· Updated · 15 min read
10 ETL Tools Compared by Ownership and Reliability
Ten ETL tools for data leaders, compared by who owns pipeline logic, how failures recover, what evidence remains, and who operates execution.
Extraction, transformation, and loading (ETL) tools leave your organization with an operating model after the first load. That model determines who can inspect changes, approve releases, maintain pipelines, and recover failed runs. Connector count and setup speed can help with an initial shortlist. However, neither proves that a pipeline is ready for production. This comparison examines control, recovery, lineage, testing, execution, and pricing across ten ETL tools.
Which operating model fits your team?
Vendor-managed execution fits organizations that want to transfer infrastructure work; customer-reviewed pipeline logic fits organizations that need internal approval and maintenance; customer-operated execution fits organizations that must control the environment. A product can combine these models.
The UK National Cyber Security Centre connects maintainable systems with version control, attributed changes, peer review, testing, and rollback. The GOV.UK guidance connects live operation with checks, alerts, recorded errors, and diagnosis.
Managed execution transfers infrastructure work, but it does not transfer accountability for pipeline behavior. The National Institute of Standards and Technology (NIST) separates cloud service models by the resources that providers and customers control.
Control, recovery, and retained evidence
The roster contains dltHub, Fivetran connectors, Airbyte connections, AWS Glue jobs, and Azure Data Factory. It also contains Informatica data integration, Qlik Talend Cloud, Matillion setup overview, Hevo Pipeline, and Stitch.
The first table summarizes control, recovery, and retained evidence.
| Tool | Best for | Pipeline logic | Changes and failures | Lineage, logs, and tests |
|---|---|---|---|---|
| dltHub | Customer-owned pipeline code | Python in customer repository | Contract actions and run logs | Runs, logs, schemas, and lineage |
| Fivetran | Vendor-maintained standard connectors | Configuration or Connector SDK code | Automatic retries and historical resync | Sync history, logs, audit events, lineage |
| Airbyte | Vendor-managed or hybrid deployment | Customer-configured connections | Schema review, repair, and resync | Timelines, logs, and audit records |
| AWS Glue | ETL within AWS | Customer-controlled job definitions | Configured retries and state bookmarks | Logs, metrics, history, audit identity |
| Azure Data Factory | ETL within Azure | Source control | Configured retries and activity reruns | Run history and optional Purview lineage |
| Informatica | Mappings with several runtime choices | Mappings, tasks, and source control | Schema refresh, restart, and resume | Logs, audit records, and lineage |
| Qlik Talend Cloud | Graphical design with engine choice | Graphical processes and run profiles | Schema-change policies and reruns from a selected step | Preview, logs, and compatible-job lineage |
| Matillion | Managed pipelines with approval controls | Components, flow, and source control | Earlier artifacts can be redeployed | Dry runs; Scale lineage; Teams/Scale audit |
| Hevo | Vendor-operated pipeline execution | Customer-configured pipeline settings | Schema-change policies and replay of failed events | Tests, failed events, Audit Tables (Early Access) |
| Stitch | Managed extraction and loading | Customer-configured replication | Schema responses and buffered delivery | Statistics, extraction logs, loading reports |
The second table summarizes execution.
| Tool | Execution |
|---|---|
| dltHub | Managed SaaS or custom deployment |
| Fivetran | Vendor-managed |
| Airbyte | Vendor-managed or hybrid deployment |
| AWS Glue | AWS-managed |
| Azure Data Factory | Azure or customer-operated runtime |
| Informatica | Hosted, serverless, or customer agent |
| Qlik Talend Cloud | Qlik or customer infrastructure |
| Matillion | SaaS or Scale hybrid |
| Hevo | Vendor-managed |
| Stitch | Vendor-managed |
How ownership, recovery, and execution narrow the list
Responsibility, rather than connector coverage, separates the operating models in this comparison. Each recommendation below points to controls documented for the named product or option.
| If your organization… | And it wants to keep or transfer… | Start with… |
|---|---|---|
| Requires changes to pass internal review | Keep control of code, source history, or approvals | dltHub; Fivetran Connector SDK; Azure Data Factory; Informatica; Matillion |
| Wants vendor-operated pipelines | Transfer execution and routine connector operation | Fivetran standard connectors; Airbyte Cloud; Hevo; Stitch |
| Standardizes execution on AWS or Azure | Keep deployment inside its named cloud environment | AWS Glue; Azure Data Factory |
| Must run work on customer infrastructure | Keep the execution environment | dltHub custom deployment; Airbyte Enterprise Flex hybrid deployment; Azure self-hosted integration runtime; Informatica Secure Agent; Qlik customer engines; Matillion Scale |
| Needs graphical pipeline design | Keep configuration while transferring some execution | Azure Data Factory; Informatica; Qlik Talend Cloud; Matillion |
These starting points do not replace a review of each entry. The product option must document the control or deployment choice that your organization requires.
ETL tools
Each entry covers control of pipeline logic, changes and failures, lineage and testing, execution, pricing, and the work that remains.
dltHub
Best for: Organizations that want pipeline code in their repository while dltHub operates the managed runtime.
Who controls and can review the pipeline logic: Your organization defines the behavior in Python and keeps it in its repository. The documented deployment process supports automated checks, human review, approval, and releases through the customer’s continuous integration process. Related dltHub articles cover the shared ground on Fivetran comparisons, production Python ETL, and portable ETL.
What happens when source data changes or a run fails: Schema contracts can accept a change, fail the run, discard a row, or discard a value. Retained run logs support failure investigation.
What lineage and testing evidence remains: The managed runtime retains runs, logs, schemas, and lineage.
Who operates execution: dltHub operates the managed service, scheduling, monitoring, and runtime maintenance. Enterprise SaaS can use private links. A custom BYOC deployment, designed with the customer, runs pipelines in the customer’s cloud.
Pricing: The published plan is USD 12,000 per month with 5,000 credits. It is billed monthly with a 12-month minimum.
What stays your job: Your organization retains pipeline code, credentials, sources, destinations, review controls, tests, and deployment decisions.
Fivetran
Best for: Organizations that want Fivetran to maintain and operate standard connectors. Connector SDK fits organizations that need customer-defined extraction code.
Who controls and can review the pipeline logic: Customers configure connections, selected data, schedules, and supported history windows. Connector SDK code can be tested locally and deployed through command-line, application programming interface, or continuous integration processes.
What happens when source data changes or a run fails: Supported schema changes propagate to the destination. A failed sync writes no destination data, and Fivetran retries it automatically. Customers can start a historical resync when recovery requires reloading data.
What lineage and testing evidence remains: Sync History records timing, stages, errors, actions, and row counts. The Platform Connector adds logs, audit events, configuration records, and column-lineage metadata.
Who operates execution: Fivetran operates hosted execution, scheduling, retries, and connector infrastructure.
Pricing: The Free plan publishes usage limits. Paid plans use monthly active rows and successful model runs; contact Fivetran for paid-plan pricing.
What stays your job: Your organization maintains access, configuration, monitoring, remediation, and resync decisions.
Airbyte
Best for: Organizations choosing between vendor-managed plans and Enterprise Flex hybrid deployment.
Who controls and can review the pipeline logic: Customers configure selected data, sync behavior, schedules, schema-change policy, mappings, and transformations.
What happens when source data changes or a run fails: Airbyte Cloud checks for schema changes before syncs. Its schema policies can propagate a change, request approval, or stop future syncs. Breaking changes pause the connection, while timelines and logs support repair and resync.
What lineage and testing evidence remains: Pro and Enterprise Flex audit logs retain the actor, operation, status, and configuration metadata for 365 days. They do not provide row-level lineage.
Who operates execution: Airbyte operates its Cloud plans and the Enterprise Flex control plane. Enterprise Flex uses a hybrid deployment in which data stays inside the customer’s boundary.
Pricing: Standard pricing starts at USD 20 per month. Plus displays USD 189 without a billing period. Contact Airbyte for Pro and Enterprise Flex pricing.
What stays your job: Your organization configures connections, investigates failures, approves applicable schema changes, and starts repairs or resyncs.
AWS Glue
Best for: Organizations that want AWS to manage serverless execution while they retain job logic, configuration, and operating decisions.
Who controls and can review the pipeline logic: Customers create, inspect, and change job definitions. A definition determines processing behavior, capacity, connections, retries, and security settings.
What happens when source data changes or a run fails: Crawlers update schemas according to customer-selected change policies. Run records expose failure outcomes, while customers configure retries. Job bookmarks support resuming or reprocessing data.
What lineage and testing evidence remains: AWS monitoring provides CloudWatch logs and metrics. Run history records execution, and CloudTrail records requester identity.
Who operates execution: AWS provisions and operates the AWS Glue infrastructure.
Pricing: AWS Glue pricing has separate usage categories. Core ETL jobs and crawlers cost USD 0.44 per data processing unit hour, with regional variation.
What stays your job: Your organization retains access, permissions, encryption choices, monitoring, retry safety, output correctness, schema policy, and failure response.
Azure Data Factory
Best for: Organizations that want Azure-managed execution with pipeline definitions in source control and an optional customer-operated runtime.
Who controls and can review the pipeline logic: Source control supports branches, peer review, test environments, and debug runs. Retained deployment artifacts permit redeployment of an earlier configuration.
What happens when source data changes or a run fails: Schema drift applies to configured mapping data flows, not to each activity. Customers set retry policies and can rerun a pipeline from its beginning or from a failed activity.
What lineage and testing evidence remains: Run records include status, timing, errors, inputs, outputs, and runtime. Native history lasts 45 days, and Microsoft Purview can add lineage.
Who operates execution: Microsoft operates the Azure integration runtime. Customers operate machines used by the self-hosted integration runtime.
Pricing: Azure pricing uses separate units for runs, activities, runtime, and operations. The page provides no verifiable numeric starting price.
What stays your job: Your organization retains connections, permissions, alerts, log retention, deployment processes, and external systems.
Informatica Cloud Data Integration
Best for: Organizations that want mappings and tasks with hosted, serverless, or customer-operated agent options.
Who controls and can review the pipeline logic: Customers configure mappings, tasks, transformations, runtime settings, sources, and targets. Source control supports history, version comparison, and reversion to an earlier version.
What happens when source data changes or a run fails: Schema refresh can apply supported additions or deletions. CDC mapping tasks can restart or resume from a checkpoint when recovery is enabled. Suspended taskflows can retry or skip a faulted step.
What lineage and testing evidence remains: Monitoring, logs, audit records, and Metadata Command Center lineage remain available. Mapping-task logs retain ten runs by default.
Who operates execution: Informatica provides Hosted Agent, serverless AWS or Azure, and customer-installed Secure Agent execution.
Pricing: Pricing uses Informatica Processing Units. Contact Informatica for pricing.
What stays your job: Secure Agent customers maintain hosts, connectivity, upgrades, logs, and troubleshooting. All customers retain configuration and operating decisions.
Qlik Talend Cloud
Best for: Organizations that want graphical pipeline design with Qlik-managed or customer-infrastructure engines.
Who controls and can review the pipeline logic: Customers define graphical processes, datasets, processors, and run profiles. Live preview and metrics help them inspect results before a full run. Qlik retired Talend Studio on 31 January 2024.
What happens when source data changes or a run fails: Schema-change policies can apply, ignore, suspend, or stop detected changes. Customers can rerun a plan from a selected step and choose only its failed tasks.
What lineage and testing evidence remains: Task logs record messages, severity, timestamps, and versions. Compatible jobs can send dataset lineage to Qlik Cloud.
Who operates execution: Cloud Engine is Qlik-managed. Remote Engine and Dynamic Engine run on customer infrastructure.
Pricing: Starter, Standard, Premium, and Enterprise use capacity-based subscriptions. Contact Qlik for pricing.
What stays your job: Your organization configures tasks, schedules, engines, logging, lineage collection, and recovery actions. Customer-engine users also operate the infrastructure.
Matillion Data Productivity Cloud
Best for: Organizations that want managed pipelines with inspectable logic, dry runs, quality checks, approval controls, and an optional Scale hybrid deployment.
Who controls and can review the pipeline logic: Customers inspect components, data flow, logic, run status, errors, row counts, and the artifact version. Source control and approval controls support governed changes.
What happens when source data changes or a run fails: Customers can revert source-control changes or deploy an earlier immutable artifact.
What lineage and testing evidence remains: Dry runs and quality checks support validation. Scale includes lineage for datasets and columns. Teams and Scale include audit logs for user, time, action, and system-event records.
Who operates execution: Full SaaS leaves runner infrastructure with Matillion. Scale can use customer-hosted Agent infrastructure and offers PrivateLink as an add-on.
Pricing: Developer, Teams, and Scale use credits through fixed annual packages. Contact Matillion for pricing.
What stays your job: Your organization defines pipelines, review controls, approvals, data access, and recovery decisions. Scale customers also operate the Agent infrastructure.
Hevo Pipeline
Best for: Organizations that want Hevo to schedule, process, and monitor execution while customers configure pipeline behavior.
Who controls and can review the pipeline logic: Customers configure sources, transformations, schema-change policy, destinations, and load settings. They can inspect progress, failures, and sample failed events and can test source connections and transformations.
What happens when source data changes or a run fails: Schema policies can allow all changes, block all changes, or allow only column-level changes. Incompatible mappings fail the batch. Failed events remain available for seven days for replay or skipping.
What lineage and testing evidence remains: Audit Tables, currently in preview (Early Access), as of 12 November 2026, retain run, ingestion, loading, failure, and destination records.
Who operates execution: The pipeline framework schedules, processes, and monitors execution.
Pricing: The pricing page lists Starter at USD 265 per month and Professional at USD 750 per month on annual plans.
What stays your job: Your organization maintains access, configuration, monitoring, schema decisions, and failure resolution before failed events expire.
Stitch
Best for: Organizations that want Stitch to execute extraction and loading while customers configure replication.
Who controls and can review the pipeline logic: Customers use the pipeline setup to configure sources, destinations, selected data, schedules, and replication methods. Stitch and Qlik remain separate product experiences within Qlik.
What happens when source data changes or a run fails: New columns replicate automatically. Removed columns remain in the destination with null values. Some structural changes require full-table re-replication, while temporary buffering protects data during destination outages.
What lineage and testing evidence remains: Stitch retains replication statistics, extraction logs, loading reports, and rejected-record details.
Who operates execution: Stitch executes extraction, preparation, and loading.
Pricing: Standard pricing starts at USD 100 without a stated period. Advanced is USD 1,500 monthly equivalent billed annually. Premium displays USD 3,000 as a monthly equivalent.
What stays your job: Your organization retains credentials, privileges, replication choices, monitoring, and destination analysis.
How do ETL tools differ from orchestration and transformation tools?
ETL includes transformation and can run under orchestration, but it is not the same as either. An academic ETL reference describes extraction, transformation, and loading into a data warehouse as its three stages. Transformation is the middle stage when data changes into the required state or format.
Workflow orchestration coordinates automated tasks across systems in the correct order. A GitHub engineering article identifies ETL pipelines as one orchestration use case. Orchestration can coordinate ETL, while transformation can form one stage within ETL.
Book a call to try dltHub with pipeline logic your team can review.
ETL pricing
The table shows each vendor’s published billing basis and starting-price information.
| Tool | What you pay for | Published starting price* |
|---|---|---|
| dltHub | Credits per month | USD 12,000 monthly; 12-month minimum. PAYG coming soon. |
| Fivetran | Active rows and model runs | Free plan; contact for paid plans |
| Airbyte | Credits or Data Workers | Standard: USD 20 per month |
| AWS Glue | Separate usage units | USD 0.44 per DPU-hour for core ETL |
| Azure Data Factory | Runs, activities, runtime, operations | No verifiable numeric starting figure |
| Informatica | Informatica Processing Units | Contact Informatica |
| Qlik Talend Cloud | Volume, executions, duration | Contact Qlik |
| Matillion | Credits and task hours | Contact Matillion |
| Hevo | Events or fixed pipeline pricing | USD 265 monthly on annual plan |
| Stitch | Monthly row volume | USD 100; period not stated |
* Pricing and product information come from each vendor's public website and documentation, linked in the text, as reviewed on 12 November 2026. Check the linked source or contact the vendor for current terms. Where a vendor does not publish a price, the article says so.
What is dltHub?
The agentic data layer on top of any data warehouse: one governed layer and a managed runtime to build, govern, heal and serve data for AI, without growing a platform team.
Was this page helpful?