Guide: AI in business › AI in sales and business departments
AI in company finance works best where there are lots of repetitive documents and data: describing invoices, matching payments, spotting unusual costs and preparing the first draft of an analysis. Financial decisions, tax settlements and signing off the financial statements stay with people. Below we explain what works, what not to hand over to a model, how to introduce AI in the finance department and how to correctly download and book a ChatGPT invoice.
In 60 seconds Finance and AI: a quick overview
- Scope: artificial intelligence used for document processing and cost control, as well as for forecasts and analysis in the finance department.
- Who gains the most: companies with a large number of cost invoices, several entities or locations, and data scattered across spreadsheets.
- Budget: general-purpose tools from a few dozen dollars per user per month, but integration with the accounting system and quality control cost the most.
- First process: describing and posting cost invoices, which since 2026 have arrived via KSeF in a structured form.
- Pitfall: a language model can confidently give a wrong figure, so check every AI analysis against the source data.
What AI in company finance is
Artificial intelligence in finance is not a single program. It is a set of tasks in which the computer takes the load off people: it reads documents, organises data, flags deviations and writes the first draft of a commentary on the results. In a medium-sized company, it usually starts with accounting and cost control and moves on to forecasts and analysis later. How finance fits into the whole topic, from law to sales, is shown in the guide to artificial intelligence in a Polish company.
According to a Gartner survey from November 2025, 59% of finance departments use AI. The most common applications are knowledge management (for example, quickly finding policies and procedures), automating the processing of cost invoices, and detecting errors and unusual items, i.e. exactly the areas where the finance department loses the most time on work whose results nobody outside it sees. For a Polish company, this is a good signpost. Repetitive tasks first, analytics later.
Where AI really helps in finance and accounting
Sztuczna inteligencja w księgowości zyskała w Polsce nowy grunt dzięki KSeF, czyli Krajowemu Systemowi e-Faktur. Od 1 lutego 2026 roku korzystają z niego obowiązkowo najwięksi przedsiębiorcy, a od 1 kwietnia 2026 roku pozostali. Faktury przychodzą więc w ustrukturyzowanym formacie. To mocno ułatwia automatyczne opisywanie i dekretowanie. AI w księgowości ma więc na czym pracować. Poniżej zastosowania uporządkowane według ryzyka.
| Application | What AI does | Risk | Who checks |
|---|---|---|---|
| Coding and posting purchase invoices | suggests the account, cost centre and project | low | the accountant approves the suggestions |
| Reconciling payments with invoices | matches transfers with documents, flags missing ones | low | the person responsible for receivables and payables |
| Detecting unusual costs | spots duplicates, inflated amounts, new suppliers | medium | financial controller |
| First draft of a commentary on the results | describes variances from budget in words | medium | CFO |
| Cash flow forecast | estimates income and expenditure based on historical data | high | the board, using source data |
| Tax settlements and decisions | shouldn’t do this on its own | high | accountant, tax adviser |
Who AI in finance is for and when it pays off
Those who gain most are companies whose finance department is drowning in documents: several hundred cost invoices a month, several companies in the group, many locations, project accounting. Every hour saved on describing documents is an hour for analysis. In manufacturing companies, where finance ties in with planning and the warehouse, also take a look at the article on AI in manufacturing and ERP systems.
The second group are companies where the board waits weeks for a report because the data has to be gathered from several spreadsheets. AI won’t fix scattered data, but it will help combine and describe it. We describe how to build a tool that shows the board the key figures in one place on the page decision dashboards. Taki kokpit, w którym liczby z systemu księgowego, ERP i systemu klientów spotykają się w jednym miejscu, budujemy w Commwork®AI system. Każdy wskaźnik ma tam źródło, datę i właściciela.
A small company with a dozen or so invoices a month and an external accounting firm stands to gain less. Instead, ask the firm which AI tools it already uses and whether it can share the results with you in a more convenient form, because at that scale building something of your own costs more time than it will ever save.
Financial analysis with ChatGPT: how to use it and what not to do
How can ChatGPT be used in financial analysis? Most safely as an assistant that helps you ask the right questions and describe the results, not as a source of figures. It works well when you need to explain where a budget variance might come from, structure a report for the board, write a spreadsheet formula or review a contract for payment terms. We show how to do this in the spreadsheet itself in the guide on AI in Excel.
Kto wpisuje w wyszukiwarkę „ChatGPT analiza finansowa”, zwykle szuka gotowego polecenia. Ważniejsze jest to, co do niego dołączysz. Dobre polecenie zawiera okres, źródło danych, poziom (spółka czy grupa) i prośbę, by model zaznaczył każdą liczbę, której nie znalazł w załączonym pliku. Taki drobiazg oszczędza godzinę sprawdzania.
What not to do? Don’t ask the model to calculate ratios from memory or to give “current” interest or exchange rates without a source. A language model can state an incorrect figure in a confident tone; we explain why this happens in the article on AI hallucinations. The spreadsheet or the system does the calculating. AI only describes.
English-language guides on AI for financial analysis often show models analysing entire financial statements. For a first review, this works well. Every figure in such a summary still has to be checked against the source document, though, because the model can confuse periods, mix up consolidated and standalone data, or quietly round an amount that matters in the statements.
Jak Commwork®AI daje zarządowi jedną wersję liczb finansowych
W wielu firmach problemem nie jest brak danych, tylko trzy różne wersje tej samej liczby: w systemie księgowym, w systemie klientów i w arkuszu kontrolera. Commwork®AI łączy te źródła, pokazuje odchylenie, gdy przekroczy próg, i zapisuje, na jakich danych zapadła decyzja. Liczby porządkuje zespół Commplace®, a decyzję zawsze podejmuje człowiek.
- Opis i dekretacja faktur 34%
- Uzgadnianie płatności 22%
- Raporty dla zarządu 18%
- Wyjaśnianie rozbieżności 16%
- Analizy i prognozy 10%
- 1.02.2026KSeF obowiązkowy dla największych firm
- 1.04.2026KSeF obowiązkowy dla pozostałych firm
- Tydzień 5Pilotaż opisu faktur na jednej spółce
- Week 12Ocena: czas, poprawki, błędy przy zamknięciu
- 1.01.2027Kary za faktury wystawiane poza KSeF
- 01Mapa danych finansowych
- 02Jedno źródło liczby
- 03Pilotaż na jednej spółce
- 04Kokpit zarządu
- 05Measuring the impact
- Łączy system księgowy, ERP, system klientów i arkusze w jeden rachunek, więc zarząd patrzy na jedną wersję liczby zamiast pięciu raportów liczonych inaczej.
- Buduje kokpit, w którym każdy wskaźnik ma źródło, datę, próg i właściciela, a odchylenie od budżetu widać w chwili przekroczenia progu, a nie na koniec miesiąca.
- Przy ważnej decyzji finansowej pokazuje symulację wariantów i naradę pięciu modeli z zaznaczonym sporem, a podpis zostawia człowiekowi. Dlaczego żadnej liczby z modelu nie bierzemy na wiarę, tłumaczymy w tekście o AI hallucinations.
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.
AI assistants and AI agents in finance: what is realistic in 2026
An AI assistant in finance answers questions and helps with the work, but does nothing on its own. It suggests a posting, explains a procedure, writes an email to a counterparty reminding them about a payment. An AI agent in finance goes a step further and carries out tasks: it retrieves an invoice from KSeF, suggests a description, checks it against the order and creates a task for the approver.
An AI agent for invoices is currently the most mature example, because the process is repetitive and the documents are structured. Even so, start in a mode where the agent prepares and a person approves. We explain how an assistant differs from an agent in the articles on AI assistants in business i o AI agents. Simpler automations, such as passing an invoice from email into the system, can be built without a developer. We show this in the article on automation with AI and n8n.
ChatGPT invoices: how to download an invoice and how to book it
This question comes up in search hundreds of times a month, so we answer it directly. As of September 2026, based on a PłatnikVAT.pl guide from July 2026. This is not tax advice. Confirm the details with your accountant.
Who issues the invoice? When buying directly from OpenAI (website, API, business plan), the issuer is an OpenAI group entity, for customers in the European Economic Area usually OpenAI Ireland Ltd. Check this on the document. When buying through the App Store or Google Play, the invoice is issued by Apple or Google, not OpenAI.
How do you get an invoice in the company’s name? In the account’s billing settings, mark the account as a business account and enter the company details and EU VAT number (the Polish NIP tax number with the PL prefix, registered for intra-Community transactions). You can download the documents in the same tab.
VAT. Buying a service from a foreign supplier is usually an import of services. Under the general rule (Article 28b of the Polish VAT Act), the service is taxed in the country where the buyer company is established, and the tax is accounted for by the buyer itself (Article 17(1)(4)), i.e. the so-called reverse charge applies. An active VAT payer then reports both output tax and, if entitled to deduct, input tax in the same JPK file. A VAT-exempt company must also account for such a purchase, having first registered for EU VAT. If the invoice shows Irish VAT because the account had no EU VAT number, you cannot deduct that tax in the Polish JPK, and the details in the settings need to be completed. Confirm the date the tax obligation arises and the exchange rate with your accountant.
Cost. A subscription used in the business may be a tax-deductible expense. The currency and exchange rate are a matter for your accountant.
- Buy the plan on a company account, not a private one.
- Before the first payment, enter the company details and EU VAT number in the billing settings.
- Co miesiąc pobieraj fakturę VAT za ChatGPT z zakładki rozliczeń i przekazuj go księgowej.
- Check the invoice for the issuer, the absence of foreign VAT and a reverse charge note.
Implementing AI in the finance department, step by step
Wdrożenie AI w finansach powinno wyglądać jak każdy dobry projekt kontrolingowy. Najpierw pomiar, potem zmiana, na końcu sprawdzenie wyniku, czyli zwykłe etapy procesu zarządzania, tyle że z nowym narzędziem. Oto pięć kroków na pierwszy kwartał:
- STEP 01Choose one process
Ideally a repetitive, well-documented one, for example describing cost invoices from KSeF or matching payments. Don’t start with forecasts.
- STEP 02Measure your starting point
How many documents a month, how long each one takes, how many corrections and errors? Without these figures you can’t assess the effect.
- STEP 03Set control rules
What AI suggests, who approves, and what always requires a person, for example every item above a set amount, a new supplier or a correction.
- STEP 04Run a pilot on part of the documents
For example, on one company or one cost category. Compare the AI’s suggestions with the accountant’s decisions and record the discrepancies.
- STEP 05Evaluate after 60–90 days
Compare the time and the number of errors with the baseline period. The fairest comparison is with the same month of the previous year, because year-end and quarter-end closes have their own rhythm. Only then expand.
How much AI in finance costs and what the budget depends on
Licences are usually the smallest part of the budget. What costs more is connecting the tool to the accounting and ERP systems, plus the team’s time spent checking results in the first months, when the accountant does their usual work in parallel and compares it with the AI’s suggestions, so for a few weeks the department’s workload tends to rise rather than fall. We compare general-purpose tools in the round-up ChatGPT, Gemini or Copilot for business.
Dla skali: pięcioosobowy dział finansów zapłaci za ChatGPT Business 100 USD miesięcznie przy płatności rocznej (5 osób po 20 USD, cennik OpenAI, stan na 24.09.2026). To mniej niż jeden dzień pracy księgowej. Automatyzacja księgowości zaczyna jednak kosztować naprawdę dopiero przy integracji z KSeF i bankiem, a potem z ERP, dlatego zanim podpiszesz umowę z dostawcą modułu AI, zapytaj go o pięć rzeczy:
Jeśli u Ciebie najdroższe nie są faktury, tylko czekanie zarządu na liczby, zacznij od jednego ekranu z najważniejszymi wskaźnikami. Tak działają kokpity decyzyjne Commplace: na Twoich danych pokażemy, które liczby da się połączyć od razu, a które najpierw trzeba uporządkować.
- Czy moduł czyta faktury bezpośrednio z KSeF, czy z plików PDF?
- Kto widzi nasze dane i czy dostawca trenuje na nich swoje modele?
- Czy każda propozycja dekretu ma zapis, kto ją zatwierdził i kiedy?
- Ile kosztuje moduł przy dwukrotnie większej liczbie dokumentów?
- Jak wyłączyć automat w trakcie zamknięcia roku bez utraty danych?
| Item | What it depends on | Example or note (as of 24 September 2026) |
|---|---|---|
| General-purpose tool for the finance department | number of users, security requirements | ChatGPT Business: USD 20 per person per month billed annually, USD 25 billed monthly |
| AI modules in the accounting system or ERP | provider, package, number of documents | often in a higher-tier package; ask the provider about the cost separately |
| Integration with KSeF, the bank, ERP | number of systems, quality of their interfaces | individual quote |
| Data housekeeping (dictionaries, cost centres, counterparties) | state of the chart of accounts and dictionaries | internal work, often underestimated |
| Quality control in the pilot | number of documents, review threshold | the accountant’s time for comparing AI suggestions |
Wdrożenie AI w księgowości w 90 dni na przykładzie typowej firmy
Plan wdrożenia na papierze wygląda prosto. W praktyce najwięcej zmienia to, czy firma ustaliła progi decyzji przed startem. Poniżej scenariusz poglądowy, a pod nim lista kontrolna, którą możesz przejść z księgową w jedno popołudnie.
Próg decyzji zapisz na pierwszej stronie planu. Przykład: rozszerzamy automatyzację faktur kosztowych na kolejne spółki, jeśli przez dwa zamknięcia z rzędu odsetek propozycji przyjętych bez poprawek się nie obniża, a liczba błędów wykrytych przy zamknięciu nie rośnie. Jeden z tych warunków nie jest spełniony? Zostajemy na pilotażu i szukamy przyczyny w danych, nie w narzędziu.
Lista kontrolna przed pilotażem:
- Wiemy, ile faktur kosztowych miesięcznie przechodzi przez dział i ile czasu zajmuje jedna.
- Plan kont, słownik MPK i kartoteka kontrahentów są aktualne.
- Jest spisany próg kwotowy, powyżej którego zawsze decyduje człowiek.
- Narzędzie AI działa na koncie firmowym z umową powierzenia danych.
- Ktoś ma w kalendarzu godziny na porównywanie propozycji AI z decyzjami księgowej.
- Wiemy, z którym okresem porównamy wynik, np. z tym samym miesiącem rok wcześniej.
When to introduce AI in finance and when to wait
A good moment is a quieter month after the quarter close. Not December and January, when the finance department is closing the year. Many companies also have a natural reason to act right now: since the switch to KSeF, documents arrive in a single format, and automation is simpler than ever before.
Wait if you are in the middle of changing your accounting or ERP system, if your chart of accounts and reference lists are in disarray, or if nobody in the company can find time to check the results during the pilot. In that case, get things in order first. We describe how to break such problems down into parts before reaching for technology in the article on problem-solving methods in organisations.
How to check whether AI in finance is delivering results
Na start wystarczą trzy wskaźniki: czas obsługi jednego dokumentu, odsetek propozycji AI przyjętych bez poprawek i liczba błędów wykrytych dopiero przy zamknięciu miesiąca. Czas spada, a błędów przybywa? Oszczędność jest pozorna. Kolejne mierniki dobierzesz z listy przykładowych KPI dla firmy, ale zacznij od tych trzech.
A faster month-end close after implementation is not necessarily down to AI. There may simply have been fewer documents. So run the pilot on one company and compare it with the others, or compare the result with the same period last year. If you’d like someone from outside to help design such measurement and implementation, visit the page Commplace® AI agency or AI analytics.
Where finance departments most often stumble with AI
A language model doesn’t calculate like a spreadsheet. It can give a plausible-looking but wrong figure. The result is a wrong board decision based on a report that nobody checked.
Financial statements, payrolls and counterparty data then end up outside the company’s control. A breach of trade secrets or data protection costs more than a business licence.
An invoice without the company’s details or with foreign VAT causes problems at settlement. And a lost deduction. Simply complete the details in the billing settings before the first payment.
Forecasts need clean historical data and a lot of oversight. A company that starts with them usually ends the pilot disappointed, because the first results diverge from reality, nobody can say why, and the board concludes that AI doesn’t work in finance, when it was actually the choice of the first task that failed.
If you don’t decide which documents a person always checks (for example above a certain amount or from a new supplier), a single incorrect posting on a large amount will wipe out a whole year’s savings.
Questions from accountants and finance directors about AI and ChatGPT invoices
How do I download an invoice from ChatGPT?
When buying directly from OpenAI, you will find the documents in your account’s billing settings. When buying through the App Store or Google Play, the invoice is issued by Apple or Google, so you look for it in the purchase history of the app store, not in your OpenAI account, where no such document will exist.
How do I get a ChatGPT invoice in my company’s name?
In your OpenAI account’s billing settings, mark the account as a business account and enter the company details and EU VAT number. Do this before the first payment. You will receive correct documents from the start.
How do I book a ChatGPT invoice?
Usually as an import of services accounted for under the reverse charge mechanism (Article 28b and Article 17(1)(4) of the Polish VAT Act), and the expense itself as a business cost if the subscription is used for business purposes. Confirm the details, including the currency exchange rate, with your accountant. This is not tax advice.
Why does the invoice from OpenAI show Irish VAT?
Most often because the account didn’t have an EU VAT number and was treated as a private customer. You won’t be able to deduct such tax in your Polish VAT return. Complete the company details in the billing settings.
Will AI replace accountants?
No. AI will take over part of the repetitive work, i.e. describing documents, matching payments and searching for unusual items. Responsibility for settlements, interpreting regulations and signing off the financial statements stays with people.
Does KSeF make it easier to implement AI in accounting?
Yes. Since 2026, invoices have reached companies in a single, structured format via the National e-Invoicing System, so automatically describing and posting documents is simpler and more accurate than with scans and PDFs, where the program first had to guess where the amount was and where the account number was.
Where should you start with AI in the finance department?
With one repetitive process, for example describing cost invoices, and with measuring how long it takes today. Then a pilot on a subset of documents, comparing the AI’s suggestions with the accountant’s decisions and an assessment after 60–90 days.
Sources
- SOURCEGartner: Finance AI Adoption Remains Steady in 2025 · 18 November 2025
- SOURCEIzba Administracji Skarbowej w Białymstoku: mandatory KSeF — deadlines · as of 24 September 2026
- SOURCEamavat.pl: KSeF deadlines and obligations 2026–2027 · updated 10 June 2026
- SOURCEPłatnikVAT.pl: How to account for an invoice from OpenAI / ChatGPT · updated 29 July 2026
- SOURCEOpenAI Help Center: ChatGPT Business — pricing and data policy · as of 24 September 2026
- SOURCEpodatki.gov.pl: Place of supply (taxation) — VAT guide · updated 15 December 2025, accessed 24 September 2026
Read next
Zaczynamy od mapy: skąd biorą się liczby dla zarządu i gdzie się rozjeżdżają. Potem wybieramy jeden proces na pilotaż, na przykład opis faktur kosztowych, i mierzymy punkt wyjścia, zanim cokolwiek zmienimy.
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.