How much does data integration cost?

There is no single answer because the platforms do not sell the same unit. Flat plans on this site start at $49 a month. Task-metered and row-metered platforms depend entirely on your volume, and the same workload can bill as one event or two thousand depending on which meter you bought. Price your own workload against each vendor's published counting rule before you compare any headline number.

What are the main iPaaS pricing models?

Five, and they behave differently under growth. Per successful action, used by Zapier and Workato. Per module run per record, used by Make. Per workflow run regardless of size, used by n8n. Per row or per volume moved, used by Fivetran and Airbyte. And flat entitlement pricing, used by Celigo and by us, where you buy connections rather than consumption. Enterprise suites such as Informatica sit on top with pre-paid credit pools.

Why is my integration bill higher than I expected?

Usually because the meter counts something you were not thinking about. Replayed runs count again on Zapier. A called sub-recipe bills its own actions on Workato. A full refresh charges every row again on Airbyte even though nothing changed. A schedule trigger on n8n counts every time it fires whether or not there was anything to do. None of these are hidden fees. They are documented counting rules that buyers rarely read before signing.

What happens when you exceed your plan limit?

The platforms split into two camps and this is the single most important question to ask. Zapier holds new runs until your billing cycle resets unless you have turned on pay-per-task. Airbyte does the opposite: it does not stop in-progress syncs and does not pause scheduled ones, and bills you in arrears. One camp risks stopping your business, the other risks an invoice you did not approve.

Is usage-based pricing cheaper than flat pricing?

At low volume yes, and it inverts as you grow. Usage pricing is attractive in the pilot, when you are moving a few thousand records and paying almost nothing. The crossover comes when a normal business year doubles your row count and the bill doubles with it, which is why teams describe it as a success tax. Flat pricing costs more on day one and stops changing after that.

What is Monthly Active Rows in Fivetran pricing?

MAR is the number of distinct rows synced from source to destination in a calendar month, tracked by primary key. The rule that surprises people is favourable: a row synced a hundred times in one month counts once. Initial syncs, new columns on existing tables and Fivetran system tables are not charged. The rule that hurts is that a re-sync counts any rows that genuinely changed.

Does a filter step cost money?

It depends on the platform and the difference is large at volume. Zapier does not charge for Filter or Paths steps, and does not charge for actions that error or never run. Workato does not count skipped conditional branches or control statements. Make counts one credit per module run per bundle, so how you structure a scenario changes the bill directly. n8n does not care at all, because the whole run is one execution.

How do I estimate integration costs before buying?

Write down one real job and count it four ways. Take the number of records it moves, the number of write steps it performs, how often it runs, and how often it will be replayed or re-synced. Then apply each vendor's counting rule to those four numbers. Do it for your busiest month rather than an average month, because the meters bill the peak and the budget has to survive it.

Is per-task pricing bad?

It is honest for low-volume automation and punishing for data movement. A few hundred tasks a month moving records between business apps is cheap and fair. The same model applied to a nightly sync of an orders table turns every row into a billable event, which is how teams end up paying four figures a month to keep two systems in agreement. Match the meter to the shape of the work.

What is an IPU in Informatica pricing?

An IPU is a pre-paid unit of credit you buy annually and then draw down across Informatica services. The complication is that each service consumes on its own scalar: Data Integration usage is measured in Compute Units, Cloud Integration Hub in Events Processed, and others differ again. One currency with several exchange rates makes mixed workloads genuinely hard to forecast, which is why the usage dashboard matters.

Data Integration Pricing: iPaaS and ETL Cost

How should you compare data integration pricing?

Compare the meters, not the prices. Integration platforms sell five different units, and the unit decides the invoice far more than the rate does. Zapier and Workato bill per successful action. Make bills one credit per module run per record. n8n bills one execution per workflow run no matter how many steps or rows are inside it. Fivetran bills each distinct row once per month and Airbyte bills the volume moved. Celigo and Adapters bill for the connections you set up and not for what flows through them. Run one real job through all five rules and the answers differ by roughly three orders of magnitude, which is why a headline price comparison is close to useless.

The practical rule: match the meter to the shape of the work. Per-action pricing is fair for a few hundred small automations and brutal for a nightly orders sync, because it turns every row into a billable event. Row-based pricing is the opposite, cheap for chatty updates to the same records and expensive for full refreshes. If you are pricing a specific vendor we keep separate write-ups of Fivetran pricing, Airbyte pricing, Informatica pricing and n8n pricing, and the shortlist itself is on best data integration tools.

What each platform actually counts

Billing units and counting rules, taken from each vendor's own documentation on 17 September 2026. We print no third-party dollar figures here, because rates change quarterly and the counting rule is the part that decides your invoice anyway. The column nobody publishes is the last one.

Nine data integration platforms compared by billing unit, what increments the meter, what is not charged, and what happens at the plan ceiling
Platform	Billing unit	What ticks the meter	What is free	At the ceiling
Adapters	Flat monthly plan	Nothing. The price is the price whatever the row count	Rows, tasks, steps, retries, backfills, connectors	No meter to exceed
Fivetran	Monthly Active Rows	Each distinct primary key inserted, updated or deleted in the month	The same key re-synced later that month, initial syncs, new columns, system tables	Usage keeps billing
Airbyte Cloud	Credits against data volume	Rows for API sources, observed data volume for databases and files	Nothing on a full refresh. Every row synced is charged	Syncs continue, billed later
Zapier	Task, meaning one successful action	Each successful action step, replays, and both ends of a sub-Zap call	Triggers, Filter, Paths, Formatter, Delay, Looping, Digest, errored or skipped steps	New runs held until reset
Make	Credit, which replaced the operation	One credit per module run per bundle, so 500 records through a module is 500	Trigger modules, which run once however many bundles they return	Usage keeps billing
n8n	Production execution	One per workflow run, per webhook request, per schedule fire	Step count, data size, sub-workflows, manual runs, error workflows, empty polls	Usage keeps billing
Workato	Task, meaning one successful connector action	Each successful action, plus the call and the child recipe actions, plus reruns	Failed actions, skipped conditional branches, control statements	Usage keeps billing
Celigo	Endpoints plus flows	Each distinct endpoint invoked and each flow run in an integration app	Transaction volume, instances, users, data loader flows	No consumption meter
Informatica	IPU, pre-paid annually	A different scalar per service: Compute Units, Events Processed and others	Nothing. Every service draws on the same pre-paid pool	The pool drains

Two rows in that last column are worth reading twice, because they are opposite failure modes. Zapier documents that without pay-per-task enabled it will hold new Zap runs until your billing cycle resets, so the sync stops and the business notices. Airbyte documents that it continues syncs as normal, does not pause scheduled ones, and bills in arrears, so the sync never stops and the finance team notices. Neither is a trap, both are published, and almost nobody checks which one they bought. Honest write-ups of the individual platforms are at Zapier, Make, Workato, Celigo and Fivetran.

One job, eight meters, answers three orders of magnitude apart

The workload: a scheduled job that reads 1,000 order records from a source and writes each of them to two destinations. Nothing exotic, and close to the most common thing anyone builds on an integration platform. The figures below are not vendor quotes. They are the arithmetic each vendor's published counting rule produces on that one job, which is the calculation a buyer has to do and which no pricing page does for you.

The same 1,000-record two-destination job counted under each platform's published billing rule
Platform	Billable units for that one run	Why, per the documented rule
n8n	1 execution	A run is a run. The docs are explicit that step count and data size do not change it
Celigo	No incremental charge	You bought the endpoints and the flow. Volume through them is not metered
Adapters	No incremental charge	Flat plan. The thousand rows and the two destinations are already paid for
Fivetran	Up to 1,000 MAR in month one	Distinct keys, counted once per month. Re-running the job that month adds nothing
Airbyte Cloud	Volume of all rows moved	Incremental sends only changed rows, but a full refresh charges every row again
Zapier	About 2,000 tasks	Two successful actions per record. The trigger and any filters are free
Make	About 2,000 credits	One credit per module run per bundle, and there are two write modules
Workato	About 2,000 tasks	Each successful connector action counts, and a called recipe bills its own actions too

Now run that job every night for a month. On n8n it is about 30 executions. On the task-metered platforms it is around 60,000 billable actions. On Fivetran it is closer to the number of orders that actually changed, because a distinct primary key synced repeatedly in one month still counts once. That single rule is why chatty syncs of slowly-changing records are cheap on a row meter and ruinous on a task meter, and it is the most useful thing a buyer can know before signing anything.

The work you are not billed for, which changes how you build

Every counting rule has holes in it, and the holes are documented. Knowing them changes how you structure a workflow, and on a metered plan that is the difference between a sensible invoice and a bad one.

Logic is free on Zapier

Filters, Paths and Formatter cost nothing

Triggers never use tasks, and neither do Filter, Paths, Formatter, Delay, Looping, Digest or Storage steps. Nor do actions that error, halt or never run. So filtering early, before the action steps, genuinely reduces the bill rather than just tidying the Zap.

Structure is free on n8n

Sub-workflows do not bill separately

Only the parent execution counts when a workflow calls another one, manual runs while you build are not counted, error workflows are not counted, and a polling node that finds nothing does not count at all. Splitting a large workflow up costs nothing.

Repetition is free on Fivetran

The same row counts once a month

A distinct primary key synced many times in a calendar month is a single Monthly Active Row. Initial syncs are free, adding a new column to an existing table does not count, and system tables are excluded. Syncing more often does not raise the bill.

Failure is free on Workato

Failed and skipped actions are not counted

Only successfully run actions become tasks. Skipped conditional branches and control statements are excluded. The counterweight is that reruns are added to usage, and a called recipe bills both the call and every action inside the child job.

The asymmetry worth noticing is that per-record meters punish the thing per-run meters ignore. Make counts one credit per module run per bundle, so a module handling 500 records is 500 credits, while the same 500 records through n8n is still one execution. If you are choosing between those two models, the deciding number is how many records a typical run touches, not how many runs you have.

Six figures worth writing into the business case

1 vs 2,000

Billing events for the same 1,000-record job, per published counting rules, read 17 Sep 2026

$5

Fivetran base charge per connection with monthly usage between 1 and 1M MAR

27 Aug 2025

Date Make replaced the operation with the credit as its billing unit, at one-to-one parity

In arrears

How Airbyte bills past your credits. Syncs are not paused and not stopped

Held

What Zapier does to new runs at the task limit without pay-per-task enabled

$49

Our Starter plan per month, flat, with no row, task or execution meter behind it

One of those dates is easy to miss. Make replaced the operation with the credit as its billing unit on 27 August 2025, converting existing balances at parity so a thousand operations became a thousand credits. Standard features still consume one credit per operation, but AI modules and some advanced tools consume a variable amount depending on processing complexity, which means a scenario that was perfectly predictable under the old unit may not be under the new one.

How to price your own workload in five steps

Step 01 / Count the records

Not the ones you have, the ones that change in your busiest month. Row-metered platforms bill the change, not the table size, and the peak month is what the budget has to survive.

Step 02 / Count the write steps

Every destination you write to is a separate billable action on task-metered platforms. A job that updates a CRM and a ledger is two, not one, and a called sub-recipe bills its own actions on top.

Step 03 / Count the re-runs

Replays, backfills and full refreshes are where estimates break. Zapier re-counts replayed steps, Workato adds rerun jobs to usage, and an Airbyte full refresh charges every row again.

Step 04 / Apply each published rule

Take those three numbers and run them through each vendor's own counting rule rather than its headline price. The spread between the cheapest and the dearest answer is routinely an order of magnitude.

Step 05 / Ask what happens at the ceiling

Get it in writing. Held runs and uncapped arrears billing are opposite risks, and which one you can tolerate depends on whether the sync is keeping your business running or filling a report.

Which meter suits which team

Finance teams syncing orders to the ledger

High row counts, low complexity, runs every night forever. This is the workload that per-task billing punishes hardest, because each order line becomes a billable event and the volume only ever grows.

RevOps keeping CRM and billing in agreement

Moderate volume, many fields, constant small updates. Row-metered platforms suit this well because a record updated twenty times in a month is counted once, so chatty syncs are cheap.

Ecommerce operators across several storefronts

Volume spikes hard in the fourth quarter. Usage meters bill the spike at exactly the moment cash is tight, which is the argument for buying a flat plan before the season rather than after it.

Engineering teams replacing cron scripts

The scripts were free, so any meter feels expensive. Execution-based and flat models map most naturally onto what the scripts were doing, and per-record billing feels like a tax on the same work.

Ops teams running a few hundred automations

Genuinely low volume across many small workflows. This is where per-task pricing is fair and where a flat platform plan is usually the more expensive choice, which we would rather say than pretend otherwise.

Regulated teams needing audit evidence

The meter matters less than the lineage and the retention. Enterprise suites price this as a pre-paid pool across services, and the forecasting difficulty is the cost of the governance you are buying.

Where a flat plan is the wrong answer

A pricing comparison written by a vendor is worth reading only if it admits where the vendor loses. Five cases where you should buy something else.

If you run a few hundred tasks a month, a metered plan on Zapier or Make will cost you less than our Starter plan. Low-volume automation is not the workload we are priced for.
We publish four prices and do not negotiate them. If your procurement process expects a bespoke annual agreement with a discount schedule, an enterprise suite is a better cultural fit.
We do not sell API management, master data management or data governance tooling. If the requirement is a regulator-facing lineage record, Informatica and Talend do things we do not.
Our connector catalogue is smaller than the open-source catalogues. If you need a long tail of niche SaaS sources, Airbyte has far more of them, with the variability that a community catalogue implies.
A flat plan removes the incentive to optimise. Teams on usage pricing tend to tune their syncs because the bill rewards it. Ours will not tell you that a nightly job could run weekly.

Four questions to put to every vendor in writing

Question 01

What exactly counts as one billable unit, in writing?

Ask for the documentation page, not the sales answer. Every vendor on this page publishes the rule, and the rules differ enough that the same job can be one event or two thousand.

Question 02

Does a retry cost me again?

This is the question that separates the quoted estimate from the real invoice. Retries, replays and full refreshes are billable on several platforms and free on others, and failures are not rare.

Question 03

What happens the month I double?

Growth is the normal case, not the pessimistic one. Ask what the invoice looks like at twice your current volume, then at five times, and compare that to a flat plan at the same points.

Question 04

Do you stop my syncs or send me a bill?

Both answers exist in this market and neither is wrong, but you have to know which one you bought. Held runs break operations. Uncapped arrears billing breaks budgets.

Priced by platform, and priced by project

This page prices the ongoing case, where two systems stay in agreement forever and a meter runs the whole time. The one-off case is different arithmetic, because the dominant cost there is people time rather than platform fees, and that is covered in what a data migration really costs alongside the tooling on data migration tools.

Questions buyers ask about data integration pricing

How much does data integration cost?: There is no single answer because the platforms do not sell the same unit. Flat plans on this site start at $49 a month. Task-metered and row-metered platforms depend entirely on your volume, and the same workload can bill as one event or two thousand depending on which meter you bought. Price your own workload against each vendor's published counting rule before you compare any headline number.
What are the main iPaaS pricing models?: Five, and they behave differently under growth. Per successful action, used by Zapier and Workato. Per module run per record, used by Make. Per workflow run regardless of size, used by n8n. Per row or per volume moved, used by Fivetran and Airbyte. And flat entitlement pricing, used by Celigo and by us, where you buy connections rather than consumption. Enterprise suites such as Informatica sit on top with pre-paid credit pools.
Why is my integration bill higher than I expected?: Usually because the meter counts something you were not thinking about. Replayed runs count again on Zapier. A called sub-recipe bills its own actions on Workato. A full refresh charges every row again on Airbyte even though nothing changed. A schedule trigger on n8n counts every time it fires whether or not there was anything to do. None of these are hidden fees. They are documented counting rules that buyers rarely read before signing.
What happens when you exceed your plan limit?: The platforms split into two camps and this is the single most important question to ask. Zapier holds new runs until your billing cycle resets unless you have turned on pay-per-task. Airbyte does the opposite: it does not stop in-progress syncs and does not pause scheduled ones, and bills you in arrears. One camp risks stopping your business, the other risks an invoice you did not approve.
Is usage-based pricing cheaper than flat pricing?: At low volume yes, and it inverts as you grow. Usage pricing is attractive in the pilot, when you are moving a few thousand records and paying almost nothing. The crossover comes when a normal business year doubles your row count and the bill doubles with it, which is why teams describe it as a success tax. Flat pricing costs more on day one and stops changing after that.
What is Monthly Active Rows in Fivetran pricing?: MAR is the number of distinct rows synced from source to destination in a calendar month, tracked by primary key. The rule that surprises people is favourable: a row synced a hundred times in one month counts once. Initial syncs, new columns on existing tables and Fivetran system tables are not charged. The rule that hurts is that a re-sync counts any rows that genuinely changed.
Does a filter step cost money?: It depends on the platform and the difference is large at volume. Zapier does not charge for Filter or Paths steps, and does not charge for actions that error or never run. Workato does not count skipped conditional branches or control statements. Make counts one credit per module run per bundle, so how you structure a scenario changes the bill directly. n8n does not care at all, because the whole run is one execution.
How do I estimate integration costs before buying?: Write down one real job and count it four ways. Take the number of records it moves, the number of write steps it performs, how often it runs, and how often it will be replayed or re-synced. Then apply each vendor's counting rule to those four numbers. Do it for your busiest month rather than an average month, because the meters bill the peak and the budget has to survive it.
Is per-task pricing bad?: It is honest for low-volume automation and punishing for data movement. A few hundred tasks a month moving records between business apps is cheap and fair. The same model applied to a nightly sync of an orders table turns every row into a billable event, which is how teams end up paying four figures a month to keep two systems in agreement. Match the meter to the shape of the work.
What is an IPU in Informatica pricing?: An IPU is a pre-paid unit of credit you buy annually and then draw down across Informatica services. The complication is that each service consumes on its own scalar: Data Integration usage is measured in Compute Units, Cloud Integration Hub in Events Processed, and others differ again. One currency with several exchange rates makes mixed workloads genuinely hard to forecast, which is why the usage dashboard matters.

If you have narrowed it to two usage-metered platforms, the head-to-head that matters most at volume is in Fivetran vs Airbyte pricing for high-volume syncs, because their two meters behave in opposite directions on the same table. For the difference between paying for movement and paying for transformation, see ETL vs ELT.

Data integration pricing compared: iPaaS pricing models, ETL pricing and what data integration really costs