How to choose an ELT tool

An ELT tool extracts data from sources — databases, SaaS apps, event streams, files — loads it into a destination warehouse or lake largely unmodified, and leaves transformation for a downstream tool such as dbt to run inside that destination. This is the practical successor to ETL, where transformation happened in the pipeline itself before loading; pushing transformation into the warehouse's own compute is usually cheaper and easier to version-control now that warehouses are fast and cheap to query. Every tool in this category solves the same basic problem — get data from A to B reliably, on a schedule or continuously, without you writing custom extraction code for each source.

You don't need a dedicated ELT platform if you have one or two sources and a script already does the job reliably; the value of these tools is in the connector maintenance (APIs change, and someone has to keep the connector working) and the operational tooling (retries, schema drift handling, monitoring) that becomes worthwhile once you have more than a handful of pipelines.

Managed connectors vs a visual canvas vs code-first

The category splits into three working styles. Point-and-click managed connectors — Fivetran, Hevo Data, Stitch, Portable — ask you to authenticate a source and a destination and handle the rest; you write no pipeline logic. Visual, low-code canvases — Matillion, Rivery, Integrate.io, Azure Data Factory — give you a drag-and-drop designer plus SQL or scripting components for anything the canvas can't express, which suits teams that want more visible control over pipeline logic than pure managed connectors allow. Code-first, open-source frameworks — Airbyte, Meltano — define pipelines as versioned code and favor engineering teams who want pipelines in Git alongside the rest of their infrastructure. Cloud-native compute services — AWS Glue, Google Cloud Dataflow — are a fourth category again: serverless Spark or Beam execution engines for teams writing their own transformation code rather than configuring prebuilt connectors.

Self-hosted control vs managed simplicity

Airbyte and Meltano are the two genuinely open-source options here, each with a free self-hosted core (Airbyte OSS, Meltano Open) and a managed cloud counterpart layered on top for teams that want the same engine without running it themselves. Everything else in this category is cloud-only or cloud-first: Fivetran, Hevo Data, Stitch, Portable, Estuary Flow and Rivery have no self-hosted mode at all. That matters for data residency and for who's on call when a pipeline breaks at 2 a.m. — with a managed tool it's the vendor's problem first; with Airbyte OSS or Meltano Open it's yours, which is also why the trade for "free" software is real operational ownership, not zero cost.

Connector breadth and the long tail

Fivetran (700+ connectors) and Airbyte (600+) lead on raw connector count, with Estuary Flow (200+), Hevo Data (150+) and Meltano's Singer ecosystem (600+ via Cloud) not far behind. But breadth numbers hide the real question: does the tool have your specific, obscure source? Portable exists specifically for that gap — it specializes in long-tail, hard-to-find connectors and will build one on request, which the larger platforms rarely prioritize for a niche SaaS tool with a small customer base.

Streaming and change data capture, if you need it

Most of these tools are batch- or micro-batch-first, syncing on a schedule measured in minutes. If you need continuous, low-latency delivery, check for it explicitly: Estuary Flow is built around unifying batch and streaming from the same pipeline definition with sub-second latency; Airbyte, Fivetran, Matillion and Rivery all offer change data capture for databases as a feature rather than the whole product's design center. If streaming CDC is your primary requirement rather than a nice-to-have, it's worth comparing against the dedicated change data capture category instead of treating it as a checkbox here.

Enterprise governance vs focused simplicity

Informatica and Talend (now sold as Qlik Talend Cloud) are the two enterprise-suite options: both bundle data integration with master data management, data quality and cataloging, aimed at large organizations that want one governed platform rather than best-of-breed tools stitched together. That breadth comes with enterprise sales cycles and quote-based pricing. At the other end, Stitch deliberately does less — simple extract-and-load, no transformation engine, minimal configuration — for teams that want a pipeline running today, not a platform to administer.

How pricing scales

This category has more pricing models than most, and the model matters as much as the rate. Usage-based on volume (Fivetran on Monthly Active Rows, Hevo Data and Estuary Flow on data volume, Rivery on consumption credits, Stitch on rows replicated, Airbyte on Cloud credits) means cost tracks growth automatically but can surprise you at scale. Flat-fee, volume-independent (Integrate.io, Portable) trades a higher fixed floor for a predictable ceiling — attractive once volume-based pricing starts to sting. Serverless compute metering (AWS Glue per Data Processing Unit-hour, Azure Data Factory across several separate meters, Google Cloud Dataflow per vCPU/memory-hour) bills for compute time rather than rows, which suits variable batch workloads better than steady, high-frequency ones. Quote-only (Matillion, Meltano Cloud, Informatica, Talend) means no public number at all — budget for a sales conversation.

A shortlist by situation

• Want the broadest managed connector catalog with nothing to operate: Fivetran.
• Want the same breadth but self-hosted, free, and open source: Airbyte.
• Want pipelines defined as versioned code, Git-native: Meltano.
• Need a specific niche connector the big platforms don't prioritize: Portable.
• Want a visual canvas with push-down transformation into your warehouse: Matillion.
• Already on Azure or AWS and want the pipeline orchestration native to that cloud: Azure Data Factory or AWS Glue.
• Writing your own Beam pipelines and want unified batch and streaming: Google Cloud Dataflow.
• Want unlimited volume for one flat monthly fee: Integrate.io.
• Want ingestion, orchestration and Reverse ETL in one consumption-billed platform: Rivery.
• Small team, want the simplest possible no-code setup: Hevo Data or Stitch.
• Large enterprise needing integration bundled with governance and master data management: Informatica or Talend.

Questions to ask a vendor or in a trial

1. Does the connector you need exist today, or is it "on the roadmap" — and if missing, what's the turnaround to build it?
2. How is a source API rate limit or schema change handled — silently retried, or does the pipeline pause and alert you?
3. What exactly is metered (rows, events, GB, compute-hours), and can you get a realistic cost estimate from your actual data volume before committing?
4. Is transformation native to the tool, or does it hand off to dbt — and if the former, can you still version-control the logic?
5. What's the actual minimum sync frequency, and does it change by pricing tier?

Common mistakes

Picking on connector count alone and finding your specific source isn't actually supported. Underestimating usage-based pricing by testing with a small pilot dataset that doesn't reflect production volume. Assuming "ELT" means no data modeling work remains — loading data unmodified is the easy part; someone still owns the transformation layer and its data catalog. And choosing a self-hosted open-source tool for its zero license cost without budgeting the engineering time to run it in production.

For head-to-head looks at specific pairs, see Airbyte vs Meltano, Fivetran vs Matillion and Informatica vs Talend. Every tool in this category is listed at /tools/category/elt-data-integration/.