Analysis
AI business transformation: how AI changes the way you see your company — and what it can realistically deliver
AI business transformation is less about a new tool than a new way of seeing the company: processes become tasks with a checkable outcome, data becomes an asset, decisions become hypotheses to test, and labour cost becomes usage cost. In studies with real workers, AI assistance raised customer-support productivity by 15% on average and cut the time spent on professional writing by 40%. The same research shows the limits: outside what AI does well, the share of correct solutions fell.678
Most conversations about AI in a company start with the question of which tool to buy. Understandable — but often a dead end: a tool bought without changing the process achieves little. The real shift happens elsewhere — in how leadership and the team see their own business.
AI pays off most not where a company buys a new tool, but where it re-describes its work: which decisions a person makes, which steps a system can take over, and how we will know it works. Technology is the easiest part of an implementation. The harder and more important part is getting processes and data in order.
— Dariusz Wtykło, Owner and CEO of Primessio
Four shifts in how you see your company
Processes: from task lists to work with a checkable outcome
A process described as “who clicks what” is hard to hand over to AI. The same process described as a task with an input, an expected outcome and a way to check it — that works. Based on interviews with organisations deploying AI agents, the OECD names three conditions: the task is structured enough to be planned, the outcome can be checked against reliable data or system records, and the cost of an error is bounded or reversible9. It is a useful yardstick for reviewing any company’s processes.
Data: from archive to a working asset
AI works on company data — documents, tickets, order history. That is why cost shifts from licences to data quality: in a Bitkom survey of German companies using AI, 50% named data preparation as a major cost item, 51% infrastructure and 41% integration with existing systems10. A company that treats data as an asset gains whichever model it chooses.
Decisions: from gut feeling to testable hypotheses
AI quickly produces options, summaries and recommendations. The decision stays with people, but its nature changes: instead of one gut call, leadership gets several reasoned options that need to be checked. That takes discipline — in McKinsey’s global survey only 27% of respondents at organisations using generative AI say employees review all AI-generated content before it is used12.
Cost model: from headcount and licences to usage cost
The cost of the model itself is falling fast: querying a model at GPT-3.5 level cost $20 per million tokens in November 2022 and $0.07 in October 202411. Routine work that used to be a fixed cost becomes a variable cost that scales with the number of tasks. Our analysis of AI implementation costs, linked at the end of this article, shows how to calculate it.
What AI can realistically deliver — what the research shows
The strongest evidence comes from studies in which some workers had access to AI and others did not. Each covers a specific task, not a whole company:
| Study | Task | Result |
|---|---|---|
| Brynjolfsson, Li, Raymond, Quarterly Journal of Economics, 2025 | customer support | +15% issues resolved per hour on average, with large differences between workers6 |
| Noy, Zhang, Science, 2023 | professional writing | time −40%, quality +18%7 |
| Dell’Acqua et al., Harvard Business School, 2023 | consulting tasks inside AI’s capabilities | +12.2% tasks completed, 25.1% faster8 |
| same study | task outside AI’s capabilities | 19% less likely to produce a correct solution8 |
The lesson is not “AI speeds everything up”. It is: AI clearly helps with tasks it handles well, and hurts where people accept its answers uncritically. The boundary is not obvious — which is why the researchers called it “jagged”8. Companies that understand this map the boundary in their own processes first, and only then scale.
Where companies stand today in the EU, Germany and Poland
| Area | Enterprises using AI in 2025 |
|---|---|
| European Union (10+ employees) | 20.0%1 |
| Germany | 26%2 |
| Poland | 8.7%3 |
| Poland, large enterprises | 42.0%3 |
In McKinsey’s global survey, cited by the Stanford AI Index, 88% of respondents report using AI in at least one business function4 — a different population (mostly large organisations) and method, so we do not set it next to official statistics. The most common reason EU enterprises give for not using AI is lack of relevant expertise (70.3%), followed by unclear legal consequences (53.6%) and data protection concerns (52.7%)5. These barriers can be addressed with method, not only with budget.
What to avoid
- A project without a problem to solve. On average only 48% of AI projects make it into production, and moving from prototype takes 8 months13. RAND lists the most common causes as a misunderstood problem, missing data and a focus on the latest technology rather than the real problem14.
- Tasks outside AI’s capabilities. There the share of correct solutions fell, because people trusted answers that sounded convincing8.
- Content and decisions without human review. Responsibility for the result stays with the company, whoever prepared it.
- Fully automated decisions about people. The Dutch data protection authority ruled that Uber violated the prohibition of fully automated decision-making and did not inform drivers sufficiently15.
Where to start
Start with a process map, not with choosing a model. Pick one task that is repetitive, has a checkable outcome and a reversible error. Decide how you will recognise success before you build anything. Sort out data and permissions, plan human oversight — and only then scale to further processes.
The Primessio AI Implementation Framework describes this path step by step — a method developed by Primessio in project practice, from process mapping to operations. If you would like to take it with a team that does the work alongside you, see AI implementation and automation services.
Frequently asked questions
How is AI changing the way businesses are run?
It shifts attention from tools to processes, data and decisions. Instead of asking which software to buy, a company asks which tasks have a checkable outcome and can move to AI, and which decisions stay with people. The OECD notes that AI works where outcomes can be checked against reliable data and errors are bounded or reversible.9
What are the realistic benefits of AI in business?
Research measures effects on specific tasks, not whole companies: in customer support, 15% more issues resolved per hour on average; in professional writing, 40% less time with higher quality. Results depend on the process, the data and the team — so we do not promise numbers, we measure them after implementation.67
How many companies use AI in Europe?
According to Eurostat, 20.0% of EU enterprises with 10 or more employees used AI in 2025. In Germany the figure was 26% and in Poland 8.7%, rising to 42.0% among large enterprises.123
Where should a company start with AI?
With a process map and a single task whose outcome can be checked — not with choosing a model. RAND finds that AI projects most often fail when teams misunderstand the problem, lack the necessary data or focus on technology rather than the problem. The Primessio AI Implementation Framework walks through this step by step.14
What should you avoid when implementing AI?
Three things: tasks outside what AI does well (there the share of correct solutions fell), content and decisions without human review (only 27% of respondents say employees review all AI-generated content before use), and fully automated decisions about people without a legal basis — the Dutch data protection authority sanctioned Uber for exactly that.81215
Sources
- Eurostat: 20% of EU enterprises use AI technologies
- Destatis: Unternehmen mit Nutzung von Technologien der künstlichen Intelligenz nach Beschäftigtengrößenklassen
- GUS: Społeczeństwo informacyjne w Polsce w 2025 r.
- Stanford HAI: AI Index Report 2026 — Chapter 4: Economy
- Eurostat: The use of artificial intelligence in the European Union — Key results
- The Quarterly Journal of Economics: Generative AI at Work (Brynjolfsson, Li, Raymond)
- Science: Experimental evidence on the productivity effects of generative artificial intelligence (Noy, Zhang)
- Harvard Business School / SSRN: Navigating the Jagged Technological Frontier (Dell’Acqua et al.)
- OECD: Agentic AI in organisations: Early insights from practitioner interviews (OECD AI Papers No. 65)
- Bitkom: Erstmals nutzt die Mehrheit der Unternehmen KI
- Stanford HAI: The 2025 AI Index Report — Research and Development
- McKinsey & Company: The state of AI: How organizations are rewiring to capture value
- Gartner: Gartner Survey Finds Generative AI Is Now the Most Frequently Deployed AI Solution in Organizations
- RAND Corporation: The Root Causes of Failure for Artificial Intelligence Projects and How They Can Succeed
- European Data Protection Board: Dutch DPA fines Uber EUR 824 990 000 for unlawful automated decision-making and insufficient transparency