Self-Serve Audience Segmentation at Production Scale
The Challenge
A global data intelligence company had 300M+ consumer records with no self-serve access layer. Every audience segment was hand-built by data science. Count queries took minutes, with no enforced PII boundary between analytical and serving workloads.
- No self-serve — every segment required manual data science effort
- Live count queries took minutes, not milliseconds
- No PII boundary between analytical and serving workloads
Stack: ClickHouse · Databricks · React / Node · Cognito · CloudFormation · S3
What We Built
Split-Plane Architecture
ClickHouse control plane for sub-second COUNT queries on non-PII attributes. Databricks data plane for all PII — never exposed in the serving layer. Two planes, one enforced boundary at infrastructure level.
Privacy by Architecture
ClickHouse holds only filterable, non-identifying attributes and boolean reachability flags. Names, emails, addresses are materialised only in Databricks at export time — presigned URL delivery, auto-expired.
One-Click Deployment
Entire control plane — reverse proxy, application server, analytics store, auth, object storage — in a single infrastructure-as-code template. Deploys in under 5 minutes. No data science bottleneck for clients.
Results
<300ms
COUNT queries on 300M+ records
Zero PII
exposure in the serving layer
Self-serve
audience creation — no DS team needed
Let's Talk About Your AI Roadmap
Whether you're exploring AI opportunities or ready to scale, we'll help you build a pragmatic path from ambition to enterprise-grade delivery.
✓ No sales pressure ✓ Actionable insights ✓ Response within 24 hours