AI in business: why most companies only use ChatGPT for emails, and where the real value is
Most companies use AI only to rephrase emails. What the numbers say, why it stalls at ChatGPT, three levels of AI use and how to bring it into real processes.
Vlado Pandžić · Founder · Senior .NET architect
Published · 4 min read
Ask your team whether they use AI. Almost everyone will say yes. Then ask what for. They rephrase an email, translate a text, summarise a document, ask ChatGPT what they used to search on Google.
That’s useful. But it’s a bit like buying a truck and only driving it to the bakery.
What the numbers say
According to Eurostat, in 2025 some form of artificial intelligence was used by 20% of EU companies with at least ten employees. Among large companies the share is 55%, among small ones 17%.
More interesting is why they don’t use it. Among companies that considered AI but didn’t adopt it:
- 71% say they have nobody with the expertise to do it
- 53% aren’t clear about what is allowed and what isn’t
- 49% are worried about data protection
- only 21% think AI wouldn’t be useful to them
So the problem isn’t that AI wouldn’t help. The problem is that nobody in the company knows how to bring it into the real work.
Why it all stalls at “ask ChatGPT”
- AI knows nothing about your company. It doesn’t know your customers, products, prices, contracts or procedures. Someone has to paste it all into the window every time.
- It can’t do anything in your systems. Someone still has to retype the answer into the ERP, the CRM or an email, which is the same problem as in the article on retyping data between systems.
- Everyone does it their own way. There are no rules, so company data ends up in public tools, and nobody knows which ones or where.
- Nobody measures it. Nobody knows whether it really saves time or just looks modern.
Three levels of using AI
| Level | What it looks like | How much it saves |
|---|---|---|
| 1. Chat for individuals | ChatGPT or Copilot for emails, translations and summaries | Minutes per person |
| 2. AI with your data | A question, and an answer from your own documents, contracts, manuals and knowledge base | Hours a week per department |
| 3. AI in the process | AI reads documents, emails and orders, sorts them and enters them into the system, and a person approves | Work that used to take someone all day |
Most companies stop at the first level. The real value is at the second and third, because there AI works with your data and inside your systems, not in a separate window.
What it looks like by department
- Sales. Customer enquiries arrive sorted, with a draft reply based on your price list and terms. The salesperson checks it and sends it.
- Customer support. Answers from your own knowledge base, with a link to the source, so you can see where the answer came from.
- Purchasing and administration. Data from supplier quotes, orders and delivery notes ends up in the system on its own, with no retyping.
- Management. Questions such as “how much of product X did we sell in the third quarter” over your own data, without waiting for a report.
- New employees. Internal manuals and procedures available through questions, instead of interrupting a colleague.
How to start
- One processLots of manual work, a clear measure
- RulesWhat may go out, what stays in
- PilotOn real data, not a demo
- MeasureTime saved, accuracy, cost
- One process. Not “we’re bringing AI into the company”, but one specific job with a lot of manual work, such as sorting enquiries or entering data from documents.
- Rules. What employees may put into public tools and what they may not. Company data can stay in your own Azure subscription instead of going to public services. Since February 2025 the EU AI Act requires companies to make sure the people who use AI know what they are doing anyway.
- A pilot on real data. A demo always works. The real question is whether it works on your documents, with your exceptions.
- Measure. How much time was saved, how accurate the AI is and what it costs per document or enquiry. Until it has proven itself, a person approves every result.
Once the first process works and shows the numbers, it’s easy to decide on the next one. How we build this in .NET applications is in the article on adding AI to an existing .NET application.
How we work
This is exactly what we do: we bring AI into companies’ business processes, not as yet another chat window, but connected to your data and systems, inside your own Azure subscription. We start with one process, measure the savings on real data, and only then expand. The first step is a free 30-minute call in which we go through your processes and see where AI would save the most.
Sources
This article is general information only, not legal, tax, financial or other professional advice. Scenarios, examples and calculations are illustrative. Terms of use and disclaimer.