Structure first, then AI

How offices can put analytical AI to work on the data they already have

AIData
Thursday, October 1, 2026

Our last article was about the risks of generative AI. This one is about the quieter side: analytical AI. It doesn't write texts; it answers questions about your own work. Which cases will take longer than usual? When will the next wave of requests roll in? Which request actually belongs with a different team?

Answers like these can be checked and repeated, and they're often worth more than any generated text. They have just one prerequisite, and it's where many projects fail: data in a usable form.

Generative or analytical? The difference in one paragraph

Generative AI creates something new: texts, images, drafts. Analytical AI recognizes patterns in what's already there. It classifies, predicts and spots anomalies. Its results can be measured against reality: a forecast comes true or it doesn't, a classification is right or it isn't. That often makes it more reliable for everyday office work – provided the data is right.

The problem rarely lies with the AI

In early 2025, Gartner predicted that through 2026, organizations will abandon 60 percent of AI projects that aren't supported by AI-ready data. According to the survey, 63 percent of organizations either didn't have the right data management practices for AI or weren't sure whether they did.

In the office, it looks like this: requests are stuck as free text in e-mails, contracts as PDFs in folders, deadlines in spreadsheets, case status in people's heads. The data is there, but scattered, duplicated and in a form no software can work with. “We type the same data three times into three systems,” we hear again and again.

Three ways to more structure

Capture it in a structured way instead of cleaning up later

The simplest path to good data is to capture it correctly from the start. A form that asks the right questions delivers structured data instead of free text from the very first moment: who, what, by when and with which documents. With FormAux, this data arrives as JSON and can be processed right away.

Process data is created along the way

When every step of a case runs as a task in the system, timestamps, responsibilities and sequences appear on their own. That shows how long a case takes and where it waits. Values that have been entered once are reused by TaskAux in every following step, because their origin is known.

Make documents searchable

An office's biggest treasure trove of data lies in its documents. A full-text search across all machine-readable files, as DocAux offers, is the first step. The next is semantic indexing: documents are additionally stored as so-called embeddings that capture their meaning. Search then finds what fits in terms of content, even without the right keyword. That's exactly what we're working on with SrxAux, which will be available soon.

Analytical AI in practice

Once the order is in place, analytical AI becomes tangible:

  • Classification: documents, tasks and messages are tagged automatically. That's what we're currently developing TagAux for, a small module that will work inside DocAux, TaskAux and MsgAux.
  • Forecasts: which cases will exceed the usual lead time? How much work will land on the team next week?
  • Automation candidates: which steps come up often and take a long time?
  • Misrouting: which requests regularly end up with the wrong team?

Why running it locally makes the difference

Analytical AI needs access to exactly the data an office has to protect most carefully. That's why it should run where this data lives: on your own infrastructure or on a dedicated server in Germany, not scattered across third-party platforms. Whoever runs the models themselves keeps control over what happens to the data.

A realistic roadmap

  1. Organize: capture requests in a structured way, map cases as chains, gather documents in one place.
  2. Measure: let it run for a few weeks and see which numbers emerge. These numbers alone answer many questions, without any AI.
  3. Analyze: only now does analytical AI come into play, with questions that grew out of the numbers.

The beauty of this order: every step pays off on its own. Even if you never get to step three, you'll end up with a better-organized office.

How we help

With FormAux, requests arrive in a structured way. With TaskAux, cases become chains whose data is created along the way. And DocAux brings documents together in one place, with full-text search across all machine-readable files. TagAux, SrxAux and StatAux, our statistics module for the whole suite, build on this and are in development.

AI isn't where digitalization begins. It's where it leads. If you'd like to know where your office stands on this path, get in touch.

Sources