How much does AI implementation cost in 2026?

Market ranges and my starting prices for an AI audit, automation, integration and an AI assistant. Plus what quotes leave out: the monthly running cost.

Dawid ZmarzlakPublishedOctober 6, 20266 min read

More and more often, instead of "how much is a website?" I hear "how much does it cost to implement AI?". The honest answer: it depends less on the AI itself and more on what it has to connect to and what a mistake costs. Below are the ranges I see in offers on the Polish market, my starting prices, and the costs that keep arriving every month after go-live. All amounts are net, in PLN.

2-8k

automating one process

PLN on the market. My price: from PLN 2,500 per process.

5-15k

AI assistant on company data

PLN on the market + 600-1,200/month. My price: from PLN 7,900 + 390/month.

≤300

PLN/month for the model API

In a typical small business. The bill grows with traffic, not with features.

Market ranges vs my prices

What you implement	Typical market range (net)	My starting price (net)
AI audit and consulting	PLN 2,500 - 5,000	PLN 1,900, counts towards the implementation
Automating one process	PLN 2,000 - 8,000	PLN 2,500 per process
FAQ chatbot (answers from a fixed list)	PLN 3,000 - 8,000	part of the AI assistant
Assistant / chatbot on company data (RAG)	PLN 5,000 - 15,000 + 600 - 1,200/month	PLN 7,900 + care from PLN 390/month
AI integration in an existing app	depends on the app	PLN 5,900
AI agent integrated with a CRM	PLN 15,000 - 45,000	quoted after the audit

The market ranges are a rough guide from public price lists and offers of Polish contractors (autumn 2026), not an industry price list. "From" with me means the smallest sensible scope I would buy myself - not a cut-down "starter" version you will have to rewrite later.

What you get in each line

AI audit and consulting - from PLN 1,900

We go through your processes and look for places where someone retypes data by hand, answers the same questions over and over, or waits for somebody else. The result is a list of 2-3 processes with an estimate of time saved, implementation cost and running cost - plus an honest note on where AI makes no sense. If we go ahead, the audit fee counts towards the implementation.

Process automation - from PLN 2,500 per process

Typical examples: a lead from a form lands in the CRM and the calendar, invoices from KSeF (the Polish national e-invoicing system) end up in a spreadsheet or the accounting system, order e-mails turn into table rows. Where it is enough, I build it in n8n or Make. When the process touches a system without an API or a large volume - I write code.

This is usually not an "AI agent", and I won't call it one. It is solid automation, often with a single step where a language model reads or classifies something. An agent is a system that chooses its own next steps - it is useful less often than the ads suggest, and it costs more because it needs more safeguards.

AI integration in an app - from PLN 5,900

A language model wired into a system you already have: summarising tickets, classifying e-mails, extracting data from documents, drafting replies for a human to approve. Included: model choice, rate limits and provider-error handling, logging, and tests on your data.

AI assistant on company data - from PLN 7,900 + PLN 390/month

A chatbot that answers from your documents, price list and FAQ (RAG - searching your own knowledge base instead of guessing), hands the conversation over to a human and collects contact details. Monthly care means updating the knowledge base, watching answer quality and reacting when the model provider changes something. The assistant says it is an AI from the first message - why that is now a legal duty is covered in the post on the AI Act and chatbots.

What drives the price

How many systems, and whether they have an API. A CRM with a good API is a matter of hours. A system without an API, a daily CSV export or an old ERP takes days.
Data quality and volume. An assistant on 20 tidy FAQ pages is a different project from one on 3,000 PDFs in five versions.
The cost of a mistake. If a model error ends with a fix in a draft, a simple flow is enough. If it goes straight to a customer or to accounting, you add human approval, tests on real cases and confidence thresholds.
GDPR and security. A model in an EU region, a data processing agreement with the provider, sometimes a locally hosted model - each of these decisions changes the architecture and the cost.
Volume. No-code platforms charge per operation. At high volume, custom code or self-hosted n8n often becomes cheaper within a few months.

Monthly costs that quotes leave out

Item	Roughly per month
n8n on your own VPS	PLN 30 - 100 (the server)
Make / Zapier, paid plan	PLN 200 - 2,000, depending on operations
Language model API (Gemini, Claude, GPT)	under PLN 300 in a typical small business
AI assistant care plan	from PLN 390

How I test this

I only show what runs in production on my side. The chat on this site runs on Gemini and sends contact details to my admin panel. The job aggregator I use collects offers from several portals, scores them with a language model, and when the main model does not respond, a backup takes over the scoring. The follow-up e-mail sequence only sends on working days, during working hours. Details and the stack are on the AI and automation page.

How to start without burning the budget

Pick one process that hurts the most and repeats every week.
Count the hours it takes per month today. If they are worth less than the cost of running the automation - skip it or wait.
Run a pilot on real data before rolling out everything.
Open accounts and access in the company's name, not the contractor's.

Want the numbers for your case? See the AI and automation offer or tell me briefly what you want to automate - I reply within 24 hours on working days.