Cheapest workflow automation tools: what they cost at real volume

Automation pricing is hard to compare because no two platforms bill for the same unit. Per task, per operation, per execution, per seat, per active workflow, per connector: each model makes a different set of workflows look cheap. A tool that costs almost nothing at signup can be the most expensive option you run by month six, and the reason is usually not a price rise but a change in what you are doing with it.

This page is about that arithmetic. It covers what the free tiers actually include, how each pricing model behaves as volume grows, and the cost categories that never appear on a pricing page. For picks based on capability rather than price, see the best AI workflow automation tools. For how the categories of tool differ, see workflow automation tools.

Prices below were checked in September 2026 and move frequently. Treat each vendor's own pricing page as the authority.

What we'll cover

• The five costs, only one of which is on the pricing page
• How each pricing model behaves as you grow
• What the free tiers actually give you
• Budget and standard managed options
• A worked comparison at volume
• The AI cost multiplier
• Picking the cheapest option for your situation
• Frequently asked questions

The five costs, only one of which is on the pricing page

Subscription. The number everyone compares, and the one least likely to decide the outcome.

Execution charges. Per-task or per-operation fees that scale with how much you actually run. On some models this dwarfs the subscription.

Engineering time. Setup, maintenance, debugging, and monitoring. On self-hosted tools this is the dominant cost, and it is the one that makes "free" misleading. Four hours a month of a developer's time at a fully loaded $75 an hour is $300 a month, which buys a great deal of managed platform.

The cost of failures. When an automation breaks and nobody notices for a week, what does that week cost? For an internal digest, very little. For customer invoicing or order routing, potentially a great deal more than any subscription you were choosing between.

Model costs. If your workflows call language models, that is usually billed separately from the automation platform, sometimes with a markup on top of the provider's own rate.

A self-hosted tool with a $0 licence and four monthly hours of maintenance costs more than a $50 managed platform that nobody has to touch. That comparison, rather than the sticker price, is what "cheapest" should mean.

How each pricing model behaves as you grow

Per task. Every action in every run is billable. A five-step workflow running a thousand times a month is five thousand tasks. This model is comfortable at low volume and becomes the most expensive of the lot once workflows get longer or run more often, because the cost scales with steps and runs multiplied together.

Per operation. Similar in shape, but operations are often counted more generously, particularly where a single module handles a batch of records rather than one record at a time. For data-heavy work this can be markedly cheaper than per-task billing on the same workload.

Per execution. One run of a workflow is one billable unit regardless of how many steps it contains. This makes complex workflows dramatically cheaper than either model above, and it rewards consolidating logic into fewer, larger automations.

Per seat. Priced on people rather than usage, which is predictable but penalises teams who want many colleagues involved even lightly. Watch for minimum seat counts, which quietly raise the entry price.

Bundled. A plan covering execution, model access, and integrations together, without per-step metering. Predictability is the benefit; the thing to check is what happens at the top of the plan.

The practical consequence is that the same workflow can differ by an order of magnitude in cost between models. Before comparing prices, count your steps and your realistic monthly runs, because those two numbers determine which model suits you far more than any headline rate does.

What the free tiers actually give you

n8n self-hosted. No licence fee. A basic VPS runs $5 to $20 a month, initial setup is a few hours, and ongoing maintenance is realistically one to two hours monthly. Cheapest on paper, and genuinely cheapest at high volume if you already have infrastructure capacity.

Windmill self-hosted. No licence fee, comparable hosting costs, with Python, TypeScript, and Go support. Suits teams who want to write real code.

Activepieces self-hosted. No licence fee and a simpler setup than n8n, with a smaller connector library to match.

Google Apps Script. Free with Google Workspace, JavaScript only, and confined to the Google ecosystem. For processes that live entirely in Sheets, Docs, and Gmail, it is hard to beat on cost.

Pipedream free tier. The most generous managed free allowance, and unusual in being genuinely production-viable rather than a demonstration.

Make free tier. A thousand operations monthly, which is enough to evaluate the product properly but not to run much on.

Zapier free tier. Single-step automations with a small monthly task allowance. Useful for trying the interface, not for running a business process.

Budget and standard managed options

IFTTT Pro, around $3 a month, covers simple consumer-style automations without branching.

Make Core, from about $9 a month for ten thousand operations, is the strongest value in managed visual automation and considerably cheaper than the equivalent Zapier tier.

n8n Cloud, from about $20 a month, removes the hosting and maintenance burden while keeping execution-based billing.

Pipedream Professional, around $19 a month, suits developers who want longer timeouts and code-level control.

Zapier Starter to Professional, roughly $20 to $50 a month, remains the easiest to learn and the most expensive per equivalent unit of work, which is a reasonable trade at low volume.

CodeWords bundles execution, model access, and more than 3,000 integrations without per-task or per-operation metering. The free plan covers light use, with Pro at $39 per month and Business at $100 per month as usage grows (pricing).

A worked comparison at volume

Take a five-step workflow running ten thousand times a month, which is a busy but unremarkable business process such as handling inbound orders.

On per-task billing that is fifty thousand billable units, which lands in the region of $150 a month on mid-tier plans. On per-operation billing the same work sits closer to $16 a month, because the counting is more generous and batching helps. On self-hosted execution billing the subscription is just the server, around $15 a month, with your maintenance time on top. On a bundled plan it is the plan price, unchanged by the step count.

Push that to a hundred thousand runs a month and the spread widens considerably. Per-task billing can pass $500 a month. Per-operation billing typically stays under $100. Self-hosted stays at server cost, though the operational burden grows with the importance of what you are running.

The lesson is not that one model always wins. It is that the winner changes as you grow, and that most teams choose while volume is low, which is precisely when the difference is least visible.

The AI cost multiplier

Once workflows call language models, a second meter starts running.

On platforms where you bring your own API keys, you pay the provider directly per call, typically fractions of a cent for short prompts and meaningfully more for long documents. That cost is proportional to how much text passes through, so a workflow summarising long reports costs far more per run than one classifying short messages.

Two things are worth checking. First, whether the platform adds a markup to the provider's rate, since some do and it is rarely prominent. Second, how many model calls a single run makes, because a workflow that loops over twenty records making one call each is twenty calls, not one.

Bundled platforms fold model access into the plan, which removes this category from the arithmetic entirely. For AI-heavy work with many calls per run, that difference tends to matter more than the subscription gap between platforms.

The costs that arrive later

Per-task pricing at scale. Covered above, and worth restating because it is the single most common budgeting mistake. Workflows get longer as they get better, and each new step multiplies against every run.

Migration. Moving platforms means rebuilding every automation. The more you have built, the higher the switching cost, which is why the platform you can grow into is usually cheaper over three years than the one that is cheapest today.

Tier jumps. Features like branching, error handling, and longer timeouts often sit behind a higher tier. Check which plan holds the capability you need rather than the plan you are quoted, because the entry price is frequently not the price you will pay.

Maintenance drift. Every connected system changes its API eventually. Whoever keeps up with that is a cost, whether it is your engineer on a self-hosted tool or the vendor on a managed one.

Picking the cheapest option for your situation

• Under a hundred runs a month, non-technical: a free managed tier is genuinely sufficient.
• Hundreds to a few thousand runs: Make Core or the Pipedream free tier, depending on whether you prefer a visual canvas or code.
• Thousands to tens of thousands: per-operation or per-execution billing, so Make Pro or n8n, self-hosted if you have the capacity and cloud if you do not.
• High volume with heavy model use: a bundled plan, or self-hosted n8n with your own API keys if you have engineering time to spend.
• Any volume where nobody can maintain infrastructure: a managed platform, and count the engineering hours you are avoiding as part of the saving.

Frequently asked questions

Is the cheapest tool always the right choice?

No, and the arithmetic usually says so plainly. A platform costing $50 a month that saves ten hours of someone's time is far better value than a free one saving two. Work out cost per hour returned rather than cost per month, and the ranking often inverts.

How do I estimate my execution volume?

Count how often the trigger fires in a normal day and multiply by thirty. Then multiply by the number of steps if you are pricing a per-task model. Do this for your busiest month rather than your quietest, since that is the bill you need to survive.

Are free tiers usable in production?

Some are. The most generous managed free tiers handle real workloads, and self-hosted tools have no licence ceiling at all. The restricted free tiers on the mainstream connectors are built for evaluation, and single-step limits rule out most genuine processes.

What is the cheapest option with AI capability?

Self-hosted with your own API keys is cheapest on subscription and most expensive in engineering time. A bundled plan costs more on paper and removes both the separate model bill and the maintenance. Which is cheaper depends on whether the hours or the subscription is the scarcer resource for you.

Does self-hosting actually save money?

At high volume with existing infrastructure capacity, yes, substantially. At low volume without it, rarely, because a few hours of engineering time each month exceeds most managed subscriptions. The honest test is whether someone already runs servers as part of their job.

How much should a small team budget?

Most small teams running a handful of processes land somewhere between $0 and $50 a month, and the figure is dominated by which pricing model they happen to be on rather than by how much work they are automating. Getting the model right early is worth more than negotiating the rate.

Why did my automation bill go up when I didn't change anything?

Almost always because volume rose rather than prices did. Automations get pointed at more sources as people trust them, and a workflow that was handling forty records a day quietly starts handling four hundred. On per-task billing that is a tenfold increase with no change you would remember making. Setting a usage alert at roughly twice your expected volume turns this from a surprise into a notification.

Should I consolidate several small automations into one?

On per-execution or bundled billing, yes, and the saving can be large, because ten separate workflows firing on the same trigger cost ten runs where one branching workflow costs one. On per-task billing it makes little difference, since the steps are counted either way. Consolidation also tends to make failures easier to trace, which is worth something regardless of the model.

Related reading

• Workflow automation tools: how the categories differ
• Best AI workflow automation tools
• Zapier vs Make vs n8n
• CodeWords pricing and templates