AI implementation in business starts with choosing one repetitive process, not with choosing a tool. A language model makes sense where someone has to read data from documents, sort incoming requests or draft a reply — and a person approves the result anyway. Indicative effort in our pricing model: AI in a process is 20–56 h of work, i.e. €920–2,600 net (excl. VAT) at a rate of €46/h net, plus an automation foundation of 10–20 h, i.e. €460–920 net.
Euro amounts converted at the NBP mid-market rate of 14 September 2026 (€1 = 4.3391 zł) and rounded. We invoice in PLN.
Where does artificial intelligence deliver real value in a company?
AI is useful where data is unstructured — that is, where a person has to read and understand something before passing it on.
Typical uses:
- reading data from documents: invoices, orders, PDFs from business partners,
- sorting incoming requests: assigning a category and priority, routing them to the right person,
- draft replies for customer service — an employee edits and sends them,
- a chatbot answering repetitive questions based on your own materials,
- generating content and analysing data that nobody has time to review.
When AI is not needed
If a process has clear rules — “when an order with status X comes in, create document Y and send a notification” — you do not need a language model. Plain rule-based automation is simpler and more predictable. We describe more examples like this in our article on business process automation.
A simple principle: rules where there are rules. AI where text has to be understood.
How to implement AI in a company step by step
1. Pick one process. The best candidate is repetitive and takes someone the same hours every week: data retyped by hand between systems, reports stitched together from several spreadsheets, reminders sent manually, mistakes made in a rush.
2. Put it in order before you automate anything. A process that changes every month, or that nobody can describe step by step, has to be written down first. Automating chaos only gives you faster chaos.
3. Decide where a person approves the result. Reading an invoice can run automatically, but posting it to the books and sending it to the client are decisions with consequences. Set the approval point at the start, not after launch.
4. Build a prototype. For us that means a data flow diagram or a clickable mock-up of the panel — before any code. This is the cheapest stage at which to change your mind.
5. Work in sprints. Every 1–2 weeks you see a working part, not a presentation.
6. Launch, monitor, then the next process. Only once the first process works and people trust it is it worth taking on the next one. We describe the whole way we work on our business process automation page.
What is the difference between a business chatbot and an AI agent?
A chatbot answers questions. An AI agent carries out steps in your systems — it opens a ticket, fills in customer data, prepares a document.
The practical difference is that a chatbot mainly needs access to good content (terms and conditions, price list, knowledge base), while an agent also needs permissions and integration with your systems via API. Without a connection to your systems, an agent has nothing to act on — all that is left is the conversation.
The second difference is risk. A wrong answer from a chatbot is a misunderstanding. A wrong action by an agent is an issued invoice or a message sent to a client. That is why we leave actions with external effects for a person to approve.
How much does AI implementation in a company cost?
We bill by the hour: development €46/h net (excl. VAT), design €35/h net. Below are the AI-related items from our pricing model — these are indicative price ranges and depend on scope.
| Item | Hours | Net cost (€46/h) |
|---|---|---|
| AI in a process: reading data from documents, sorting requests, draft replies | 20–56 h | €920–2,600 |
| Automation foundation: architecture, task queue, logs, error handling, deployment | 10–20 h | €460–920 |
| An AI feature in a web or mobile app: chatbot, content generation, analysis | 16–44 h | €740–2,020 |
| Connection to each system (e.g. ERP, CRM, shop) | counted separately for each system | depends on the system and its API |
On top of that comes the fee for the language model charged by the AI service provider — billed separately and dependent on the number of queries. This is a running cost, not part of the implementation.
What pushes the price up and what brings it down
The cost goes up when a system has no API or documentation — in that case we add around 60% to the hours for the connection. It also goes up with many exceptions in the process, data that needs cleaning and integrations with several providers at once.
The cost goes down when the process is clearly described, you have access to API documentation and you start with one process instead of five. You will find the full hour tables for automation and integration in our article on the cost of automation and system integration, and you can calculate a price range for your own scope in the cost calculator — including for a feature described in your own words.
When does an AI implementation pay for itself?
The maths is simple: hours recovered per month × the cost of an hour of work, set against the one-off cost of the implementation.
Example (an illustration, not an offer). Assume that handling documents takes 30 hours a month in your company and an hour of work costs €14. That is €410 a month. Implementation: AI in a process 30 h + automation foundation 15 h = 45 h × €46 = €2,050 net (excluding connections to systems, which are counted separately). The payback comes after about 5 months, assuming the process really does disappear from people's calendars rather than simply taking half as long.
That is only a calculation based on assumptions. Your numbers will be different — they start with how many hours the process actually takes today.
AI and customer data under GDPR — what to take care of
If you send customer data to an external AI service, that service is usually a processor. That means a data processing agreement setting out, among other things, the subject matter, duration, nature and purpose of the processing, the type of data and the categories of data subjects.
The rules also imply a few things worth checking before you start:
- the provider must offer sufficient guarantees of appropriate technical and organisational measures,
- it may not engage a sub-processor without your written consent,
- once the service ends, it deletes or returns the data — depending on your decision.
On our side it looks like this: we work on the basis of a contract and an NDA, the code goes into the client's repository, and external services are registered in the client's name. Access to the data stays under your control. CodePort does not provide legal advice — confirm the scope of your obligations with a lawyer or data protection officer.
The AI Act — what applies to your company today?
The Artificial Intelligence Act (Regulation (EU) 2024/1689, as amended by Regulation (EU) 2026/1744) applies in stages. Obligations depend on your role: a provider of an AI system has different ones from a deployer using it.
| What | From when |
|---|---|
| Prohibited practices under Article 5, and supporting AI literacy among staff | 2 February 2025 |
| Application of the regulation, including transparency obligations (a chatbot informs you that you are talking to an AI; labelling of synthetic content) | 2 August 2026 |
| New prohibitions added by the 2026 amendment (Article 5(1)(ba) and (bb), (1a) and (1b)) | 2 December 2026 |
| Labelling of outputs by providers of systems placed on the market before 2 August 2026 | 2 December 2026 |
| Obligations for high-risk systems under Annex III (including recruitment, CV screening, employee evaluation) | 2 December 2027 |
| High-risk systems under Annex I | 2 August 2028 |
The AI literacy obligation means taking measures that support the development of knowledge about AI among your staff and the people operating systems on your behalf, taking into account their experience and the context of use. It does not require specific individuals to reach a defined level.
Transparency in practice: the provider of a chatbot has to ensure that the chatbot informs the person that they are dealing with an AI system, unless this is obvious — and it must do so at the latest at the first interaction, in a clear way. A deployer using AI for deepfakes discloses that the content has been artificially generated. The same applies to text published in order to inform the public on matters of public interest — unless it has undergone human review or editorial control and someone holds editorial responsibility for it.
The uses described in this article — reading data from invoices, sorting requests, draft replies — are not listed in Annex III. Recruitment and employee evaluation are, so plan for more headroom there. This is not legal advice; if in doubt, you need a lawyer.
How AI works at our company — two examples
The first is our own website. In the cost calculator and the project brief builder, a language model estimates the effort for features you describe in your own words and prepares an initial technical specification from a completed brief.
The second is an automated blog for the BKS Rocky boxing club: the system picks topics based on Google suggestions, writes articles, checks prices and contact details against the club's own facts, and publishes the texts in four languages — once the owner approves them. The calculator's estimate is an indicative price range — a detailed quote is prepared by our team.
How to prepare for a conversation about an AI implementation
To get a credible price range, prepare four things:
- a step-by-step description of the process — who does what, and in what order, today,
- a few sample documents or requests, including the unusual ones,
- a list of systems to connect and whether they have API documentation,
- the name of the person who approves the result.
We reply to an enquiry within one working day, and we prepare a free price range and schedule within 2–3 working days.
If you have a process in your company that eats the same hours every week, start by describing it — we will work out the rest together on our business process automation page.
Frequently asked questions
How to implement AI in a company with no IT department?
You do not need developers of your own — you need someone on your side who knows the process and can approve decisions. The rest, including access rights and service configuration, can be handled remotely; we register services in the client's name, so the accounts stay with you.
Does a business chatbot have to state that it is artificial intelligence?
Yes — from 2 August 2026, AI systems intended to interact directly with people must inform them that they are talking to an AI system, unless this is obvious. The obligation lies with the system provider, so when choosing or commissioning a chatbot, check that it does this. The information must be clear and appear at the latest at the first interaction.
Do we have to replace the systems we use in order to implement AI?
Usually not. We first check whether your existing tools can be connected via API — replacing a system is a business decision, not a condition for implementing AI.
How much does it cost to maintain an AI solution after launch?
With our applications, post-launch support (monitoring, backups, updates, further development) typically comes to 5–15 hours a month, billed at the hourly rate. On top of that comes the running fee for the language model at the provider, depending on the number of queries.
Does the code belong to us after the project?
Yes. Once the project is settled, full intellectual property rights to the code and design pass to the client, and the code goes into their repository. We work on the basis of a contract and an NDA.