Use Cases

Workload patterns that drive real-time analytics

Fit is determined by workload shape, not by industry or company size. Each pattern below records the data shape, typical scale, the metrics that matter and the objections it tends to attract.

Fraud detection & investigation
Financial services · Payments · Marketplaces · Gaming
strong fit

Problem

Fraud signals arrive as high-volume events, but rules and analysts operate on stale aggregates, so losses are found after settlement.

Why real-time matters

The decision window is milliseconds to minutes. Feature lookups and analyst pivots must run against fresh data, not yesterday's batch.

Data shape

Append-only transaction and behavioural events, wide attribute sets, very high cardinality identifiers.

Typical scale

10 GB–10 TB/day ingest, 90 days–3 years retention, hundreds of concurrent analyst queries.

Architecture patterns
  • Kafka → ClickHouse raw events with materialized views for per-entity aggregates
  • Feature serving for real-time scoring via low-latency point aggregations
  • Analyst investigation workbench with arbitrary pivots over raw history
Metrics that matter
  • Fraud loss rate
  • False positive rate
  • Time to detect
  • Analyst investigation time
  • Rule iteration cycle time
Business value
  • Reduced fraud losses through faster detection
  • Fewer false declines, protecting revenue and customer trust
  • Faster rule iteration because analysts can test hypotheses on full history
Discovery questions
  • How long between a fraudulent event and your team seeing it?
  • How much history can an analyst query interactively today?
  • What is the current monthly fraud loss, and what portion do you consider addressable?
  • How long does it take to ship a new rule from idea to production?
Reference architectures

Kafka / Confluent → ClickHouse

Consume existing event streams directly into ClickHouse, with materialized views building query-ready aggregates at insert time.

Producers → Kafka / Confluent / Kinesis → ClickHouse ingest (Kafka engine or ClickPipes) → Materialized views → rollups → Real-time apps & dashboards

Postgres → CDC → ClickHouse

Keep the operational database as the system of record and stream row changes into ClickHouse for analytics and customer-facing reporting.

PostgreSQL (OLTP) → Debezium / ClickPipes CDC → ClickHouse (ReplacingMergeTree) → Dashboards & product analytics

Relevant customer stories

SEON

Fraud prevention (vendor)

A fraud vendor building its product on ClickHouse for real-time signals and tenant analytics.

Instacart

Grocery marketplace

Marketplace fraud and operational analytics moved from batch to near-real-time.

Mercado Libre / Mercado Pago

E-commerce & payments

One of LATAM's largest platforms running real-time analytics on ClickHouse.

Qonto

Business banking

Regulated European banking analytics running on ClickHouse.