Tier im Recht (TIR) is a Swiss animal welfare organisation that closely monitors how animal welfare legislation is applied in practice. Each year, TIR receives, via the Federal Food Safety and Veterinary Office (FSVO), all animal welfare criminal-law decisions from across Switzerland that the cantons report to the federal government. TIR then maintains a nationwide database of these cases. Based on this information, TIR publishes regular analyses focusing on enforcement practices, the categories of animals affected and the level of penalties imposed for individual violations.

However, processing these criminal-law decisions is demanding. Around 2,000 animal welfare cases are prosecuted in Switzerland every year. The corresponding  decisions come in 26 different formats and multiple official languages (including German, French, and Italian),  and are delivered to TIR as scanned, anonymised PDFs of varying OCR quality. Manually copying and pasting this information into TIR’s nationwide database is slow, repetitive, and time-consuming work. Before working with Archipanion, TIR’s staff had to open each decision document and manually fill in a database form, field by field. TIR wanted to find a way to reduce this repetitive work and refocus staff time on data analysis and advocacy.

To address this challenge, Archipanion developed and piloted a solution that combined an AI-supported data extraction workflow with an internal review dashboard. The workflow automatically extracted the required case information from each criminal-law decision document, while the dashboard allowed staff to verify, correct, and add to the extracted data. In this pilot phase, a subset of criminal-law decisions was processed to test if AI could extract the required data fields with sufficient accuracy.

Following the successful pilot, Archipanion and TIR moved from testing to an operational workflow built around three key components:

1. AI-supported data extraction. This process included defining a clear set of database fields that mirrored TIR's existing animal welfare criminal case database structure (for example canton, court, decision date, offence category, animal category, and relevant law articles). In addition to these structured fields, AI was also asked to generate short summaries of the case facts and the criminal-law decision outcome. The team specified what information AI should extract or summarise from each decision. As a result, case information was automatically extracted from the decision documents and pre-filled in the quality review dashboard, ready for checking before import into TIR's animal welfare online criminal case database.

2. Expert quality review. TIR staff start their work in the quality review dashboard, where the database fields for each case are pre-filled by AI, saving significant staff time and resources. The pre-filled fields are displayed alongside the original decision document, so reviewers can move through the document and the database in parallel. Some fields remain empty by design and are reserved for TIR's expert judgement. Reviewers focus on checking that AI has extracted the correct information and on completing the expert-only fields that AI cannot reliably derive on its own. Once a case has been reviewed, it is marked as complete and the data is exported to TIR's central database, which supports both the analyses and the public-facing case search on TIR's website.

3. AI accuracy review. Because every TIR staff edit is tracked within the quality review dashboard, the Archipanion team can measure where AI extraction is consistently accurate and which fields still require more frequent corrections. These insights support ongoing technical improvements and help identify where AI-supported data extraction delivers the greatest benefit.

Screenshot of the quality review interface showing expert verification before export

With the AI-supported workflow in place, TIR now has:

• A significant reduction in manual data entry. Staff no longer start from an empty form. Instead, they review and refine AI-extracted case data, redirecting their time from repetitive data entry to quality review.

• Structured, consistent data across thousands of cases. Automated extraction ensures that case data continues to be captured in a consistent way, even though the underlying documents differ by canton, layout, and language.

• More staff time for advocacy and analysis. With much of the manual data entry work handled by AI, TIR can focus more of its staff time on analysing and interpreting the data.

Although this service focuses on animal welfare criminal-law decisions, the same pattern – AI-supported data extraction combined with an expert review dashboard – can be applied to any organisation that needs to turn large volumes of complex legal or regulatory documents into structured, analysable data. In the longer term, there is also potential to enable new ways of working with this structured data, such as:

• Chat-based analysis over the structured case data, allowing TIR staff to ask questions like “Show recent cases involving horses in rural regions” or “Summarise trends in sanctions for specific offence types”. 

• Support for drafting report sections, where AI could generate a first draft report on topics such as particular animal categories or regional developments, always subject to expert review.

If implemented, these developments could make the data easier to analyse and communicate, while ensuring that animal welfare expertise remains central to the process.

If your organisation works with large volumes of complex documents - whether court decisions, administrative records, or other structured files - AI-supported data extraction can transform how you search, analyse, and use your data.