out honored end to end. A suppressed user drops out of the identity graph and every downstream model.
Consent Mode & consent enforcement	Compliant	Consent state captured at the event layer and enforced before modeling, so un-consented touchpoints never earn attribution credit.
Cookieless / privacy-safe measurement	In progress	We architect toward first-party, aggregated, and modeled measurement (MMM, conversion APIs, incrementality) as third-party signal degrades. The engineering is ours; the privacy policy call is yours.

Free

AI Marketing Stack Audit

A review of your attribution stack: data quality, modeling assumptions, identity resolution, and where the numbers are over- or under-crediting a channel.

Get the audit

Data analyst working at a workstation with multiple data views

Attribution, personalisation, and scale in production

Marketing AI products live or die on data quality, latency, and whether the numbers hold up under real traffic. We build for the dashboard that a CMO opens at 8am on Monday.

Frequently asked questions

They answer different questions, so we usually ship more than one. Last- and multi-touch models are good for operational, journey-level credit; MMM is better for budget allocation and privacy-safe, aggregate measurement; incrementality tests are the closest thing to ground truth for causation. We build the layer that lets you compare them side by side instead of betting the product on a single method.

Building an AI attribution product? Schedule a meeting.

We’ve shipped the identity resolution, multi-touch and MMM modeling, and ROI dashboards that marketing-AI teams run in production. Tell us what you’re building.

Discuss my project Back to AI Marketing