Start with the work
An application might create an order, reserve inventory, and read one customer's profile. A report might scan millions of orders to calculate monthly revenue.
Both use data, but their access patterns and operating priorities differ.
The transactional side
Online transaction processing, or OLTP, commonly involves many small reads and writes, concurrent users, and correctness requirements around individual business operations.
A useful design question is: how does the system keep one order correct while other orders are changing?
The analytical side
Online analytical processing, or OLAP, commonly involves scanning, joining, and aggregating substantial portions of a dataset.
A useful design question is: how can the system calculate a large result efficiently without interfering with the application's operational workload?
| Question | Transactional workload | Analytical workload |
|---|---|---|
| Typical scope | A small set of records | Many records and aggregates |
| Example | Update one order | Revenue by month and region |
| Common priority | Predictable transaction latency | Efficient scans and computation |
| Design focus | Constraints and access paths | Data layout and aggregation |
These are common patterns, not absolute boundaries. Some engines support mixed workloads, and small analytical queries may run well in an application database.
Decide when to separate
Separation introduces ingestion, duplicate storage, freshness lag, and another system to operate. It becomes useful when analytical work disrupts transactional performance or when analytics needs a different data organization.
Before moving a report, measure its actual impact and define how fresh the analytical result must be.
Follow the data
A typical flow starts with application transactions, then captures or exports changes, transforms them into analytical models, and serves reports.
Each boundary needs an answer for schema changes, late data, corrections, and reconciliation. Choosing a warehouse does not solve those pipeline questions by itself.