Sla direct naar inhoud
Implementation

What Does an AI Workflow Cost? Costs and Returns

By Mike van BeaumondAI Engineer · company brain & automationpublished on

6 min read
What Does an AI Workflow Cost? Costs and Returns

The cost of an AI workflow has four parts: the one-off build, the platform it runs on, the AI usage itself and the maintenance afterwards. What the build costs depends on scope, not on a fixed package; the rest you can work out in advance, and so can what the workflow earns you.

Quick Take: An AI workflow costs money in four places: build (one-off), platform (for n8n and Make often a few tens of euros a month), AI usage (at light volumes usually a few euros a month) and maintenance (time). Whether it pays off, you work out per task with your own numbers: how often, how long and what an hour costs, minus the running costs.

The question I get most about AI workflows is not "can this be done", but "what does it cost, and what does it deliver?". The honest answer has two sides: the price depends on a few things you can check beforehand, and the return you can calculate with your own numbers before you sign a quote. For the wider context, also read building AI workflows: what to check first.

Which three things actually determine the price?

How many systems are involved

A workflow that stays within one system (for example, just your CRM) is cheaper than a workflow that needs to combine multiple systems (CRM, an external data source, a reporting tool). Every extra connection means extra work, extra error handling and therefore extra cost.

How much judgement is needed

A workflow that moves data according to a fixed rule is relatively simple. A workflow that needs to judge who is a good lead, which client example fits best or what should go into a report requires more thinking during the build and therefore more time.

How many exceptions need to be covered

A workflow that only handles the ideal scenario is faster to build than one that also knows what to do when a website cannot be read, a customer is not found or a step fails once. That error handling is often the least visible part of the work, and the part that takes the most time. So price it in from the start.

A quote that comes in very fast and very low without anyone ever looking at your systems or processes is a signal, not a bargain. A price should only be fixed after a scoping conversation, not before it. Always compare quotes on scope, not on the final number.

What costs does an AI workflow consist of?

Count on four cost items. Only the first is a one-off amount.

Cost itemOne-off or runningWhat it depends on
BuildOne-offNumber of systems, judgement needed, exceptions, data cleanup
Platform (n8n, Make)Per monthNumber of runs or credits, cloud or self-hosted
AI usage (Claude API)Per monthNumber of runs × tokens per run, chosen model
MaintenancePer monthChanges in your processes, failures, quality checks

Build

The build cost is the part that differs per project. I build workflows as an add-on to your brain; more on that further down. The fixed price is in the proposal after the intake. What you can do beforehand is work out what the workflow returns. You will find that further down.

Platform

The workflow has to run somewhere. The two platforms I come across most, with their price as listed on the pricing page:

  • n8n Cloud: Starter 20 per month (billed annually) for 2,500 workflow executions, Pro 50 per month (billed annually) for 10,000 executions, in the currency the page shows for your region (price as of October 2026, source: n8n.io/pricing). One execution is one run of the whole workflow, however many steps it has. Self-hosting is possible with the free Community Edition; then you pay no licence, but you do pay for hosting and maintenance.
  • Make: Core $9 per month and Pro $16 per month, both at 10,000 credits per month. Each action in a scenario generally costs 1 credit (price as of October 2026, source: make.com/en/pricing).

How these two compare to Zapier is covered in n8n vs Zapier vs Make. The platform price says little about total cost, by the way: at light volumes the AI usage is the smallest item.

AI usage

As soon as the workflow calls an AI model, you pay per token. For Claude Sonnet 5.5 that is $2 per million input tokens and $10 per million output tokens, and the Batch API gives 50 percent off both if the work does not need to be instant (price as of October 2026, source: platform.claude.com/docs/en/about-claude/pricing). A smaller model such as Haiku 5.5 costs $0.10 and $0.50 per million tokens (same source and date). The calculation: cost per run = (input tokens × input price) + (output tokens × output price), multiplied by the number of runs per month.

Maintenance

A workflow does not stand still. Your offer changes, an integration gets an update, a website the workflow reads changes its structure. So count on maintenance time, internal or external. Agree beforehand who receives the alerts when something goes wrong and who picks them up. This is the item that is forgotten most often.

What does an AI workflow return? How to calculate it

The same order applies to every workflow I discuss. It is a calculation method, not a promise: the outcome depends entirely on the numbers you fill in.

  1. Measure what the work costs now. Tasks per week × minutes per task = hours per week. Hours × hourly rate (salary including employer costs) = what it costs now. Use a measured average, not a feeling.
  2. Decide what share the workflow takes over. Do not calculate with the maximum from a brochure. An AI workflow usually delivers a first version that a person still reviews. Start cautiously and adjust once you measure.
  3. Count all running costs: platform, AI usage and maintenance, as above.
  4. Look at what you do with the freed-up time. Saving time is not automatically saving money. It only becomes value when the team does something useful with it. Leave that extra return out of your base calculation; mention it separately as a bonus.
  5. Allow for a ramp-up period. The first months go to getting used to it and adjusting. Count the return from the moment the workflow runs stably.

The formulas:

  • Value per month = tasks per week × minutes per task ÷ 60 × share taken over × hourly rate × weeks per month
  • Running costs per month = platform + AI usage + maintenance
  • Net value per month = value per month − running costs per month

How to fill in the sum yourself

I deliberately give no worked example with amounts here. The outcome depends on your numbers. What you need per task:

  • How often does the task occur, per week or per month?
  • How long does one time take now? Measure it, do not guess.
  • Who does it, and what does an hour of that person cost?
  • Which costs run: the build, the platform, the AI usage and the maintenance?

More on what a company brain delivers. This is a calculation with your own numbers, not a promise. An honest baseline matters more than any number from a brochure.

Where the calculation usually goes wrong

  • Using the best-case percentage. Anyone who enters "up to 80 percent time saved" gets a nice number that is not reached in practice. Start lower and let measurement adjust it.
  • Counting only the build. Platform, AI usage and above all maintenance belong in the sum.
  • No baseline. Without a measurement of how much time the work takes now, any ROI calculation is a guess.
  • Wanting to express everything in euros. Quality, fewer errors and faster turnaround are real, but hard to price. Mention them, but do not calculate with them.

When does it cost more than expected?

Two situations where the total cost often ends up higher than the first estimate: when the scope changes during the build (an extra system gets added, an exception turns out to be more common than expected), and when the existing data is messy, requiring cleanup before the workflow can work with it properly.

Both are preventable with a proper audit upfront, not by picking a low price and hoping it works out. A good scope describes the moment the workflow starts, the sources that go in and a check that rejects an empty or too-short result. Examples of my own work are in the cases.

Why a company brain makes the workflow cheaper to maintain

A workflow that has to know your business needs that knowledge: your offer, your ideal customer, your tone of voice. You can capture that in every workflow separately, but then you change several workflows with every update. One source is better.

A company brain is your business's knowledge source: vision, offer, customers, brand style and processes in one place, with an owner per topic. Your team and AI tools use it. If your offer changes, you change it in one place and all workflows read the new offer. That is a design choice that keeps maintenance simpler; how much it saves depends on your situation. If you are unsure whether you need a workflow or a custom application, read what it costs to develop an app with AI.

My take

The price of an AI workflow says less about how clever the workflow is, and more about how much reality it has to absorb. A workflow that can only handle the ideal scenario is almost always priced too low for what it will do for you in practice. So first work out what the workflow returns, then ask for a proposal based on scope.

From cost question to a workflow that knows your business

If you want to know what a workflow costs for your process, I start with an intake of your systems and your process. A proposal follows with a fixed price, stating what is and is not included. I build workflows as an add-on to your brain; the price is in your proposal. On AI workflow automation I explain how I approach it and what a workflow can use from your company brain.

Sound familiar? Let's talk.

In 30 minutes we look together at which team your brain starts with. No obligation.

Frequently asked questions

It depends on the number of systems, the judgement needed and the exceptions that have to be covered. I build workflows as an add-on to your brain. The fixed price is in the proposal after the intake. What a workflow returns, you work out per task with your own numbers.

Yes: the platform (n8n Cloud from 20 per month billed annually, currency depends on your region, price as of October 2026, source: n8n.io/pricing), the AI usage per token and maintenance time. At light volumes the AI usage is often only a few euros a month.

Value per month = tasks per week × minutes per task ÷ 60 × share taken over × hourly rate × weeks per month. Subtract the running costs. That gives you the net value per month. Start with a measured baseline and a cautious percentage.

Only if the scope is already fully clear. A serious builder runs a short intake first before naming a price, because the price depends heavily on the number of systems and the judgement required.

Often because the scope was interpreted differently. A low quote sometimes covers only the ideal scenario, without error handling, while a higher quote includes it. Always ask explicitly what is and is not included in the price.

No. Saved time only becomes money when the team does something useful with it. So in the base calculation count only the hours, and mention extra return, such as more leads handled, separately as a bonus.