ETL tools copy data out of the apps and databases your business runs on. They clean and reshape it on the way, then load it into a data warehouse where people can report on it. There are hundreds of them. However, they differ less in what they do than in who runs them and how they charge. This guide therefore sorts 22 widely used ETL tools into four types and explains how each one charges. It ends with the tests worth running before you sign.
A note on bias: Peliqan publishes this guide, and Peliqan sells a data platform that includes ETL. So we kept our own product out of the list. Instead, a separate section near the end explains where it fits and where it does not.
The short answer
There is no single best ETL tool. In practice, the right one depends on three things: who runs it, where the data lands and how you pay. Most shortlists therefore come down to one of these situations.
- A data warehouse, a budget and no time for maintenance: start with Fivetran, then compare Hevo Data and Airbyte Cloud.
- Engineers who want full control: self-hosted Airbyte, dlt or Meltano, scheduled by Airflow.
- Everything on one cloud or data platform: try its own service first, such as AWS Glue, Databricks Lakeflow or Snowflake Openflow.
- Database changes needed within seconds: a change data capture specialist such as Estuary.
- A large, regulated company with many systems: Informatica, Qlik Talend Cloud or IBM DataStage.
- A small team without data engineers: Hevo Data, Skyvia or Integrate.io, or an all-in-one platform.
What ETL tools do, and how ETL differs from ELT
An ETL tool automates the three steps of ETL. First, it extracts data from sources such as a CRM, an ERP or a database. Next, it transforms the data into a consistent shape, and finally it loads it into a destination. It then repeats that on a schedule, so only new and changed records move after the first load. In addition, a good tool handles the dull parts: logins, API limits, retries and alerts when a run fails.
Most tools sold as ETL tools today actually do ELT. In other words, they load raw data into the data warehouse first and transform it there, on its own compute. That is cheaper to build and easier to change, because the raw data is always there to rebuild from. Our data warehouse examples show what that destination looks like in practice.
Still, the classic order has its place. For example, you may need to mask sensitive fields before they reach the data warehouse at all. Our ETL vs ELT comparison explains when that matters.
Several tools that appear in ETL lists are neither, though. For instance, orchestrators such as Airflow schedule code you write. Transformation tools such as dbt model data after it lands. Meanwhile, reverse ETL tools push data from the data warehouse back into apps.
Still, they all belong to the wider family of data integration tools, and most teams use two or three together.
ETL tools compared
Below, the table sums up all 22 tools. After that, each tool gets a short review, grouped by type.
Managed ELT services
Managed services run the connectors for you. You choose a source, sign in and pick a destination. The vendor then deals with API changes, retries and schema drift. In exchange, you get a usage-based bill and less control over how each connector behaves. They suit teams that would rather pay than maintain code, which is also the idea behind most low-code integration platforms.
1. Fivetran
Fivetran is the reference point for managed ELT. It lists 750+ connectors that handle incremental syncs and schema changes automatically. In addition, it runs as SaaS or in a hybrid model that keeps data in your network. In June 2026, it also completed its merger with dbt Labs, so ingestion and transformation now come from one company. Our explainer on the dbt and Fivetran merger covers what changed.
The catch is the bill. Fivetran charges by monthly active rows, which counts each row that changes in a month. So a source that touches many records costs more, whether or not those rows matter to you. If cost is the problem, our list of Fivetran alternatives compares the usual replacements.
Best for: funded data teams that want zero-maintenance ingestion into an existing data warehouse.
2. Hevo Data
Hevo Data is a no-code managed ELT service with 170+ connectors. It charges by events, where an event is each record inserted, updated or deleted. Its free plan covers up to 1 million events a month on a limited set of connectors. Meanwhile, dbt-based modelling inside Hevo is still in beta.
Compared with Fivetran, Hevo is simpler to start with, but its catalog is smaller. So check that it covers your long-tail sources, or see our Hevo Data alternatives if it does not.
Best for: lean data teams that want managed pipelines with a free starting tier.
3. Stitch
Stitch is a simple extract-and-load service with 130+ sources. Pricing is based on row usage, with a 14-day trial and no free plan. In 2023, it became part of Qlik when Qlik acquired Talend. Since then, its own site says Qlik is building the best of its technology into Qlik Talend Cloud.
For that reason, Stitch makes most sense where a pipeline already runs reliably on it. For a new build, ask Qlik about the roadmap first, because the direction of travel is clearly Qlik Talend Cloud.
Best for: small teams with simple replication into one data warehouse.
4. Estuary
Estuary focuses on real-time pipelines. It reads database change logs and streams changes within seconds, and it also covers batch SaaS sources through 200+ connectors. You can use it as SaaS or, on the Enterprise plan, run it privately or in your own cloud account.
Pricing combines the data you move with the connectors you run. In addition, a free developer plan covers up to 10 GB a month with two connectors. Because cost follows gigabytes, it stays predictable for steady database streams.
Best for: teams that need database changes in their data warehouse within seconds.
5. Integrate.io
Integrate.io is a low-code platform for ETL, ELT, CDC and reverse ETL, with 200+ sources and destinations. Unlike most managed tools, it charges a fixed fee, without per-row or per-connector metering. Recent releases also add Helm, which builds pipelines from a plain-language description.
The flat fee makes budgeting simple. On the other hand, the entry price is higher than a small team would pay on a usage plan. So it suits steady, substantial volumes best. Our Integrate.io alternatives list compares it with usage-priced tools.
Best for: mid-market and RevOps teams that value a predictable bill.
6. Skyvia
Skyvia, from Devart, is a browser-based, no-code platform with 200+ connectors. It covers import, export, replication, sync and backup, and it now also offers MCP endpoints for AI tools. Skyvia prices each product separately, so data integration costs follow the records you move each month, with 10,000 free.
Because it puts many small jobs on one bill, it suits teams that juggle several simple integrations. Still, check each plan’s limits against your volumes before you commit, or compare Skyvia alternatives.
Best for: non-technical teams with several simple integrations and a small budget.
7. Matillion
Matillion is a visual ELT tool for teams on Snowflake, Databricks, BigQuery or Redshift. Its pipelines push work down into the data warehouse, so transformations run as SQL on the data warehouse’s own compute. In 2025, it also launched Maia, an “agentic data team” that builds and maintains pipelines. Since then, the platform itself carries the Maia name.
Matillion charges consumption-based credits, but only while pipelines run. Keep in mind that pushdown also moves compute cost onto your data warehouse bill. Our Matillion alternatives post covers lighter options.
Best for: enterprise data teams on one cloud data warehouse that want visual or AI-assisted pipeline building.
Open-source and code-first tools
Open-source tools cost nothing to license and real money to run. After all, someone has to host them, upgrade them and fix a connector when an API changes. So they make sense when you have engineers, need an unsupported source, or must keep data on your own servers. Many teams also pair them with plain Python ETL scripts for the odd source.
8. Airbyte
Airbyte is one of the most widely used open ELT platforms, with 700+ connectors and a no-code connector builder. You can self-host Airbyte Core for free, or use Airbyte Cloud, either fully managed or in your own VPC. Note, however, that the platform uses the Elastic License 2.0, which is source-available rather than a classic open-source licence.
In September 2026, Airbyte changed its cloud pricing. Standard and Plus plans now bill by volume through credits, while larger plans use capacity-based pricing. Self-hosting avoids those fees but adds infrastructure and upkeep, and our Airbyte alternatives compare the managed options.
Best for: engineering teams that want a self-hostable tool with a long tail of connectors.
9. dlt
dlt is an Apache 2.0-licensed Python library for loading data. Instead of a UI, you write a short Python script, and dlt handles schemas, incremental loading and pagination. Because it is just a library, it runs anywhere Python runs, from a laptop to Airflow. In addition, its maker dltHub sells a managed runtime, paid with credits.
dltHub also targets engineers who build with AI coding agents, since a pipeline is just code an agent can write. The trade-off is ownership: every source you add is code your team maintains. For a middle ground, see our take on low-code Python for data integration.
Best for: Python data engineers who want pipelines as code.
10. Meltano
Meltano is an MIT-licensed, code-first ELT framework built on the Singer standard of taps and targets. In addition, Meltano Hub lists 600+ sources and destinations. Pipelines live in version control and run in CI/CD, and a managed Meltano Cloud is also available. Better still, the project is actively maintained, with a new release in October 2026.
Singer connectors vary a lot in quality, though, so test each tap you plan to use. Our Meltano alternatives post covers tools with maintained catalogs.
Best for: engineering teams that already manage infrastructure as code.
11. Apache NiFi
Apache NiFi is an Apache 2.0-licensed system for routing and transforming data flows, built on a drag-and-drop canvas of processors. It handles files, events and APIs well, and it also records the provenance of everything that passes through. Version 2 is current, since the 1.x line reached end of support in December 2024.
NiFi is also the engine under Snowflake Openflow, which raised its profile in 2025. Running it yourself, however, takes real operations work, from clustering to upgrades. Our overview of data automation tools puts it next to lighter options.
Best for: teams that move files, events and API data between many systems and want a visual flow editor.
12. Apache Airflow
Airflow is not an ETL tool in the strict sense. Rather, it is the Apache 2.0-licensed standard for scheduling and monitoring pipelines written as Python code. The current major version is Airflow 3, and Airflow 2 reached end of life in April 2026.
So the common mistake is to adopt Airflow expecting connectors. Instead, teams pair it with an ingestion tool such as Airbyte or dlt and keep Airflow for scheduling and dependencies. Managed versions also exist, such as Astro, Cloud Composer and Amazon MWAA. Our guide to data pipeline tools then shows how the pieces fit.
If Airflow feels too heavy for your team, our Airflow alternatives cover simpler schedulers.
Best for: engineering teams that orchestrate complex, custom pipeline logic.
Cloud-native services
If all your data already sits in one cloud or data platform, its own service is often the shortest path. That way, there is no new vendor, contract or security review. However, these services assume engineering skills, and they get weaker the further a source sits from their ecosystem.
13. AWS Glue
AWS Glue is a serverless data integration service. You write or visually build Spark jobs, and AWS then runs them without clusters to manage. It connects to 100+ data sources and works closely with S3, Redshift, Athena and Apache Iceberg tables. Glue 6.0 also adds Spark 4 and declarative pipelines in SQL and Python.
Pricing is per DPU-hour, billed by the second, so you only pay while jobs run. As a result, Glue shines for large datasets landing in S3 or Redshift. By contrast, it is heavy for a team that just wants a CRM synced every hour.
Best for: AWS teams with Spark skills that process large datasets.
14. Azure Data Factory and Fabric Data Factory
Azure Data Factory (ADF) is Azure’s managed pipeline service, with 90+ built-in connectors. In addition, a self-hosted runtime reaches on-premises sources. It can also run existing SSIS packages in the cloud. Pricing combines pipeline runs, data movement hours and activity hours.
However, Microsoft now tells new users to start with Data Factory in Microsoft Fabric, the next generation of ADF. Fabric connects to 170+ sources and bills through capacity units instead. According to Microsoft, Fabric already offers about 90% of ADF activities, though SSIS support there is still undecided.
For options outside Microsoft, see our Azure Data Factory alternatives.
Likewise, our Microsoft Fabric alternatives cover the wider Fabric platform.
Best for: Microsoft-centric companies, with ADF for existing estates and Fabric for new projects.
15. Google Cloud Dataflow
Dataflow is Google Cloud’s fully managed service for running Apache Beam pipelines. It handles batch and streaming in one programming model and scales workers automatically. Meanwhile, billing is per second, for vCPU, memory and data shuffled. In fact, it is one of the strongest options here for large-scale streaming, such as clickstreams or IoT events landing in BigQuery.
Like Glue, however, it assumes engineers, because Beam pipelines are code. For a visual option on Google Cloud, Cloud Data Fusion offers drag-and-drop pipelines, priced per instance hour.
Best for: Google Cloud teams with streaming workloads and Beam skills.
16. Databricks Lakeflow
Lakeflow is Databricks’ data engineering suite, generally available since June 2025. It combines Lakeflow Connect for ingestion, Spark Declarative Pipelines for transformation, Lakeflow Jobs for orchestration and a visual Lakeflow Designer. Managed connectors cover SaaS apps such as Salesforce, HubSpot, Jira and Workday, plus CDC from MySQL, PostgreSQL and SQL Server.
Managed connectors run on serverless compute. Databricks then bills them in Databricks units (DBUs), with a daily free allowance per workspace. Because the data lands in your Databricks lakehouse, Lakeflow mainly suits teams already on Databricks. Our Databricks alternatives cover lighter platforms.
Best for: Databricks customers who want ingestion, transformation and orchestration in one place.
17. Snowflake Openflow
Openflow is Snowflake’s ingestion service, built on Apache NiFi and generally available since May 2025. It offers dozens of ready-made connectors and hundreds of NiFi processors for structured and unstructured data. You can run it inside Snowflake on Snowpark Container Services, or in your own AWS account.
Snowflake bills it in credits based on uptime and compute, so the cost lands on your existing Snowflake bill. For teams already standardized on Snowflake’s data cloud, that removes a vendor. For everyone else, though, it adds a dependency.
Best for: Snowflake customers who want ingestion without another vendor.
Enterprise suites
Enterprise suites bundle integration with data quality, master data management and governance. Large organizations buy them because they run many systems, face strict audits and have dedicated platform teams. Licences and projects are large, and so is the cost of leaving.
18. Informatica
Informatica’s Intelligent Data Management Cloud (IDMC) combines integration, data quality, master data management and governance. Salesforce completed its acquisition of Informatica in November 2025, and the products now carry the “Informatica from Salesforce” brand. Pricing uses consumption-based Informatica Processing Units (IPUs).
Meanwhile, Informatica set the end of support for the on-premises PowerCenter 10.5.x on 31 March 2026. Its path forward is PowerCenter Cloud Edition on IDMC, which runs existing mappings without a rebuild. If you face that move anyway, our guide to data management helps frame what you actually need.
Best for: large enterprises where governance, lineage and MDM matter as much as pipelines.
19. Qlik Talend Cloud
Qlik Talend Cloud brings Talend and Qlik’s own data integration products together under Qlik, which acquired Talend in 2023. It covers pipelines, data quality, governance and CDC replication, and it connects to hundreds of sources and targets. Pricing comes in four editions, measured on data volume moved, job executions and execution duration.
Qlik retired Talend’s free Open Studio in January 2024. As a result, teams that chose Talend for open source now face a commercial suite. In return, they get strong CDC replication from operational databases, based on the Attunity technology Qlik bought in 2019. Recently, Qlik also added an MCP server for AI agents.
Best for: enterprises that want integration, data quality and CDC in one commercial suite.
20. IBM DataStage
DataStage is IBM’s enterprise ETL engine, known for parallel processing of very large volumes. It now sits inside IBM watsonx.data integration. Also, it runs on-premises, in the cloud or in hybrid setups with a remote execution engine. For the managed DataStage as a Service, IBM charges per capacity unit-hour.
In October 2026, IBM also made agentic data integration generally available, adding AI help for pipeline design and operations. Still, implementations remain substantial projects, with licence and consulting costs to match.
Best for: IBM-centric enterprises with high-volume, hybrid workloads.
21. Oracle Data Integrator
Oracle Data Integrator (ODI) is Oracle’s E-LT tool: you design mappings, and ODI pushes the work to the target database. Version 14c adds Python for its Knowledge Modules, and ODI is also available on Oracle Cloud Marketplace. It also pairs naturally with Oracle GoldenGate for real-time replication.
Inside Oracle-standardized companies, it remains a solid choice. Elsewhere, though, the skills pool is narrow, and newer tools fit modern cloud data warehouses more easily.
Best for: enterprises with large Oracle estates and existing ODI skills.
22. Microsoft SSIS
SQL Server Integration Services (SSIS) ships with SQL Server Standard and Enterprise, including SQL Server 2025. So for many Microsoft shops, it is effectively already paid for. Moreover, it is mature, well documented and familiar to a generation of database administrators. Besides, existing packages can move to Azure through the SSIS runtime in Azure Data Factory.
However, Microsoft’s investment has moved to Data Factory and Fabric. SQL Server 2025 even removed some SSIS components, such as the Attunity CDC components and the Oracle connector. So existing SSIS estates will run for years. But new pipelines that start in SQL Server and end in the cloud usually belong elsewhere.
Best for: on-premises SQL Server environments with existing SSIS packages.
How ETL tools charge
Five pricing models cover almost every tool on this list. The model matters more than the list price, because it decides what makes your bill grow. Our guide to data pipeline pricing goes deeper. Below are the basics, with figures from vendor pricing pages in November 2026.
Rows or events that change
Fivetran, Hevo Data and Stitch charge for the records that change. At the low end, free tiers are generous: Fivetran’s covers 500,000 monthly active rows, and Hevo’s covers 1 million events. Yet cost follows how much your data changes, not how much of it you use. For example, a CRM that stamps every record with a new date each night can multiply the bill.
Data volume
Estuary charges per gigabyte moved plus a monthly fee per connector: $0.50 per GB and $100 per connector instance. Similarly, Airbyte Cloud’s Standard and Plus plans now bill by volume through credits. Volume pricing is easy to forecast for steady streams, although large historical backfills still add up.
Compute time
Cloud-native services charge for the compute your jobs use. AWS Glue, for instance, lists Spark jobs on Glue 6.0 at $0.308 per DPU-hour in US East. Billing then runs per second, with a one-minute minimum. So a nightly 15-minute job on 10 DPUs uses 2.5 DPU-hours, or about $0.77 a run and $23 a month. In the same way, Matillion credits, Databricks units and Snowflake credits track compute. Therefore, efficient jobs are cheap and slow ones are not.
Flat fees
Integrate.io lists a flat fee of $1,999 a month without per-row metering. Likewise, other platforms charge a fixed price per connected source. Budgeting is therefore simple, but you pay the same for a quiet month as for a busy one.
Licences and capacity
Enterprise suites sell editions, capacity units or processing units, usually through a sales process. SSIS is the exception, since it comes with SQL Server licences. In both cases, the upfront commitment is large, but monthly costs barely move.
What to test before you choose
Demos use clean sample sources, but your real sources will not be clean. For that reason, a two-week trial on your two hardest sources tells you more than any comparison table, ours included. These checks catch most bad fits. Once a tool is running, our ETL best practices guide covers what comes next.
Connector depth, not connector count
A connector count says nothing about what each connector actually syncs. So ask for the field-level schema of your top three sources and compare it with what your reports need. In particular, check custom fields, deleted records and history. Some connectors only keep the current state of each record, so “what changed last quarter” becomes hard to answer.
This matters most for SaaS apps, because each API exposes data differently, as our guide to SaaS data integration explains.
API limits and the first load
SaaS APIs cap how fast any tool can read. For example, HubSpot’s API usage guidelines limit privately distributed apps on Free and Starter accounts to 100 requests per 10 seconds. On top of that, every app on the account shares a daily allowance of 250,000 requests. As a result, a first load of five years of history can eat that allowance and slow your other integrations. Ask how the tool backs off, resumes after an error and spreads a large backfill.
Incremental syncs and change data capture
After the first load, a tool either asks the source what changed since the last run or reads the change log. The first method usually relies on an “updated at” timestamp. Log-based change data capture (CDC) also catches deletes and every change in between. However, it needs work on your side. PostgreSQL, for instance, only allows it once you set wal_level to logical on the source database.
Similarly, MySQL, SQL Server and Oracle need their own settings. Debezium, the open-source CDC project, documents what each database needs. So its docs are a useful reference even if you never run it.
Schema changes
Sooner or later, someone adds a custom field or renames a column. So find out whether the tool adds new columns automatically, pauses the sync, or silently drops the field. Equally important, find out whether it tells you when that happens.
Where the data lands, and how you leave
Check which destinations each tool supports and in what format. Tools that write to open table formats such as Apache Iceberg keep data readable by other engines. That way, a later switch hurts less. Also ask what happens to your pipelines and history if you cancel, because that is your real exit cost.
No destination yet? Then start with our guide on how to build a data warehouse, since it shapes which tools fit.
Security and data residency
Ask for the SOC 2 Type II report itself, not just the badge on the website. Then check where the tool processes data, not only where it stores it. For EU personal data, any transfer outside the EEA needs an adequacy decision or safeguards such as standard contractual clauses. The European Commission’s rules on international data transfers explain the options. Consequently, a managed tool that routes data through US regions needs that paperwork in place.
Whether AI will read the data
More pipelines now feed AI assistants and agents, not only dashboards. If that is on your roadmap, check whether the tool or your data warehouse offers a Model Context Protocol server. That way, assistants query clean, synced tables instead of calling every app’s API on the fly.
Where Peliqan fits
Peliqan is an all-in-one data platform, so it is not a pure ETL tool. It suits teams that would otherwise buy four products: connectors, a data warehouse, a transformation tool and a way to push data back into apps. It comes with 300+ connectors, and the team builds a missing connector on request within two weeks.
Synced data lands in a built-in data warehouse, Postgres with Trino for queries, or in your own data warehouse, such as Snowflake, BigQuery, Redshift or SQL Server. On top of that, you transform data in SQL or low-code Python in the same place.
From there, the data can flow back into business apps through reverse ETL and writeback. It can also feed BI tools such as Metabase, which you can deploy in one click from the Peliqan Marketplace.
In addition, AI assistants such as Claude and ChatGPT can query the same data through Peliqan’s MCP server. That way, answers come from synced tables rather than live API calls.
Pricing is a flat monthly fee per connected data source, listed on the pricing page. Also, Peliqan is SOC 2 Type II and ISO 27001 certified, GDPR compliant and hosted on AWS in Frankfurt.
For example, CIC Hospitality brought data from 50+ sources into board-level reports with Peliqan. As the CIC Hospitality case study describes, that saves about 40 hours a month of manual Excel work.
Where Peliqan is a weaker fit
Peliqan is a weaker fit in three cases. First, some teams only need to move very large volumes from one production database into an existing data warehouse within seconds. A CDC specialist does that more directly. Second, some teams already run a managed ELT tool, dbt and a cloud data warehouse that work well. For them, another platform for one extra source rarely pays off. Third, heavy Spark workloads belong on a cloud-native engine. For a side-by-side with the most common managed tool, see Peliqan vs Fivetran.
Bottom line
Start from your team and your sources, not from a feature list. For instance, teams with engineers and unusual sources get the most from open-source tools. Funded data teams with a data warehouse usually pay for managed ELT to avoid maintenance. Meanwhile, single-cloud teams should try their platform’s own service first, and large regulated organizations still buy enterprise suites.
Whichever type you choose, test it on your two hardest sources before you sign. Also, model the bill with your own data volumes. And if you want a second opinion on your setup, talk to us.
