Data Integration Updated: 2026-01-21

What is Fivetran?

A fully managed data integration platform that automatically syncs data from hundreds of sources to data warehouses and lakes with minimal configuration.

From the field

Fivetran is the tool you pick when you want ingestion to stop being your problem. Connect a source, pick tables, and the data lands in Snowflake/BigQuery/Databricks with schema changes handled automatically, incremental syncs, and normalized tables — usually within an hour of setup. The appeal is not the 300+ connectors; it's that someone else is on call at 3 AM when the Salesforce API changes.

Where it shines, where it bites

Fivetran shines for standard SaaS sources (Salesforce, HubSpot, Stripe, Google Ads) where building connectors yourself is undifferentiated work. It bites when your source is high-churn — a table where every row updates daily — because their Monthly Active Rows (MAR) pricing treats every updated row as one MAR, and bills explode. I've seen a single chatty Postgres table (audit log with update-heavy rows) push a monthly bill from $2K to $18K in one quarter. The fix is usually to either exclude that table, switch to log-based CDC with volume-priced billing, or move it to Airbyte self-hosted.

A quick field story

One fintech moved 43 sources to Fivetran in three weeks — a project that would have taken a year to build in-house. Two months in, their bill was 6x the initial estimate because three MongoDB collections were fully re-syncing nightly (a bug on our side in configuring history mode). The Fivetran billing dashboard showed it, we fixed the sync mode, and the bill dropped 70%. Lesson: Fivetran is worth it for the 40 sources you don't think about, but you need a MAR budget alert on every connector, and someone should review the top-spending connectors monthly. The ease is real, and so are the invoices.

Fivetran is a fully managed ELT (Extract, Load) platform that automates data replication from SaaS applications, databases, and other sources to data warehouses and lakes. It handles the E and L, leaving the T to tools like dbt.

How Fivetran Works

  1. Connect: Select a source (e.g., Salesforce, PostgreSQL)
  2. Configure: Provide credentials and select tables
  3. Sync: Fivetran automatically extracts and loads data
  4. Transform: Use dbt or SQL for transformations

Key Features

  • Automated Schema Migration: Handles source changes automatically
  • Incremental Updates: Syncs only new/changed data
  • 300+ Connectors: Pre-built integrations to popular sources
  • Normalized Schemas: Standardized table structures
  • Transformations: Built-in dbt Core for basic transforms

Popular Connectors

CategorySources
CRMSalesforce, HubSpot
MarketingGoogle Ads, Facebook Ads
DatabasesPostgreSQL, MySQL, MongoDB
SaaSStripe, Zendesk, Jira
FilesGoogle Sheets, S3

Fivetran vs Alternatives

  • vs Airbyte: Fivetran is managed, Airbyte is open-source
  • vs Stitch: Similar but Fivetran has more enterprise features
  • vs Custom Code: Fivetran eliminates maintenance overhead

Pricing Model

Fivetran uses Monthly Active Rows (MAR) pricing:

  • Pay based on rows updated each month
  • Predictable costs for stable datasets
  • Can be expensive for high-change data

Key Points

Gotchas

Fivetran pitfalls I see repeatedly:

  • MAR pricing and high-churn tables are a trap. A table where every row updates daily costs as much as a fully-new table every day. Review the MAR breakdown monthly and set per-connector alerts. Pull rarely-used tables out of the sync.
  • History mode / sync mode defaults matter. Some sources default to "full table sync" rather than incremental, which is expensive and slow. Audit sync modes per connector. For MongoDB specifically, history mode can 10x your MAR.
  • Schema drift breaks dbt models. Fivetran adds new columns when sources add them; your dbt models can silently pick up extra columns or fail when types change. Use dbt tests on source freshness and column presence.
  • Deletes are connector-dependent. Not every Fivetran connector captures hard deletes — some mark as deleted, some don't capture at all. Read the per-connector docs before assuming your warehouse reflects source reality.
  • API rate limits push latency. Fivetran honors source API rate limits. Salesforce, HubSpot, Google Ads all have quotas. Sync frequency higher than every 15 minutes often pushes you into rate-limit backoff, with no benefit.
  • Normalization decisions are sticky. Once Fivetran normalizes a nested JSON structure into relational tables, downstream dbt depends on that shape. Re-configuring denormalization later causes a re-sync (expensive) and can break models.
  • Custom connectors are limited. Fivetran's connector SDK exists but is constrained. For genuinely custom sources, Airbyte's Connector Builder or a custom Python + Dagster pipeline usually wins.
  • No mid-stream transformation. Fivetran is EL, not ETL. Transformations happen in the warehouse via dbt after loading — not during transit. If you need to mask PII before it lands in the warehouse, Fivetran isn't the right tool (look at Airbyte with hashing, or custom ingestion).

Frequently Asked Questions

What is Fivetran used for?

Fivetran is used to automate the ingestion layer of a data platform — it syncs data from SaaS apps (Salesforce, HubSpot, Stripe), databases (Postgres, MySQL, MongoDB), and files (S3, Google Sheets) into your data warehouse (Snowflake, BigQuery, Databricks, Redshift) without you writing or maintaining connector code. Schema changes are handled automatically, sync is incremental, and normalization is pre-built. Teams use it to stand up ingestion for 20-100 sources in weeks instead of quarters, and to avoid the perpetual maintenance burden of custom connectors.

Is Fivetran an ETL tool?

Fivetran is an EL (Extract, Load) tool, not traditional ETL. It intentionally leaves the T (transform) to tools like dbt running inside the warehouse. This aligns with the modern ELT pattern — raw data lands first, transformations happen in Snowflake/BigQuery/Databricks where compute is cheap and version control (via dbt) is easy. Fivetran does include some basic transformation via dbt Core integration, but the core product is connectors-to-warehouse without mid-stream transformation. If you need mid-transit transformations (e.g., PII masking before load), Fivetran is not the right fit.

How does Fivetran pricing work?

Fivetran uses Monthly Active Rows (MAR) pricing — you pay per unique row that is inserted or updated during a month. A row updated 10 times still counts as 1 MAR. This is predictable for stable datasets (most of your SaaS sources) but becomes expensive for high-churn data (audit logs, IoT event tables, fast-changing Postgres tables). Tiered pricing drops the per-MAR rate at higher volumes. Budget carefully for the first 60 days — the first bill is usually a surprise. Turn on per-connector alerts and review MAR breakdown monthly.

Fivetran vs Airbyte: which is better?

Fivetran for reliability, polish, and enterprise support — managed service, best-in-class connector quality, someone else on call when a source API breaks. Airbyte for cost control and flexibility — self-host for free, build custom connectors visually, keep data in your network. In practice, many teams use Fivetran for critical SaaS sources (Salesforce, HubSpot, Stripe) where reliability matters and Airbyte self-hosted for long-tail or custom sources. Don't treat it as all-or-nothing. Both support 300+ connectors; the difference is SLA and operating model.

Does Fivetran support Change Data Capture (CDC)?

Yes — Fivetran supports log-based CDC for databases like PostgreSQL, MySQL, Oracle, and SQL Server, reading the transaction log (WAL, binlog) directly rather than polling. This is the efficient path for capturing inserts, updates, and deletes in near-real-time without load on the source. For SaaS sources, Fivetran uses the appropriate API pattern (webhooks, incremental polling, snapshot + incremental) depending on what the source supports. CDC mode is more efficient than incremental polling but requires source-side configuration (replication slot on Postgres, binlog format on MySQL).

How do you reduce Fivetran costs?

Five tactics: (1) exclude high-churn tables that aren't actually needed — audit tables, log tables, frequently-updated staging tables; (2) reduce sync frequency on sources where 1-hour or daily is acceptable; (3) switch high-volume sources to a cheaper alternative (Airbyte self-hosted, custom CDC to Kafka, Snowpipe Streaming); (4) turn off history mode on MongoDB / DynamoDB unless you actually need full change history; (5) set per-connector MAR budgets and review monthly. The MAR breakdown in Fivetran's UI makes it easy to spot the top 3 connectors eating your budget — they almost always follow the 80/20 rule.

Is Fivetran secure enough for regulated industries?

Fivetran is used by healthcare, financial services, and government-adjacent orgs, and carries SOC 2 Type II, HIPAA, GDPR, and PCI certifications. It supports private networking (AWS PrivateLink, Azure Private Link, GCP Private Service Connect), customer-managed keys, and column-level hashing for PII. Gaps: for stringent data-residency requirements (data must never leave your network), you'll want the Hybrid Deployment option (Fivetran runs inside your VPC). Review their security page and your compliance requirements carefully — for most teams it's fine; for some specific regulatory requirements self-hosted Airbyte is easier to approve.

Related Terms

ETL (Extract, Transform, Load)Airbytedbt (Data Build Tool)Data WarehouseChange Data Capture (CDC)

Related Tools

AirbyteStitchdbtSnowflakeBigQuery

Learn More

Browse All Terms Read Articles