Public Sector & Education

The registry office has its case system, the treasury its budget system, the university its campus management. Each produces its own reports, and a view across offices, sites or years hardly exists. Figures for committees are often put together by hand.

We build a shared data foundation for this: fixed metric definitions, access rights by role and data flows you can trace. Business intelligence is the core of the work. AI comes in where documents and inquiries are involved, always with a review step by a person.

What does “processing time” actually mean?

Before a dashboard exists, it has to be clear what is being measured. Processing time can start at receipt of the letter, at the complete application or at assignment to a caseworker, and every procedure counts slightly differently. Without a shared definition you end up comparing counting rules instead of offices. The same goes for backlogs, workload and the share of applications that need additional documents.

So we fix the metrics first, together with the departments that will later be measured by them. Only then does it make sense to refresh the reports for committees, controlling and oversight automatically instead of reassembling them every quarter.

Connecting case systems without touching them

Case systems for registration, building permits, social services or administrative offences often come from different vendors, each with its own data model. We change nothing in them. Regular extracts through interfaces or export files go into the data warehouse, and the analysis runs there, not in the case system. Budget systems, electronic records, application portals and, in education, school administration systems come in the same way.

The hard part is rarely technology. Free-text fields, legacy records and missing mandatory fields only show up once someone analyzes the data, and then they show up in the report. Checks in the data pipeline make such gaps visible early. We also plan access rights from the start: a dashboard with individual case data does not belong in every hand, and adding role-based permissions afterwards is tedious.

Budgets, school places and online applications

In budget execution, the current balance matters less than the question of where funds will go unspent by year-end or where overruns are coming. Plan, actual and commitments sit in the budget system, and the outlook to year-end comes from the analysis.

For demand planning we add forecasts for student numbers, daycare places or incoming applications, derived from population data, registrations and birth and migration figures. They are decision aids, not instructions. Whether a school site is expanded is decided by professionals and politicians, not by a model.

Germany’s Online Access Act (OZG 2.0, in force since July 24, 2024) establishes the once-only principle, among other things. The more services move online, the easier it becomes to see where applicants drop out and which form triggers follow-up questions. Privacy-conscious measurement is enough for that. The same sources can also feed datasets for open data portals. In the EU, the Open Data Directive governs reuse of public sector data, implemented in Germany by the Datennutzungsgesetz, and an implementing regulation defines high-value datasets such as geodata and statistics.

Where AI helps in administration, and where it does not

AI is most useful with text. Incoming letters and applications can be read and assigned to a case type and responsible unit. An assistant can answer questions from statutes, leaflets and internal guidelines and name its source, based on a knowledge base with access rights.

In both cases the output is a draft. A person confirms or corrects it, and the decision stays with the responsible unit. Without that step, errors enter the procedure and are hard to trace later. AI should not decide alone on individuals’ entitlements or rights, legally or practically.

The EU AI Act draws a line here. Annex III lists, among others, systems for admission, evaluation of learning outcomes and exam proctoring in education, as well as systems that help decide on entitlement to public benefits. Under the AI Omnibus, these high-risk obligations apply from December 2, 2027, and public bodies that deploy such systems must also carry out a fundamental rights impact assessment. Transparency duties for chat assistants have applied since August 2026. More on our EU AI Act page.

Cloud, procurement and data protection in the public sector

What is permitted depends on the level of government and how sensitive the data is, and federal, state and local bodies sometimes have their own rules. In Germany, the Administrative Cloud Strategy of the IT Planning Council aims for choice and the ability to switch between cloud offerings, and the federal government is working on the Deutschland-Stack. We plan platforms so that they can run in the EU or in your own data center.

Procurement is where this pays off. If open formats, documented interfaces and the transferability of results are written into the specification, you can change contractors later. Closed formats and missing interfaces usually only hurt at that point.

On data protection, state data protection and school laws apply alongside the GDPR, and data on children and young people is particularly sensitive. For planning and steering, aggregated or pseudonymized data is usually enough. Sign-off in each case belongs with data protection, information security and legal counsel.