Cost to automate a process: the formula, not the range

Search for what it costs to automate a process and you will find the same thing everywhere: ranges from €1,500 to €60,000, a list of factors that move the price, and a closing line that sums it all up. The exact figure comes out of a prior audit. That is true. And it is no use at all when what you need is to know whether it is worth it before booking a call with anyone. This article takes another route: what drives the cost of process automation, the options from no-code tools to AI, the recurring charges quotes often leave out, and a formula with three numbers you already know.

How much does it cost to automate a process?

There is no fixed price. The cost to automate a business process varies with its complexity, the number of systems it connects, how often it runs, the quality of the data and whether AI is involved. The figure that makes any proposal meaningful is a different one: what the process costs you today, done by hand.

Put the question to a provider and the first answer is often «it depends». That is true, and it also suits them. It is true: the price depends on your systems, your volumes and how many exceptions have to be handled. And it suits them: a range from €1,500 to €15,000 commits nobody, and forces you to ask for a quote. The problem is that the range is so wide it does not help you decide. Between €1,500 and €15,000 there is a factor of ten.

What the range is missing is one figure: what doing the process by hand costs you. With that figure, a quote becomes comparable.

What drives the cost of process automation?

Five factors drive the cost more than anything else: complexity and scope, the systems and integrations involved, exception handling, volume and data quality. Two business processes that cost you the same today can call for very different builds.

Complexity and scope

Every step, decision and branch in a workflow is logic that someone has to build and test. Moving form entries into a spreadsheet is a simple workflow; a business process with approvals, rules that vary by client and several outputs is a complex one. Before you approach anyone, write down which steps will be automated, which stay manual and what counts as finished: a loose scope is a common reason for a budget to grow halfway through a project.

Systems, integrations and APIs

Connecting two apps that already offer a ready-made integration is quick. Each additional system adds an integration to the workflow, and each one relies on the other app keeping a stable API, the interface through which programs exchange information. Some business tools have no public API, or only open it on higher plans, which forces workarounds such as reading exported files or emails. The more third-party pieces, the more points to watch.

Exceptions

The standard case is rarely the expensive part; edge cases are. The invoice with two VAT rates, the client who pays half, the order cancelled after it ships: each exception calls for a rule in the workflow, an AI step or a clear way to pass the item to a person. List the exceptions you deal with in a normal month: that list says more about the final bill than the number of steps.

Volume

Volume works in both directions. The more often a process runs, the more time automation saves, so the value of automating rises. But on platforms with usage-based pricing, the bill rises too.

Data quality

An automation is only as good as the data it receives. Duplicate records, scanned documents, formats that change from one supplier to the next and fields typed in manually all have to be cleaned up first, and that clean-up is a cost too. Where the data sits often decides a project more than its complexity.

No-code, custom development or RPA: options and pricing models

There are three broad routes for business processes, each with its own pricing model. AI is not a fourth route: it is a layer that can be added to any of them, and choosing the right tool comes down to the apps you use, how often the process runs and who will look after it.

No-code automation platforms

Tools such as Zapier, Make or n8n connect the apps you already use through a visual editor: an event in one app, such as a new form entry or an incoming email, starts a chain of steps in others, and many now offer AI steps too. The entry cost is low, often with a free plan to start. Pricing usually grows with usage, counted in tasks, operations or runs each month, so a workflow that is cheap at first can become costly as it scales. n8n can also be self-hosted, which swaps the subscription for hosting and updates. If you are building the workflows yourself, the cost is your time: see our article on no-code automation for small businesses.

Custom development

A developer writes code that connects your systems directly. The build costs more, because everything is made to measure, but there are fewer subscriptions, and the code can handle workflow logic that visual tools struggle with. Hosting, monitoring and support remain. The main risk is depending on a single person: make sure the code, the logins and documentation of the logic are handed over.

RPA: software robots for systems without an API

Robotic process automation (RPA) uses software robots that click and type in an application's screens the way a person would. It helps with an older system that nothing else can connect to, but it is fragile: when a screen changes, the robot stops. RPA tools are often designed for enterprise teams and can involve licences, which can make them hard to justify for small businesses.

Attended and unattended robots, and process mining

IBM, among other vendors, describes two main types of RPA:

• Attended RPA: the robot sits on an employee's desktop and helps that person on demand, for example copying details from forms into a legacy screen while they deal with a client.
• Unattended RPA: robots operate on their own, often on a virtual machine, and perform steps automatically on a schedule or when a file arrives, with no one at the keyboard.

The difference matters for cost, because many enterprise RPA solutions charge differently for each type: per person for attended robots, per robot or per machine for unattended execution. Larger vendors also offer process mining and task mining, tools that analyse system logs and on-screen activity to discover which processes are worth automating. These are useful for a large organisation with multiple departments; a small business can learn the same thing by timing the job for a week, as described below.

What tool plans count, and what to check before you choose

Each platform measures consumption in its own unit, so compare plans by what they count rather than by the headline figure:

• Zapier counts tasks: each successful action step in a Zap uses one.
• Make counts credits, previously called operations: each module that runs in a scenario consumes them.
• Microsoft Power Automate is licensed mainly per person or per unattended robot: cloud flows connect online services, while desktop flows add RPA for desktop applications, attended or unattended.

No-cost plans are available on several of these platforms, with caps on executions, users or features, so you can explore and try a flow before you purchase a paid tier. Before you commit, confirm on each vendor's official plans, at the time you buy, as terms change:

• how many credits or executions each tier includes per month, and what happens when you reach the maximum;
• whether data storage, file size caps in MB or GB, and advanced or premium connectors cost extra;
• whether every person who creates or manages flows needs their own account;
• whether annual billing brings discounts, and whether unused credits roll over or are lost.

Here is how the options compare:

What does AI automation cost?

AI automation is usually more expensive to operate than rule-based automation: on top of the build and the subscriptions, AI is typically billed by usage, and its output calls for testing and human review. It earns that extra cost when a process involves text that fixed rules cannot handle.

AI vs fixed rules: when a process actually needs AI

Fixed rules work when the input is predictable: a form with the same fields, a file with the same columns. AI earns its place when the input varies: an email written in countless ways, delivery notes in each supplier's layout, requests to sort by topic. Many processes combine both, with AI reading and extracting the details and the workflow's rules deciding what happens next. If yours follows clear rules, start without AI: it will be cheaper and more stable.

AI agents: more autonomous, more to supervise

AI agents go a step further than an AI step inside a flow: they receive a goal, decide the next steps and can chain multiple tools on their own, from reading a chat to updating a sales or marketing record. Automation platforms increasingly offer built-in agent features. That autonomy can help with varied requests, but each extra decision the agent takes is one more point to review, so start with a narrow, well-defined job and keep a person in the loop for anything that reaches a client.

How AI pricing works, and what else it adds

AI models are usually billed by the amount of text they process, so each email or document read adds a small cost. With a few documents a day that line stays modest; with thousands, it becomes a monthly cost to calculate before you start. AI also interprets rather than follows rules, so it can get things wrong. Budget for trials on real examples, for a human review of anything a client will see, and for care with personal data: under the GDPR, which applies in Spain, you remain responsible for the data you send to an AI service.

Build, subscriptions and upkeep: the three parts of the bill

The total automation cost has three parts: the one-off build, the monthly running costs and the ongoing upkeep. Many quotes show only the first, so ask for all three in writing.

The one-off build

The build covers mapping the process, designing the workflow, building the integrations, the logic and any AI steps, testing end to end with real examples and going live. Providers charge for it at an hourly rate, as a fixed project or in relation to what the process costs you; whatever the pricing model, the scope sets the figure.

Monthly running costs

Once live, an automation keeps costing money every month: the automation platform subscription, any plan upgrade a connected app requires before it allows integrations, AI usage and, if the platform is self-hosted, hosting. These fees rise with usage, so calculate them at your real volume, and at double it.

Upkeep and what quotes leave out

Connected apps change: an app updates its interface, a supplier redesigns its invoices, a form gains a field. Someone has to notice and adjust the workflows, and monitoring, with an alert when a workflow run fails, keeps an unnoticed error from costing more than the automation saves. We look at this in our article on maintenance, the hidden cost of automation. Other items rarely appear on a first quote:

• Testing: trying the automation on real examples, odd ones included, before you trust it.
• Data preparation: removing duplicates, standardising formats, sorting out paper files.
• Training: your team must understand what the automation does, what to do when it fails and whom to tell.
• Your own time: explaining the process and checking results in the first weeks.
• Human review: if AI is involved, someone checks its output, at least at the start.

Our price guide, published

As a guide, we price the build at around 12 % of the annual cost of the task being automated. If a task costs you €20,000 a year, the project would come to around €2,400; if it costs €4,000, around €500. The figure starts from what the job costs you today; the fixed quote, sent within 48 hours of the video call, takes your systems and edge cases into account, and it is what you pay. Maintenance comes to around 20 % of that build a year, and tool subscriptions are extra.

Whatever the method behind a quote, you can hold it up against the same calculation, using your own figures:

• Annual cost of the process = times a day × hours each time × hourly cost × 240 working days.
• Real saving = that annual cost × the share automation actually recovers.
• Months to pay back = build cost ÷ (monthly saving − monthly running costs).

If the result is a few months, the automation makes sense; if it runs to several years, look again at the process you chose before blaming the provider.

How to estimate the current cost of a process

Everything starts with three figures, and you have all of them:

• Frequency: how many times a day it is done. If it happens weekly, convert it to days, based on a normal month rather than a closing day.
• Time: how long it takes to complete, in minutes or hours, from start to end, including checks and back-and-forth emails.
• Hourly cost: the gross salary of whoever does it plus the company's Social Security contributions, divided by the hours worked. If you run the business alone, count what you would bill in that time.

Multiply the three by 240 working days and you have the annual cost. No spreadsheets, no odd formulas. Measure for a week rather than relying on memory: a simple log avoids errors.

A worked example, with the numbers on show

Imagine a small company in Elche that keys in supplier delivery notes twice a day, twenty minutes each time, with someone costing €25 an hour fully loaded. That is 160 hours and €4,000 annually. The build comes to around €480 and maintenance to around €96 annually. Now for the other half of the calculation: how much of those €4,000 automation actually recovers.

You almost never recover 100 %

This is where many payback calculations go wrong. No automation removes the whole job: exceptions, checks and items that do not fit all remain. That residual work has to be deducted before counting the saving, and what you really recover depends on the data: kept in a business system, with operations that resemble one another, the recoverable share is far larger than on paper, with every case different. That difference decides the entire project.

Two scenarios, two paybacks

Suppose the delivery notes arrive as digital files and the business software accepts data through an API. In that case, a good share of the time comes back and the build pays for itself within a few months.

Now suppose they arrive on paper, each supplier with its own layout, and someone has to verify every line. Even with AI reading the scans, only a small share is recovered, and the same investment takes several times longer to come back. Same process, same build cost, a very different payback.

Return on investment: when is automation worth it?

Not all processes are worth automating. For small businesses, automation is worth it when the real saving, net of monthly charges, repays the build within a reasonable number of months: the payback formula above gives you that number.

That payback period says more than any return-on-investment percentage: when it is very long, a single change in a tool's pricing or in your process is enough for the investment never to come back.

When the answer is no

We only propose a project when it pays back in under 24 months. Below that threshold there is a case; above it, there is not. In practice that rules out tasks of a few minutes a day, those whose rules change every quarter, and those whose data sits on paper without having been sorted out first. Not because they are impossible: because the cost of keeping them running eats the saving.

There is a fourth group: work where the personal touch is what the client values, such as answering a complaint. The first conversation with a hesitant prospect is better not left to an AI chatbot. Before deciding, read our guide to which repetitive tasks to automate first and our article on what you should never automate.

Calculation mistakes that skew the decision

The formula is simple, but it is easy to fool yourself when filling it in. Watch out for these points:

• Counting all the time as saved. Something almost always remains to be checked by hand.
• Forgetting the cost of keeping it going. An automation that saves time but needs almost as much to stay alive saves far less than it seems.
• Counting hours that turn into nothing. If you don't spend them billing, selling or serving customers, the saving only exists on paper.
• Taking an exceptional period as the reference. Use a normal month, not a closing or a campaign.

How to budget and compare automation quotes

Compare quotes on the total cost and the payback, not on the build alone. Whoever you approach, a freelancer, an agency or a software vendor, ask for the same things:

• a fixed price or a published pricing method, rather than an open-ended figure;
• a written scope: which steps are automated and which stay manual;
• what the build includes: testing, documentation and training;
• the monthly running costs, subscriptions and AI usage included, and who pays them;
• support and monitoring terms, set out separately;
• if the solution uses AI, how errors are caught and who reviews them;
• a projection of the saving based on your own figures.

Then start small: automate one workflow that matters to your business, measure the result for a month and only then decide on the next. To see the whole journey, read our guide to how automating a process works, step by step; to avoid the classic traps, see the costliest automation mistakes; and if you are weighing a new hire against automation, see hire someone, or automate?

You can run these numbers yourself with a calculator, or let a free diagnostic do it: four questions and about two minutes give you the annual cost of your task; six more give you how much of it is recoverable, the minimum technology required and how many months it takes to pay back. No appointment, no talking to anyone. And if it turns out not to be worth it, it tells you that too.