Management Reporting, Job Adverts, Onboarding: AI in the Administration of Mid-Sized Companies

The monthly report takes several days and is finished when the numbers are already out of date. Vacancies stay unfilled for a long time, and the job adverts get written on the side. New colleagues need months to find their feet, and until then they tie up the experienced ones. Three routines that run in every company with fifty or more people, and which hardly anyone looks at as a process.

We have deliberately kept this group small. In machinery manufacturing, administration is not where AI makes the difference, and one of the three applications falls into the high-risk category under the EU AI Act. They still belong in the library, because the question comes up in almost every conversation.

Where the time goes

In the reference company with a hundred employees, the three activities account for 22 hours a week, spread across management, controlling and HR. The report is the largest item and the application with the shortest payback.

Three applications, costed

Automating management reporting

The report is built from ERP and BI, including commentary on the variances. The numbers stay the same, they just arrive three days earlier, and the commentary is the part that nobody used to write.

In the model company: 10 hours a week, 40 percent of which can be automated, €15,300 a year. Setup €3,500, running costs €300 a month, payback after 3.6 months. Implementation: builds on the shared knowledge platform, 3 to 6 weeks. Tools: Microsoft 365 Copilot, Power BI Copilot, Claude. Data protection: not critical, internal data with no personal data involved.

Job adverts and initial screening

Adverts are drafted from the role profile, and incoming applications are pre-sorted by professional fit. Pre-sorting yes, automatic rejection never. Under the EU AI Act, candidate selection is a high-risk case: the final human decision and the documentation are mandatory, not a recommendation.

In the model company: 6 hours a week, 40 percent of which can be automated, €7,020 a year. Setup €4,000, running costs €90 a month, payback after 8.1 months. Implementation: licence and training, no infrastructure of your own, 2 to 4 weeks. Tools: Microsoft 365 Copilot, Personio, Claude. Data protection: high-risk under the EU AI Act, the final human decision and documentation are mandatory.

Supporting onboarding

An assistant answers the common questions of the first few weeks from the company’s own knowledge base. It builds on the knowledge platform and therefore only makes sense if that is already in place.

In the model company: 6 hours a week, 35 percent of which can be automated, €6,142 a year. Setup €3,500, running costs €210 a month, payback after 11.6 months. Implementation: builds on the shared knowledge platform, 3 to 6 weeks. Tools: Cognigy, Nextcloud Assistant, Azure AI Search, FORGE Document Intelligence (our own product). Data protection: not critical, internal data with no personal data involved.

Where to start

Reporting first: paid back in under four months and immediately noticeable for the management team, though as an add-on to the knowledge platform. Without the platform, the only option left in this group is the job advert, and only with a clearly defined human decision and proper documentation. Otherwise, better not. Onboarding comes as the last add-on, once the platform already exists from documentation or service. The honest summary: in machinery manufacturing, administration is an add-on, not a starting point.

Application	Hours per week	Savings per year	Setup	Payback
Automating management reporting	10	€15,300	€3,500	3.6 months
Job adverts and initial screening	6	€7,020	€4,000	8.1 months
Supporting onboarding	6	€6,142	€3,500	11.6 months

How the figures are calculated

All figures come from the calculation model of the use case library, not from the survey. Reference company with 100 employees, 45 productive weeks a year, fully loaded costs of €65 an hour for commercial staff and €85 for technical staff, licences at €45 per user per month. The model counts only working time saved against the full costs, and the savings rates are deliberately set at the lower end of what is known from projects and studies. No quality gains, no avoided cost of errors, no effects on lead time. Replace the assumptions with your own figures and the model calculation becomes your calculation.

Further reading

The full calculation for all thirty applications is in the Use Case Library for Machinery Manufacturing, and the managing directors’ answers are in the survey results (in German).

Where is your biggest lever? Let’s talk it through briefly, no obligation.