How Much Does AI Automation Cost in 2026?

Most quotes for AI automation hide three different costs inside one number. This article separates them so you can compare offers on equal terms and estimate what a workflow will really cost to run.

The three costs behind every automation

Whatever the provider calls it, an automation project has three separate cost lines:

• Implementation — the one-time work of mapping the process, designing the workflow, building it, testing it and handing it over.
• Running costs — the automation platform, AI model usage, hosting and any telephony or OCR services the workflow uses each month.
• Care — monitoring, error handling, regression tests after model updates and small changes as your process evolves.

A quote that shows only the first line is incomplete. A quote that bundles the second line into the first is hiding a margin you cannot see. Ask for all three, itemised.

What drives the implementation price

The word “automation” covers everything from a two-step form-to-CRM sync to a document pipeline with OCR, approval routing and accounting integration. Price follows complexity, and complexity is driven by a handful of factors:

• Number of connected systems. Each system means authentication, field mapping, error cases and testing.
• Branching and approvals. A workflow that routes cases to people and waits for a decision needs state, reminders and timeouts.
• AI steps and their variability. Classifying a clean form is cheap. Extracting fields from supplier invoices in twelve layouts is not.
• Custom APIs and legacy software. Anything without an established connector needs development and more maintenance later.
• Sensitive data. Personal, financial or health data adds design work for minimisation, access control and review.
• Volume. High throughput changes platform choice, cost caps and how failures are handled.

We price in automation credits so that “ten automations” cannot quietly become ten enterprise systems. A simple workflow (one trigger, up to three routine actions, one or two apps) is one credit. A standard workflow with branching, an approval step and one controlled AI step is two. Advanced workflows with custom APIs, knowledge retrieval, OCR variability or sensitive data are four to eight or more. The final classification is confirmed in discovery, before you pay for a build.

Our implementation ladder

These are our published prices. All exclude VAT, and third-party platform, model and hosting fees are paid by you in your own accounts.

Package	Price (excl. VAT)	Typical fit	Timeline
One Workflow Pilot	from €1,250	One clear process, up to two apps, one AI step	10–15 business days
Solo	€2,750	Five credits, one department, up to four apps	2–3 weeks
Team	€6,900	Twelve credits, approval routing, monitoring view, training	3–6 weeks
Business	from €16,500	Up to 30 credits, custom APIs, test environment, security review	6–12 weeks
Enterprise	discovery from €5,000 / implementation from €40,000	Architecture, governance, self-hosted options, staged rollout	10–24+ weeks

Optional care plans start at €290/month (excl. VAT) for up to five production workflows and include monitoring, a monthly health check, error alerts and a small allowance of minor changes. Full details are on the pricing page.

Typical third-party running costs

Running costs depend on the platform and how much the workflow does. Published pricing signals as of the sources listed below:

Platform	Pricing model	Public signal
n8n	Execution-based; self-hosting possible	Hosted Starter from €20/month, Pro from €50/month (billed annually)
Make	Credit (operation) based	Free tier through Enterprise plans
Zapier	Task-based	Free from 100 tasks/month; paid tiers by task volume
Microsoft Power Automate	Per user or per bot	Premium $15/user/month; Process $150/bot/month

Add AI model usage on top. For most SME workflows — classification, extraction, drafting — model costs are small compared with platform fees, but they scale with volume and with the length of documents you process. We estimate running costs in every proposal and set cost caps at launch so that a runaway loop cannot surprise you.

What makes a quote go up

If two quotes for “the same” workflow differ by a factor of three, one of these is usually the reason:

• The process was never mapped, so one provider priced the happy path and the other priced the exceptions.
• One quote includes testing on edge, failure and privacy scenarios; the other assumes clean inputs.
• One includes documentation, a runbook and recorded handover; the other ends at “it works on my screen”.
• One builds in accounts you own; the other builds on the provider’s subscription and charges you to leave.

How to compare quotes

Put every offer through the same questions before you compare the totals:

1. What exactly is included — how many systems, which AI steps, which approval points?
2. What are the written acceptance criteria, and who decides when “done” is reached?
3. Who owns the accounts, workflows, prompts and documentation afterwards?
4. What are the estimated monthly running costs, and are they paid by me directly?
5. What happens after launch — how long is support, and what does care cost?
6. What triggers a change order, and how is it priced?

Hidden costs most quotes leave out

Maintenance. APIs change their authentication, fields get renamed, and connectors are deprecated. A workflow that nobody monitors will eventually fail silently.

Model updates. AI providers retire and replace models. A prompt tuned for one model can behave differently on its successor. Regression tests after updates are part of the real cost of an AI step.

Exception handling. The 5% of cases the workflow cannot handle still need a person, a queue and a process. If that is not designed, the savings on the other 95% erode.

Adoption. A workflow that the team routes around because they do not trust it costs the full implementation fee and saves nothing. Training and a gradual rollout are cheaper than a second build.

A payback example

This is an illustrative example, not a client result. Suppose a Team package at €6,900 automates a set of workflows that recover 80 hours a month at a loaded cost of €30 per hour, prevents about €600 of rework each month, and costs €300 a month in platform, model and care fees.

• Gross time value: 80 × €30 = €2,400/month
• Net monthly benefit: €2,400 + €600 − €300 = €2,700
• Payback: €6,900 ÷ €2,700 ≈ 2.6 months

Whether your numbers look like this depends entirely on your baseline. Change every assumption in the ROI calculator, or read how to calculate automation ROI for the method and its common mistakes.

Risks to price in

Automating a process nobody has measured means you cannot prove it paid off. Automating a broken process makes the problem faster. And building on a provider’s accounts makes the cheapest quote the most expensive one when you want to leave. None of these are technical risks; all of them show up on the invoice.

Next actions

• Pick one process and write down its monthly volume, hours and error rate before you ask for a quote.
• Ask every provider for the three cost lines separately: implementation, running costs, care.
• Run the free Opportunity Scan to get a fit score and a package suggestion without a sales call.