$ hightouch sync run customer_360_to_crm

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

$ hightouch sync run customer_360_to_crm
Warehouse model to CRM, only what changed
WAREHOUSE · customer_360 emailltvsegment alice@900regular sam@300regular ryan@120regular CRM · contacts emailsegment alice@regular sam@regular ryan@regular alice@1300VIP maya@2100VIP refreshed by dbt at 06:00 SYNC ENGINE diff vs last run 2 changed · 2 same upsert by email 2 records sent, not 4 alice@VIP maya@VIP Sales sees "VIP" on Alice's contact; a CRM workflow assigns an account manager.
  1. A customer_360 model in the warehouse and the CRM's contact list. Right now they agree.
  2. The nightly dbt run updates the model: Alice crossed the VIP threshold and Maya is a new customer.
  3. The sync engine compares the model with what it sent last time. Only 2 of 4 rows changed.
  4. 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.
  5. 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

See also: ELT, CDC