Reverse ETL
Reverse ETL copies modeled data out of the data warehouse and into the operational tools people work in every day: the CRM, the marketing platform, the ad network, the support desk. Regular ETL and ELT bring data into the warehouse for analysis; reverse ETL sends the results of that analysis back out, so a salesperson sees a customer’s lifetime value in Salesforce instead of in a dashboard they never open.
The pattern is also called data activation or operational analytics, and it is a core feature of composable customer data platforms (CDPs).
How a sync works
- A customer_360 model in the warehouse and the CRM's contact list. Right now they agree.
- The nightly dbt run updates the model: Alice crossed the VIP threshold and Maya is a new customer.
- The sync engine compares the model with what it sent last time. Only 2 of 4 rows changed.
- It upserts just those rows through the CRM's API, matched on a stable key (email or CRM id), within the API's rate limits.
- The insight now lives where the work happens. No one has to open a dashboard.
Why not just call the API from a script?
You can, and for one field it works. A reverse ETL tool, or a well-built script, has to solve the same problems:
- Diffing: send only rows that changed, or you will burn API quotas and overwrite edits made in the CRM.
- Identity mapping: match warehouse rows to CRM records (email, external ID, CRM object ID) and decide what to do with duplicates.
- Rate limits and batching: Salesforce, HubSpot and ad platforms all cap calls per second and per day.
- Partial failures: one invalid email must not fail the other 9,999 records. Failed rows need a log and a retry.
- Field mapping and types: picklists, required fields, date formats and custom objects differ per destination.
Typical use cases
| Team | Data synced | Destination |
|---|---|---|
| Sales | Lifetime value, product usage, lead score | Salesforce, HubSpot |
| Marketing | Audiences such as “churn risk” or “bought twice” | Braze, Iterable, Google and Meta Ads |
| Support | Plan, open invoices, VIP flag | Zendesk, Intercom |
| Product / finance | Usage-based billing quantities | Stripe, internal services |
Common mistakes
- No owner for synced fields. If both the CRM user and the sync can edit
segment, they will overwrite each other. Make synced fields read-only in the destination. - Syncing raw tables. Sync from tested, documented models with a primary key, not from raw data.
- Full syncs every run. They are slow, expensive and can trigger thousands of CRM workflows at once.
- Treating it as real time. Reverse ETL is usually batch (minutes to hours). If a feature must react in seconds, it needs an event stream, not a warehouse sync.
Tools in 2026
Hightouch, Census (acquired by Fivetran in 2025), RudderStack and Segment Reverse ETL are the common managed options. The DIY route is a scheduled job (Airflow, Dagster) that diffs a model against the last snapshot and calls the destination’s bulk API.
Learning Resources
- Fundamentals of Data Engineering, Chapter 9 “Serving Data for Analytics, Machine Learning, and Reverse ETL”
- Hightouch: What is Reverse ETL?
- Segment: What is reverse ETL?
- What is Reverse ETL? (Explained by the Creator)