Fivetran

What it is

Fivetran started as managed ELT: you pick a source — Salesforce, HubSpot, Zendesk, NetSuite, a Postgres replica — point it at Snowflake, BigQuery, Databricks or Redshift, and Fivetran builds and maintains the pipeline, including the schema changes that break hand-written scripts. The pricing page now lists 700+ connectors. Over the last eighteen months it stopped being only that. Fivetran acquired Census in May 2025 and relaunched it as Fivetran Activations, a reverse ETL product that pushes warehouse rows back into CRM and CS tools, priced on its own meter from February 1, 2026. It bought Tobiko Data, the company behind SQLMesh and SQLGlot, in September 2025. And on June 1, 2026 it completed an all-stock merger with dbt Labs, with George Fraser as CEO and Tristan Handy as president. The same day, dbt Core v2.0 shipped with the Fusion engine released under the Apache 2.0 license. Coverage of the deal puts the combined company near $600 million in ARR; the merger press release itself cites 100,000+ data teams and names no revenue figure.

So the question the page has to answer: now that one vendor sells ingestion, transformation and activation, do ops teams still need three contracts to get clean data into the warehouse and back into Salesforce?

Why it shows up in ops stacks

• Connector maintenance is the job it takes off your plate. When Salesforce adds a field or HubSpot deprecates an endpoint, a Fivetran connector absorbs the change; a Python script written by a contractor two years ago does not. For a RevOps team that wants a pipeline-and-bookings model in the warehouse, that is the scoped use case: land Salesforce, HubSpot and billing data in one schema, model it in dbt, and never own the extraction code.
• The transformation layer is the de facto standard. dbt is how most warehouse teams write models, and the Standard plan includes dbt Core integration. Post-merger, the ingestion and modeling roadmaps sit under one company, which removes the finger-pointing when a sync lands data a model cannot parse.
• Activations closes the loop without a second vendor. Scored accounts, health scores or legal-matter status modeled in the warehouse can be written back to Salesforce, Gainsight or Zendesk on the same account. The free plan includes 3,500 activation MAR to test it.
• An official MCP server exists. Fivetran publishes a self-hosted server at github.com/fivetran/fivetran-mcp that you clone and run in your own environment. It answers operational questions — which connections are broken, when a sync last succeeded — from Claude or another MCP client. It manages pipelines; it does not query your data.

Pricing reality

The meter is monthly active rows (MAR): rows inserted, updated or deleted in the destination, counted per connection each calendar month. Initial syncs and unchanged rows are free. Three changes since 2025 matter more than the list itself:

• March 2025: MAR is tiered per connection, not summed across the account, so many small connectors no longer share one volume discount.
• January 1, 2026: every standard connection generating between 1 and 1 million MAR carries a $5 monthly base charge, deletes count toward paid MAR, and history-mode changes are billed. Annual contracts pick these up at renewal.
• February 1, 2026: Activations gained its own MAR meter.

The free plan covers 500,000 connection MAR, 3,500 activation MAR and 5,000 monthly model runs. Transformation runs above that cost $0.01 each up to 30,000, $0.007 up to 100,000 and $0.002 beyond. Annual commitments save up to 22%.

Buyer data is the better guide to what you will pay. Vendr reports a median annual contract of $48,137 across 767 purchases, a range of $15,000 to $129,240, and 17.49% average negotiated savings. Read it as: a RevOps team syncing CRM, marketing automation and billing lands between $15,000 and $50,000 a year; a company replicating production databases with CDC is a six-figure line item. The dbt platform is billed separately — Starter is $100 per user per month with 15,000 successful model builds included.

Best for

RevOps, CS ops and data leads at companies with a warehouse and at least one person who writes SQL, who want Salesforce, HubSpot, support and billing data modeled in one place and are tired of maintaining extraction scripts. It is the default pick when a buyer wants one vendor relationship across ingestion, dbt and reverse ETL.

Do not buy it if you have no warehouse and no one to model data — Fivetran moves rows, it does not decide what an active customer is. Do not buy it for a single low-volume source either: the $5 connection floor and per-connection tiering make it expensive relative to the job.

Versus the alternatives

Airbyte is the other volume leader and the open-source one. Airbyte Core is free to self-host; the managed Standard plan starts at $20 a month with 5 credits, and Plus runs from $189 a month for 40 credits. Pro and Enterprise Flex bill on capacity (Data Workers) rather than rows. Pick Airbyte when high-churn tables would make a MAR bill unpredictable, or when data must stay inside your network. Pick Fivetran when you would rather pay more than run and patch connectors yourself.

Hightouch is the alternative for the activation half. Its main rival, Census, now lives inside Fivetran. Pick Hightouch when activation is the harder problem — audiences, identity resolution, many ad and lifecycle destinations — and keep ingestion with whichever vendor you prefer. Pick Fivetran Activations when the use case is a handful of CRM and CS writebacks and one invoice matters more than destination depth.

dlt, the open-source Python library from dltHub, is the fastest-growing entrant: PyPI shows roughly 4.8 million downloads in the last month. Pick dlt when a data engineer already runs Python pipelines and wants ingestion as code in the repo at zero license cost. Pick Fivetran when nobody on the team should be maintaining ingestion code at all.

If none of these fit because the real problem is that sales, finance and CS define “customer” differently, you have a modeling problem. Fix the definitions in dbt first; any of these tools will ship a wrong definition just as reliably as a right one.

Watch-outs

• The 2026 billing changes raise bills without any change in usage. Deletes, history mode and the $5 connection floor all land at renewal for annual contracts. Guard: before renewal, export per-connection MAR from the usage dashboard, recompute last quarter under the 2026 rules, and consolidate or retire connections that generate fewer than 1 million MAR and little business value. Negotiate the overage rate and the annual uplift into the order form.
• High-churn sources burn MAR. A Salesforce object where a workflow rewrites a timestamp on every record nightly bills as if every row changed. Guard: run the free plan against your actual sources for two weeks, find the tables driving MAR, and exclude columns or tables that change without business meaning before you commit to a volume.
• The dbt open-source community is watching the merger. An IDC analyst quoted by TechTarget named keeping community trust as the real test, and another flagged Fivetran’s reputation for aggressive consumption pricing. Guard: keep your models in dbt Core in your own repo, on the Apache 2.0 engine, so switching the ingestion or orchestration vendor is a configuration change and not a rewrite.
• Activations is newer and narrower than the incumbents. Guard: list your writeback destinations and required sync frequency before buying, and confirm each destination and field mapping in a trial. If you need more than a few CRM and CS targets, price Hightouch against it.
• Warehouse compute is not on the Fivetran invoice. Frequent syncs and dbt runs are queries that Snowflake, BigQuery or Databricks bill for. Guard: set sync frequency from business need — hourly is plenty for pipeline reporting — and track warehouse spend attributable to Fivetran and dbt jobs for the first 60 days.