Guide: AI in business › AI implementation, tools and team skills
AI implementation in a company starts with a process map, not with choosing a tool. Companies that succeed start with three repetitive tasks and measure time and quality before changing anything, and a person approves every piece of material that goes out. Below you will find a 90-day plan for a marketing and PR team, the cost structure, obligations under GDPR and the AI Act, and how to check the effect.
In 60 seconds The implementation plan in a nutshell
- What it involves: you change how a few specific tasks are done. There are rules, an owner and measurement. Buying licences on its own is not yet an implementation.
- Who will find it useful: companies where the team already uses AI in its own way, and the board doesn't know with what effect or on what data.
- Where the money goes: licences are usually the smallest item. The biggest cost is the team's time: the pilot, integrations, quality control.
- First move: choose the three tasks that take up the most hours and measure how long they take today.
- What to watch out for: personal data in public tools, invented figures stated in a confident tone, and obligations under the AI Act.
What AI implementation in a company is — and what it isn't
AI implementation is a lasting change in how specific tasks are done, in which you establish who uses the tool and for what, what data it works on, who checks the result and how you will know that things have improved. A few ChatGPT or Copilot licences are just a purchase. It becomes an implementation when the process changes and the change can be measured.
Poland is clearly lagging behind here. According to Eurostat, in 2025, 20.0% of companies in the European Union with at least 10 employees used AI technologies. In Poland it was 8.4%, one of the lowest figures in the EU. Across the EU, AI is most often used for text analysis, generating images and sound, and writing texts. For marketing and PR, that happens to be good news, because artificial intelligence in business starts where people work with text. In other words, with you.
Implementation is not an IT project that ends on launch day. Nor is it a one-off training session. If, after three months, people are working in the same way and just have an extra browser tab, there was no implementation. We have gathered the bigger picture, from strategy to regulations, in guide to artificial intelligence in business; this article is about the implementation itself in a communications team.
Who AI implementation is for and in what situation it makes sense
ABOUT AI implementation for businesses is something boards most often ask about in one of four situations. Each requires a different starting point.
- The team already uses AI in its own way. Everyone has a different tool, and nobody knows what is being fed into external models. Start: an AI policy for the company and a list of approved tools, and only then new uses.
- Marketing produces a lot, but slowly, because reports, media monitoring summaries and yet more text variants take up days the team would rather spend working with clients. Start: measure the time taken by the three most repetitive tasks.
- The board wants ‘AI in the company’ but doesn't know where to start. There are twenty ideas and a budget for one. Start: a diagnosis and an order of priorities, not another tool presentation.
- The company operates in a regulated industry or processes a lot of personal data. Start: a lawyer and the data protection officer at the table from day one, before anyone pastes the first document containing customer data into a tool.
Where should a company start with AI implementation?
With three questions. Which tasks take up most of the team's hours? Which decisions do you currently make by gut feeling? What data must not end up in external tools? Tools are chosen only after that. The reverse order — licences first, then looking for uses for them — usually ends with several subscriptions used by one person.
Polish data shows where the risk lies. In a ZFPR (Polish PR Agencies Association) survey from June 2026, 90% of PR agency boards admitted to using AI for research and analysis. At the same time, 67% of employees have low or very low trust in AI output, and 76% of boards consider security the most important area. The conclusion is simple. Quality control and data protection rules must work from day one, otherwise people will stop trusting the tool before it has had a chance to be useful.
Look at the whole process, not at a single step. If you don't have a documented workflow, draw a simple diagram; we explain how to optimize the marketing process. AI will speed up every stage. Including the one that should disappear.
How to implement AI in your company in 90 days, step by step
Ten plan sprawdza się w marketingu i PR, a ten sam rytm działa też w sprzedaży czy obsłudze klienta. Każdy etap kończy się konkretnym wynikiem. Nie „dalszą analizą”. Harmonogram, zadania i wskaźniki z każdego etapu najlepiej trzymać w jednym miejscu. Tak działa Commwork®AI cockpit, w którym każdy wskaźnik ma właściciela, próg i źródło.
- STEP 01Days 1–15: diagnosis
You draw a process map, measure time and quality ‘before’ and list sensitive data, and at the end you choose three tasks for the pilot, each with a calculated target, for example a campaign report in 2 hours instead of 6.
- STEP 02Days 1–15 in parallel: rules
A short AI policy: which tools are allowed, with what data, who approves the output. One page is enough.
- STEP 03Days 16–45: pilot
You prepare tools and instructions for the three tasks, train the team on its own examples and check every result before it leaves the department. Outcome of this stage: a comparison of time and quality with the ‘before’ measurement.
- STEP 04Days 46–75: scaling
Expanding what works, switching off what doesn't. Outcome: work instructions and an owner for each process.
- STEP 05Days 76–90: review and decision
Pomiar efektu, raport dla zarządu, plan na kolejny kwartał. Zestaw miar ustal jeszcze w diagnozie; pomoże lista marketing indicators, które da się liczyć co miesiąc. Wynik to jedna z trzech decyzji: rozwijamy, zmieniamy albo kończymy.
| Stage | Days | What we do | Result |
|---|---|---|---|
| Diagnosis | 1–15 | process map, ‘before’ measurement, inventory of sensitive data | 3 tasks for the pilot, each with a target |
| Pilot | 16–45 | tools and rules for the 3 tasks, training, quality control | comparison with the ‘before’ measurement |
| Scaling | 46–75 | expanding what works; switching off the rest | work instructions, process owner |
| Control | 76–90 | measuring the effect, report for the board | decision: expand, change, stop |
Jak Commwork®AI prowadzi wdrożenie AI przez pierwsze 90 dni
Wdrożenie AI najczęściej upada nie na narzędziu, tylko na braku pomiaru i właściciela. Commwork®AI zaczyna od mapy procesów i programów, które firma już ma, a potem prowadzi harmonogram, wskaźniki i decyzje w jednym kokpicie. Za wdrożeniem stoi zespół Commplace®, a każdą decyzję podejmuje człowiek.
- Team time for the pilot and learning 36%
- Integracje z systemami 24%
- Quality control 18%
- Licencje i opłaty za modele 12%
- Maintenance 10%
- 2.02.2025Art. 4 AI Act: firma dba o wiedzę pracowników o AI
- 29.06.2026ZFPR: 67% pracowników agencji PR mało ufa wynikom AI
- 2.08.2026Art. 50 AI Act: zasady przejrzystości treści i chatbotów
- Week 4Pilotaż trzech zadań z celem liczbowym
- Week 12Raport dla zarządu: rozwijamy, zmieniamy albo kończymy
- 2.12.2026Termin technicznego znakowania treści dla starszych systemów
- 01Diagnoza (dni 1–15)
- 02Pilotaż (dni 16–45)
- 03Skalowanie (dni 46–75)
- 04Decyzja (dni 76–90)
- W pierwszym miesiącu mapuje procesy i programy, które firma już ma, i pokazuje, ile z nich naprawdę wykorzystujecie.
- Prowadzi harmonogram wdrożenia z kamieniami milowymi i bramkami decyzji zarządu, a każdy wskaźnik ma źródło, próg i właściciela.
- Zapisuje decyzje w rejestrze: kto, kiedy, na jakich danych i z jakim skutkiem, a wpływ sprawdza grupą porównawczą. Jak wybrać partnera do takiego projektu, porównujemy w tekście AI software house or AI agency.
The view reproduced is modeled after the dashboards provided to customers. The names, numbers, and terms are examples and show the way the system works, not the results of a specific company.
Which marketing processes should you automate first?
To start with, choose tasks that come back every week and are easy to check before publication, and leave for later those where relationships, legal liability or tone at a difficult moment matter.
What does the use of AI in business look like in a communications team? For example: the morning media monitoring summarised in five points for the board, meeting notes with a list of tasks and deadlines, three headline variants for testing, a first draft of a reply to a typical customer question. We describe more uses on the page AI in marketing. You will find specific programs in the roundup AI tools in marketing and in our overview of AI tools for businesses, organised by task.
| Good to start with | With caution | Don't hand over to AI |
|---|---|---|
| summaries of media monitoring | first drafts of website copy | publication decisions |
| organising customer data | answers to typical customer questions | conversations with journalists |
| campaign reports from a single data source | graphic and headline variants for testing | crisis statements |
| meeting notes with tasks and deadlines | competition analysis | financial commitments and contracts |
How much does implementing AI in a company cost?
Uczciwa odpowiedź brzmi: zależy od zakresu, a licencje to zwykle najmniejsza część rachunku. Dla orientacji: Microsoft 365 Copilot dla firm kosztuje na polskiej stronie Microsoftu od 15,60 euro netto za użytkownika miesięcznie przy rozliczeniu rocznym (cena promocyjna do 31.12.2026, stan na 24.09.2026). Przy dziesięciu osobach to około 156 euro netto miesięcznie. Mniej niż tydzień pracy specjalisty, który poprawia skutki źle wdrożonego narzędzia.
When several variables are uncertain, it is better to show a range with a probability than a single amount, and the method for this is Monte Carlo, i.e. repeatedly simulating many cost scenarios. The board then hears ‘with 80% certainty we will stay within the range X–Y’, rather than a single figure that is bound to be wrong. Our AI implementation diagnosis and strategy costs from PLN 8,000 net, and the first conversation is free of charge (as at 24 September 2026).
Chcesz, żeby ktoś policzył ten rachunek na Twoich procesach, zanim kupisz pierwszą licencję? Agencja AI Commplace zaczyna od diagnozy, po której wiesz, które trzy zadania wziąć do pilotażu i ile godzin zespołu one pochłoną.
| Cost item | What it depends on | What to watch out for |
|---|---|---|
| Licences and model usage fees | number of people, plan (business or consumer), usage | providers' price lists change every few months |
| Integrations with company systems | email, CRM, drive, analytics (how many there are and how old they are) | the most frequently underestimated technical item |
| Team time for the pilot and learning | number of people and tasks in the pilot | the most frequently underestimated item overall |
| Quality control | risk of error in a given task | someone has to check the results, especially at the beginning |
| Maintenance | changes to tools, staff turnover | updating the rules and training new people |
Implementing AI in line with GDPR and the AI Act: legal aspects
AI regulations for businesses currently come down to two main sources of obligations: GDPR, i.e. personal data protection, and the EU regulation on artificial intelligence, the AI Act. A full discussion of the regulations, with dates and examples, can be found in the article The AI Act in Poland. Here we give only the dates that concern the marketing team.
Since 2 February 2025, companies using AI have been required to look after their employees' AI literacy (Article 4), and in July 2026 an amending regulation known as the Digital Omnibus came into force, which relaxed this provision. The obligation remains, but without a prescribed level of knowledge: measures should suit the tools used, the roles in the team and the risks. The European Commission explains that no certificate is needed; an internal record of training is enough.
Since 2 August 2026, the transparency rules under Article 50 have applied: informing people that they are talking to a chatbot, and labelling certain AI-generated content. The Omnibus gave providers of systems placed on the market before that date until 2 December 2026 to introduce technical marking of content. We have set out what this means for press releases and graphics in our article on labelling AI content. Obligations for high-risk systems, e.g. in recruitment, have been postponed to 2 December 2027.
On the GDPR side, the questions are simple, but they need to be asked before you start: does the tool process personal data, will the provider sign a data processing agreement, is company data excluded from model training, and where is it stored?
Copilot, ChatGPT or AI agents to start with
If the company works on Microsoft 365, the natural first step is often a Copilot rollout, because it works in the email, Teams and documents the team has known for years. Companies on Google Workspace more often turn to Gemini, and creative teams to ChatGPT. We have set out all three routes side by side in the comparison ChatGPT, Gemini or Copilot for business, while licences and the rollout process at Microsoft are covered in a separate guide to Copilot for business.
Deploying AI agents only makes sense once the process has been documented and measured. An agent is a program that carries out several steps in sequence on its own: it collects data, writes a report and sends it for approval. It will speed up a good process. It will replicate a bad one just as efficiently. We describe what an agent can realistically do in marketing and how much it costs in the article AI agents in marketing and PR.
The same rule applies at every stage. AI prepares, and a person decides: on publication, on spending, on commitments to clients. At Commplace we call this a human in the decision loop.
When to implement AI — and when it's better to wait
Dobry moment to taki, w którym zespół ma czas na pilotaż. Start w szczycie sezonu kampanii albo tydzień przed dużą premierą kończy się przewidywalnie: przy pierwszej presji ludzie wracają do starych nawyków. To zwykły opór przed zmianą, a o tym, jak prowadzić zmianę w organizacji, żeby przetrwała pierwszy trudny tydzień, piszemy osobno.
Signs that it's a good time: repetitive tasks are growing faster than the team, the department uses AI without rules, the board asks for reports that nobody has time to put together. Signs that it's better to wait: no one to take responsibility for the implementation, an ongoing reorganisation, or a lack of the basic data AI would need to work with.
If someone in the team wants to go deeper into the theory, they might consider a university course; we compare programmes in our overview postgraduate studies in AI. For the implementation itself, practical AI training for businesses based on the team's own tasks is enough. We explain how to plan it in our guide on how to choose AI training for businesses.
Wdrożenie AI w marketingu w typowej firmie, czyli te 90 dni w praktyce
Wdrożenie AI w marketingu według planu z tabeli wyżej wygląda porządnie na papierze. Poniżej scenariusz poglądowy, jak zwykle przebiega w życiu. To nie historia konkretnego klienta, tylko typowy układ: producent B2B, pięcioosobowy dział marketingu i PR, zarząd, który słyszał, że sztuczna inteligencja w przedsiębiorstwie „daje oszczędności”, i chce wiedzieć jakie.
Pierwsze dwa tygodnie. Spis narzędzi pokazuje, że zespół używa kilku różnych programów AI, każdy na innym koncie. Nikt nie wie, co do nich trafiło. Powstaje jednostronicowa polityka i trzy zadania do pilotażu: raport z kampanii, poranne streszczenie media monitoring i notatki ze spotkań. Dla każdego zapisany jest czas „przed”. O zmianie mówicie zespołowi wprost, bo plotka o „zastępowaniu ludzi przez AI” rozchodzi się szybciej niż polityka; pomoże tu dobra internal communication.
Dni od 16 do 75. Pilotaż, a potem skalowanie. Zwykle jedno z trzech zadań odpada. Najczęściej to, w którym poprawki zjadają cały zysk, bo model podaje pewnym tonem liczby, których nie ma w źródle. To nie porażka wdrożenia, tylko jego normalny wynik. Dwa pozostałe zadania dostają instrukcje pracy i właściciela. Wspólna biblioteka poleceń, o której piszemy przy prompts for marketing, skraca naukę nowych osób.
Ostatnie dwa tygodnie. Pomiar efektu i raport dla zarządu. Przykład z tego artykułu: raport z kampanii w 2 godziny zamiast 6. Zanim ktokolwiek go ogłosi, sprawdzacie czas całego procesu razem z poprawkami i porównujecie z grupą, która w tym samym miesiącu pracowała po staremu. Wdrożenie AI w biznesie kończy się decyzją, a nie prezentacją.
Próg decyzji przyjmij na starcie. Przykładowa reguła: zadanie zostaje, jeśli czas całego procesu spadł co najmniej o jedną trzecią, a liczba błędów, które wyszły poza firmę, nie wzrosła. Licencja nieużywana przez miesiąc wraca do puli. Jak wykorzystać AI w biznesie dalej, decydujecie dopiero z tymi liczbami w ręku. Gotowe wzory mierników znajdziesz w zestawieniu przykładów KPI.
Jeśli wdrożenie prowadzi ktoś z zewnątrz, zadaj mu cztery pytania. Co dokładnie dostaniemy po diagnozie? Z jaką grupą porównawczą zmierzycie efekt? Czy wykorzystacie licencje, które już mamy, zanim zaproponujecie nowe? Kto u nas zostanie właścicielem procesów, gdy projekt się skończy? Podobnie wygląda wdrożenie ChatGPT czy Copilota: pytania są te same, zmienia się tylko cennik.
Lista kontrolna na pierwszy dzień:
- Spis narzędzi AI, których zespół już używa, także na prywatnych kontach.
- Lista danych, które nie mogą trafić do zewnętrznych modeli.
- Jednostronicowa polityka AI zatwierdzona przez zarząd.
- Zmierzony czas „przed” dla trzech zadań do pilotażu.
- Jedna osoba odpowiedzialna za wdrożenie i zastępca.
- Prawnik i inspektor ochrony danych wpisani do kalendarza pierwszego tygodnia.
- Data decyzji zarządu po 90 dniach wpisana już dziś.
What are the risks of implementing AI and how do you measure the effect?
Cztery ryzyka wracają w niemal każdym projekcie. Każde ma proste zabezpieczenie, a jeśli wolisz omówić je na przykładzie własnej firmy, Commplace AI agency zaczyna każde wdrożenie od diagnozy procesów i ryzyk.
- Data leaks. Confidential information ends up in public tools. Safeguard: an inventory of sensitive data and tools covered by a data processing agreement.
- Errors stated in a confident tone. AI can invent a figure or a quote. Safeguard: a person reads the text before publication (we explain where such slip-ups come from in our article on AI hallucinations).
- Illusory savings. The text is produced faster, but corrections can eat up the entire gain, so at the end of the week the team hasn't saved a single hour, even though everyone feels they are working faster. Safeguard: measure the time of the whole process, not a single step.
- Legal obligations, i.e. the rules on transparency and employees' AI literacy. Safeguard: an AI policy and a record of training.
Where AI implementations most often go off the rails
The team pays for tools before it knows what it will use them for. The cost: several months of fees with no effect, and disillusioned people who will say ‘we've been through this already’ next time round.
Without a baseline, any result can be presented as a success or a failure, so the board has no basis for a decision on scaling, and the budget goes wherever someone argues the loudest.
An invented figure or quote in a press release is a reputational problem, and sometimes a legal one, which ends in a correction, loss of journalists' trust and, in extreme cases, a crisis.
Ten pilots at the same time means none of them gets attention, responsibility becomes blurred, and after a quarter the board doesn't have a single convincing result on which to base a budget decision.
Customer data in a tool without a data processing agreement means a risk of breaching GDPR. The cost can be painful: the implementation stalls halfway and part of the work has to be redone.
Questions boards ask before their first pilot
How do you implement AI in a company, step by step?
Start with a diagnosis: choose three repetitive tasks and measure how long they take today. Write down rules for using AI, run a pilot with quality control, and after 90 days compare the results with the baseline measurement.
How much does implementing AI in a company cost?
It depends on the scope: the number of people, integrations with systems and the team's time for the pilot. Licences are usually the smallest part. For example, Microsoft 365 Copilot for business costs from EUR 15.60 net per user per month on annual billing (promotional price until 31 December 2026, as at 24 September 2026). The biggest costs are people's time and quality control.
Who should be responsible for AI in the marketing department?
One person from the team, with the authority to decide on tools and rules, supported by IT and a lawyer, because diffused responsibility ends with everyone using a different tool on different terms. This person also keeps the tool inventory and the training record.
Does AI implementation have to comply with GDPR?
Yes, if the tool processes personal data, which in marketing and sales is almost always the case. You need to check the data processing agreement with the provider, where the data is stored and whether company data is used to train the model.
Does a company have to train its employees in AI?
Since 2 February 2025, the AI Act (Art. 4) has required companies using AI to look after their employees' AI literacy. Following the changes of July 2026, the provision does not prescribe a specific level, and the European Commission explains that no certificate is needed. An internal record of role-appropriate training is enough.
How does an AI agency differ from a software company?
A software company builds a program to a specification. An AI agency starts earlier, with the question of which decision and which task need improving, and is responsible for the effect in the process, not just for the code.
How can you check whether an AI implementation has paid off?
Compare the time and quality of the same tasks before and after the implementation, ideally in two groups — with and without AI — over the same period. Count the time of the whole process, including corrections, not just the first draft.
Sources
- SOURCEZFPR: survey on the use of AI in PR agencies · 29 June 2026
- SOURCEEurostat: 20% of EU enterprises use AI technologies · 11 December 2025
- SOURCEEuropean Commission: transparency obligations under Article 50 of the AI Act · as of September 2026
- SOURCEEuropean Commission: AI literacy — questions and answers (Art. 4) · updated 27 July 2026
- SOURCE
- SOURCEMicrosoft: Microsoft 365 Copilot pricing for business (Poland) · as of 24 September 2026
- SOURCEEUR-Lex: Regulation (EU) 2026/1744 (Digital Omnibus on AI), Polish text · of 8 July 2026, OJ EU of 24 July 2026, accessed 24 September 2026
Read next
Zaczynamy od diagnozy: które zadania w marketingu, PR i sprzedaży dadzą najwięcej i jak zmierzymy efekt. Dostajesz mapę procesów, trzy zadania do pilotażu z celem liczbowym i plan na 90 dni. Pierwsza rozmowa jest bezpłatna.
Sebastian Kopiej, President of Commplace®. In public relations since 1996. Text prepared with the help of AI tools and fact-checked by the author. Data current as of September 29, 2026.
Read more on this topic