Should we build a CDP or just buy Segment / mParticle?

For most marketing-AI products the honest answer is a warehouse-native CDP, not a packaged one. Your customer data already lives in Snowflake or BigQuery; a composable CDP models profiles and segments on top of it, so you own the data, the SQL, and the cost curve instead of paying per-event to a vendor whose schema you can’t inspect. We’ll tell you when a packaged CDP genuinely fits, but it’s rarer than the sales decks suggest.

How do you handle identity resolution across devices and channels?

Deterministic-first: we stitch on shared keys (email, user ID, hashed identifiers) before we ever reach for probabilistic matching. Where probabilistic stitching is warranted, every merged profile carries a match-confidence tier so downstream systems can choose how much to trust it. The identity graph is rebuildable from source events, so a bad merge can be unwound instead of being baked in forever.

What is reverse-ETL and do we need it?

Reverse-ETL syncs modelled data out of your warehouse into the tools your users act in: ad platforms, email, CRM, messaging. If your platform builds audiences or computed traits, you need activation, otherwise the insight dies in a dashboard. We build it on Hightouch or Census when that fits, or as custom sync workers with per-destination mapping, retries, and delivery observability when you need tighter control.

How do you stop bad data from poisoning every downstream segment?

Event contracts and schema enforcement at ingestion, plus freshness, volume, and uniqueness tests in the transformation layer, so a malformed or missing upstream event is caught before it reaches a profile. Lineage lets us trace any segment back to the exact models and events that produced it, so when a number looks wrong, we can find the cause instead of guessing.

Can this feed our downstream AI models and personalization engine?

That’s the point of building it properly. A governed profile store with resolved identities and computed traits is the foundation every downstream AI marketing system depends on: attribution, personalization, scoring, and audience models all read from the same source of truth. We build the platform so those systems consume clean, consented, well-modelled data instead of each re-inventing its own broken pipeline.

Marketing Data Platform Development

We build customer data platforms (CDPs), identity resolution graphs, and audience activation pipelines for marketing-AI products. Warehouse-native, consent-aware, and wired to reverse-ETL every segment back into the channels your users run.

Discuss my project See what we build

What we build

Customer data platform (CDP) builds

A unified customer profile store over your warehouse: first-party events, traits, and transactions modelled into one queryable schema, not a black-box SaaS you can’t inspect.

Identity resolution

Deterministic-first stitching across email, device, cookie, and account, with probabilistic matching only where it earns its place. Every merged profile carries a clear match-confidence tier.

Ingestion & data modelling

Streaming and batch pipelines from web, mobile, CRM, and product events, transformed with dbt into governed, documented, tested marketing models the rest of the stack can trust.

Audience & segment builders

Self-serve segment tooling for your end users: rule and SQL-based audiences, computed traits, and materialised segments that refresh on schedule or in near-real-time.

Activation & reverse-ETL

Warehouse-to-destination syncs that push resolved audiences and traits into ad platforms, email, CRM, and messaging, with per-destination mapping, retries, and delivery observability.

Data quality & governance

Schema enforcement, event contracts, freshness and volume tests, and lineage so a bad upstream event is caught before it silently poisons every downstream segment.

Marketing data that resists wishful thinking

Reference architecture

Event & Source Ingestion
Warehouse (raw)
Identity Resolution
Unified Profile & Segments
Activation / Reverse-ETL
Channels & AI Systems

A typical marketing data platform: first-party events and source systems land raw in the warehouse, identity resolution collapses them into unified profiles, segments and computed traits are modelled on top, and reverse-ETL activates those audiences back into ad, email, CRM, and downstream AI systems. Every profile traces back to the events that built it.

Integrations shipped across 27+ partners.

Event & CDP tooling

Segment
RudderStack
Snowplow
Jitsu
PostHog

Warehouses & lakehouse

Snowflake
BigQuery
Redshift
Databricks
Postgres

Transformation & orchestration

dbt
Airflow
Dagster
Python

Reverse-ETL & activation

Hightouch
Census
Custom sync workers

Destinations

Meta Marketing API
Google Ads API
HubSpot
Salesforce
Braze
Customer.io

Identity & enrichment

Clearbit
Clay
FullContact
Custom match graphs

Which engagement fits

Project Build

For a new customer data platform: a focused build covering ingestion, identity resolution, the profile store, segment tooling, and activation end to end.

Explore Project Build

Embedded Squad

For extending a platform as sources and destinations grow: a dedicated team that ships new connectors, models, and audience features alongside yours.

Explore Embedded Squad

Tech Audit

For a data stack that’s drifting: a 5-day diagnostic of pipeline health, identity-match quality, data governance, and where segments are quietly built on bad data.

Explore Tech Audit

Compliance considerations

Compliance posture for marketing data-platform 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 access and erasure that cascade through the identity graph and every activated destination, and retention enforced at the schema level. See /compliance/gdpr.
CCPA / CPRA	Compliant	"Do not sell / share" honored end to end: a suppressed profile drops out of every segment and reverse-ETL sync, not just the UI toggle.
Consent capture & enforcement	Compliant	Consent state stored on the profile and enforced before activation, so an un-consented user is never pushed to an ad audience or messaging destination.
Data residency & PII minimisation	In progress	We architect region-scoped storage and hash/tokenise PII before it leaves the warehouse. The engineering is ours; the residency and DPA policy call stays with your team.

Free

AI Marketing Stack Audit

A review of your data platform: pipeline reliability, identity-match quality, data governance, and where segments are being built on data you can’t fully trust.

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

For most marketing-AI products the honest answer is a warehouse-native CDP, not a packaged one. Your customer data already lives in Snowflake or BigQuery; a composable CDP models profiles and segments on top of it, so you own the data, the SQL, and the cost curve instead of paying per-event to a vendor whose schema you can’t inspect. We’ll tell you when a packaged CDP genuinely fits, but it’s rarer than the sales decks suggest.
Deterministic-first: we stitch on shared keys (email, user ID, hashed identifiers) before we ever reach for probabilistic matching. Where probabilistic stitching is warranted, every merged profile carries a match-confidence tier so downstream systems can choose how much to trust it. The identity graph is rebuildable from source events, so a bad merge can be unwound instead of being baked in forever.
Reverse-ETL syncs modelled data out of your warehouse into the tools your users act in: ad platforms, email, CRM, messaging. If your platform builds audiences or computed traits, you need activation, otherwise the insight dies in a dashboard. We build it on Hightouch or Census when that fits, or as custom sync workers with per-destination mapping, retries, and delivery observability when you need tighter control.
That’s the point of building it properly. A governed profile store with resolved identities and computed traits is the foundation every downstream AI marketing system depends on: attribution, personalization, scoring, and audience models all read from the same source of truth. We build the platform so those systems consume clean, consented, well-modelled data instead of each re-inventing its own broken pipeline.

Building a marketing data platform? Schedule a meeting.

We’ve shipped the ingestion pipelines, identity graphs, segment tooling, and reverse-ETL activation that marketing-AI teams run in production. Tell us what you’re building.

Discuss my project Back to AI Marketing