Enterprise AI & Strategy · Process analysis
The Future of Enterprise AI Starts Within Your Processes
A language model answers questions. A process asks for more: exceptions, internal rules, data from several systems and someone who signs off at the end. AI delivers the most value where it is built for exactly that workflow. The way there starts with analysing the process.
Julia Rose
Key takeaways
- According to a 2025 IW survey, 37% of companies in Germany use AI, mostly through free tools. Only 3.6% develop AI themselves.
- The study concludes that custom and in-house solutions are needed to unlock the full potential of AI.
- According to McKinsey, AI high performers are nearly three times as likely to have fundamentally redesigned individual workflows.
- A process analysis clarifies decisions, data, exceptions and systems before a model is chosen.
Why AI without context hits its limits
General AI tools write, summarise and answer questions remarkably well. In daily business, though, a result depends on things no general model knows: exceptions and special cases, internal terminology, rules nobody wrote down, and data that sits in ERP, CRM or document storage.
Most companies are still early in their use of AI. The German Economic Institute (IW) surveyed more than 1,000 companies in Germany in 2025:
| IW survey finding | Share |
|---|---|
| Companies using AI | 37.0% |
| use free AI tools | 29.3% |
| buy AI applications from other companies | 13.0% |
| develop AI themselves | 3.6% |
Source: IW-Report, Wettbewerbsfaktor Künstliche Intelligenz, July 2025 (German)
Free tools are a good way to gain first experience. The study’s authors still draw a clear line: custom and in-house AI solutions are needed to unlock the full potential of AI.
What AI leaders do differently
McKinsey’s State of AI 2025 paints a similar picture. The organisations that see measurable results from AI rarely layer it on top of existing processes. They are nearly three times as likely to have fundamentally redesigned individual workflows as other organisations.
Source: McKinsey, The State of AI in 2025
That matches our experience. A strong demo model is quick to build. A system that reliably takes work off people’s hands every day only emerges once it is clear how documents are structured, how exceptions are handled and which rules apply.
What a process analysis clarifies
The most successful AI projects start with the process. These questions belong at the beginning:
- Decisions: which decisions are made in the workflow, and which of them can a system prepare or take over?
- Data: which information is available, in which format and in which systems?
- Variants and exceptions: which cases are typical, which are rare, and what happens when something is missing?
- Systems: where does the result need to go, for example ERP, CRM or the ticketing system?
- Control: where does a person review, and how does the system recognise that it is uncertain?
- Value: how much time and effort does the process cost today, and how will success be measured?
The answers show whether AI pays off, which use case comes first and which architecture fits. Sometimes the result is that simple automation is enough.
From process to AI system
Production enterprise AI rarely consists of a single model. It combines several building blocks, each taking over one task in the process:
| Building block | Task in the process |
|---|---|
| Models | read, classify, extract and write |
| Knowledge modules | answer from company documents with the source, for example via RAG |
| Agents | plan steps and call tools |
| Integration | fetch data from existing systems and write results back |
| Control | check results, route uncertain cases to people and keep a log |
How several agents work together and connect to existing systems is described in our article on multi-agent systems.
In practice
In practice: Fr. Meyer’s Sohn
The logistics company processes thousands of emails a day with shipment, routing and planning data, in German and English and in no fixed format. Rule-based extraction failed on the variety. A GPT-based pipeline on the client’s own servers cuts manual extraction effort by 80%.
In practice: Radaway
The bathroom equipment manufacturer already had a first LLM system for orders from customer emails. In three weeks we developed it into a solution ready for production: 90% less manual intervention and more than 95% accuracy in product matching.
In both projects the key was the process logic: what the emails really look like, which details can be missing and how the result gets into the existing systems.
What production-ready AI systems have in common
- They are tied to reality: they work with the company’s real data, exceptions and rules.
- They are traceable: results can be checked, and uncertain cases go to people.
- They can grow: further steps and use cases build on the same architecture.
Our article How companies implement AI successfully shows how the analysis turns into a structured adoption process.
Sources
- McKinsey & Company: The State of AI in 2025: Agents, innovation, and transformation
- Engels B, Scheufen M, Schmitz E: Wettbewerbsfaktor Künstliche Intelligenz. Empirische Befunde und Handlungsempfehlungen zum Einsatz von KI in deutschen Unternehmen. IW-Report, Cologne, 4 July 2025 (German). PDF
Which process costs you the most?
In the process analysis we map your workflows, rank the use cases by value and propose an architecture that fits your systems.
Request a process analysisFrequently asked questions
Because processes depend on exceptions, internal rules and data from several systems. A general model does not know this context and cannot write results back into existing systems.
Which decisions are made in the workflow, which data is available, which variants and exceptions exist, where results go, where people review and how success is measured.
According to a 2025 survey by the German Economic Institute, 37.0% of companies use AI, but only 3.6% develop it themselves. Most rely on free tools.
At theBlue.ai the analysis is available at a fixed price from €3k. Implementation typically runs between €25k and €100k, with a scoped estimate at the end of the analysis.
Julia has been part of theBlue.ai since 2019 and has accompanied the development of AI applications in the enterprise environment since the company’s early days. In her role as Marketing Lead, she works closely with the engineering and consulting teams and makes complex technical topics understandable and accessible for decision-makers.
In her articles, she writes about practical experience from enterprise AI projects, as well as the challenges and opportunities of using AI in companies.