AI in business processes: process digitalisation that relieves instead of keeping you busy

Fastest where people currently spend a lot of time on unstructured information: reading and answering enquiries, summarising documents, preparing quotations, writing reports, transferring data from emails and PDFs into systems, finding knowledge in manuals and instructions. In manufacturing companies, this concerns order clarification, purchasing, technical documentation, maintenance, quality and customer service. AI takes over the draft, the search or the pre-structuring, and the person checks and decides. The benefit is time freed up for value creation, and fewer errors caused by retyping. A use case analysis of your processes shows exactly where the biggest lever lies.

In five steps. First, check readiness, meaning data quality, systems, skills and culture; second, identify use cases, prioritise them by benefit, data situation, risk and effort, and select two or three pilots. Third, map, streamline and standardise the processes concerned before technology comes into play; fourth, train managers and teams and set rules for the handling of data and tools. Fifth, implement the pilots, support them, measure them, and only then roll out. The most common mistakes are starting with the tool instead of the process, and bypassing the people who are supposed to work with it.

Small, concrete and with rules. Choose a process that noticeably costs time and carries little risk, such as answering recurring enquiries, summarising reports or preparing quotations. Define which tools are permitted and which data may go in; without this rule, shadow AI emerges. Train the two to five people who will work with it, and after four weeks measure how much time was saved and what the quality is like. Only then move on to the next process. An AI workshop for senior management beforehand helps to assess the potential and limits realistically.

Rarely because of the technology. Usually because of four things: the process to be supported is not standardised, so the AI produces variants of the chaos, and the data is incomplete or scattered, so the tool delivers unusable results. The employees were not involved, see the AI as a threat or a gimmick, and do not use it. And there are no rules and no one responsible, so the pilot fizzles out as soon as the initiator turns to something else. All four causes are process and change issues, not technical ones; that is why I treat AI introduction as a Lean and change project.

Because without rules, two things happen: either employees secretly use AI tools with company data, with all the risks for data protection, trade secrets and quality. Or, out of uncertainty, they do not use them at all, and the potential is left untapped. AI governance defines which tools are permitted, which data may and may not go in, who checks results, how use cases are approved and who is responsible. For SMEs and mid-sized companies, a few pages and one responsible person are enough; what matters is that the rules are known and lived. Governance is the prerequisite for employees to be able to use AI safely.

This can be answered in five dimensions. Processes: are the workflows to be supported documented and standardised, and is the necessary data available digitally, completely and in one system, or scattered across emails and Excel? Systems: can your applications be connected, or are they closed? Skills: can managers assess AI use cases, and are there employees who can use the tools? Culture: are mistakes and experiments allowed, or is every innovation put on hold? I assess this AI readiness in the assessment area IT, Systems and Digitalisation of my Maturity Check and derive from it where you should start.

In the administrative environment: automatic capture and clarification of orders from emails and PDFs, quotation drafts from enquiries, summaries of supplier correspondence, searching technical documentation and work instructions by asking a question instead of browsing a folder structure, creating maintenance and quality reports from keywords, translating instructions for multilingual workforces. In planning: support for capacity and scheduling, detection of patterns in breakdown data. Closer to the machine: predictive maintenance and image-based inspection in quality, which however require specialists and good sensor data. For getting started, I recommend the administrative cases: low risk, fast impact, no new hardware.

An AI agency or AI service provider develops and implements solutions: they build the application, integrate it into your systems and operate it. AI consulting, as I offer it, works before and alongside the technology: it clarifies which processes are suitable, prepares these processes, defines rules, trains managers and teams and guides the introduction so that the solution is actually used. Both are needed, and both should not come from the same source, because a vendor who sells the solution cannot select the use cases neutrally. I work together with implementation partners and remain your neutral point of contact for process and change.