Quick answer
Reverse ETL copies data out of a data warehouse and loads it into operational tools such as a CRM, email platform or support desk, so the calculated results are available where teams work.
What is reverse ETL?
Reverse ETL is the process of copying data out of a data warehouse and loading it into the operational tools that business teams use every day, such as a CRM, an email platform, a support desk or an ad platform. Ordinary ETL moves data from many sources into the warehouse. Reverse ETL runs the other way: it takes what the warehouse has already cleaned and calculated, and puts it where someone can act on it.
The typical case: the warehouse can compute "customers who bought twice in 90 days and have not opened an email in 30", but the sales rep works in the CRM and has no access to the warehouse. Reverse ETL writes that flag, per customer, into a CRM field.
ETL, ELT and reverse ETL compared
| ETL / ELT | Reverse ETL | |
|---|---|---|
| Direction | Sources → warehouse | Warehouse → business tools |
| Typical sources / destinations | Databases, apps, event streams → warehouse | Warehouse → CRM, email, support, ads, spreadsheets |
| Main user | Data engineers, analysts | Marketing, sales, support, product teams |
| Goal | Consolidate and clean | Make the result usable where work happens |
| Typical unit | Tables, large batches | Records (one customer, one account) and fields |
A customer data platform often bundles collection, a profile store and this outbound sync. Reverse ETL is the narrower idea: it assumes the warehouse is already the system of record and only handles the outbound leg.
How it works, step by step
- Model. In the warehouse, build a table with one row per person or account and the columns you want to send (for example
lifetime_value,last_order_date,churn_risk). Using SQL, this is usually a view or a scheduled table. - Choose the key. Decide which column identifies a row in the destination: email, CRM ID or account ID. This is the step where identity resolution matters.
- Map fields. Match each column to a destination field, with its type and allowed values.
- Sync. On a schedule or on a trigger, compare the current table with the last synced state and send only the rows that changed (inserts, updates, sometimes deletes).
- Handle failures. Rows that the destination rejects are logged and retried, or set aside for review.
- Monitor. Track rows sent, rows failed, run time and lag.
A worked example: the numbers behind a sync
Sample data: a warehouse table customer_ltv_30d has 20,000 customers. It is synced to a CRM keyed by email. The CRM's batch endpoint accepts 100 records per call and allows 5 calls per second.
| Quantity | Value | Calculation |
|---|---|---|
| Rows in the table | 20,000 | |
| Rows that match a CRM contact | 18,400 | 18,400 ÷ 20,000 = 92% |
| Rows with no match | 1,600 | 20,000 − 18,400 = 8% |
| API calls for a full sync of the matched rows | 184 | 18,400 ÷ 100 |
| Time for that full sync at 5 calls/s | about 37 s | 184 ÷ 5 = 36.8 |
| Rows that changed since the last run | 340 | |
| API calls for an incremental sync | 4 | 340 ÷ 100 = 3.4, rounded up |
| Rows rejected for invalid values (2% of matched) | 368 | 18,400 × 2% |
Three lessons sit in this table. First, an 8% match rate gap is an identity problem, not a sync problem: 1,600 people never get the field. Second, sending only changed rows (4 calls instead of 184) matters once tables are large or rate limits are tight. Third, even a 2% rejection rate leaves 368 customers silently out of date unless failures are surfaced and retried.
How often should it sync?
Pick the frequency from the cost of a stale value, not from what the tool offers.
- If acting on old data does real harm (emailing someone who just unsubscribed, discounting a customer who just bought), sync near real time or on the event that changes the value.
- If a rep reads the field once a day, a daily batch is enough and far cheaper.
- If the value is a slow-moving score (a monthly health score), weekly is usually fine.
Sample data: the 20,000-row table above, synced hourly with incremental changes only, costs 24 runs a day of about 4 calls each (about 96 calls); the same table synced as a full refresh costs 24 × 184 = 4,416 calls a day. Change detection is what makes frequent syncs affordable.
Why teams do it: common use cases
- Sales. Put product usage and a health score into the CRM, so a rep sees that an account stopped using a feature before the renewal call.
- Marketing. Sync a segment such as "bought in the last 30 days" to an email platform or to an ad audience, so the same definition drives every channel.
- Support. Show plan, lifetime value and recent errors on the ticket.
- Finance and operations. Push a calculated metric to a spreadsheet or a planning tool.
The benefit in each case is the same: the definition is written once, in the warehouse, instead of rebuilt in each tool.
What goes wrong
- Freshness. A daily sync means an email can still go to someone who unsubscribed this morning. Decide per field how stale it may be.
- Identity mismatch. Emails in the warehouse and the CRM differ in case, aliases or typos; match rates below 90% are a warning.
- Overwriting. A sync that writes a field the sales team also edits will fight them. Own each field in one place.
- Silent failures. Without a failure log and an alert, a broken sync looks like a quiet week.
- Privacy. Moving personal data to more systems spreads it further, and consent and deletion requests must follow it. A person who asks for deletion must be removed from every destination, not only the warehouse.
- Cost and load. Frequent full syncs of big tables cost warehouse compute and may trip destination rate limits.
Reverse ETL and TapCub
TapCub is not a reverse ETL tool: it does not push segments or fields into a CRM, email platform or ad platform. It is closer to the first leg of the chain, a source of behaviour data. Here is how it fits into the example above.
- Collect. Events and user properties arrive through the web script, SDKs or the server HTTP interface. The public API is for writing data in, not reading it out.
- Define the audience. In All-platform analytics › Users › Segments & tags, build the segment (for example, bought twice in 90 days) so the definition is written down and its size trend is visible for 30 days.
- Move data out. For raw detail, create an asynchronous export in All-platform analytics › Data › Data export. Kinds include events, sessions and persons; the file is JSONL, kept for 48 hours, and rows older than the retention period are already gone. It is a batch job, not a live feed, and the console cannot download the file yet, so delivery needs support. The SQL console in All-platform analytics › Behavior › Metrics & SQL is for querying inside TapCub.
- Keep identities consistent. Identify logged-in people with
tapcub.loginso the same ID can be joined in the warehouse. Anonymous visitors have no profile and do not enter segments. - Close the loop in the warehouse and the destination. Loading to a CRM, and checking match rate and rejected rows, happen in your pipeline, not in TapCub.
See also: What is data activation? From warehouse to ad email and chat tools.
Sources
- Twilio, What is Reverse ETL (and how does it work)?.
- DataCamp, Reverse ETL: a complete guide.
Was this explanation clear?
We use your answers to fix and expand this page.
Couldn’t send. Please try again in a minute.
Thanks for the feedback. We’ll use it to improve this page.You’ve already sent feedback on this page. Thanks.

Data Analyst
Explores traffic sources, conversion funnels, and retention, using clear metric definitions and concrete examples to explain what the numbers mean.