Fuzzy Search for Documents

Deloitte’s Fact Finder delivers relevant and comprehensive search results - fast

Deloitte’s Fact Finder employs advanced text-mining functions optimized for semantic search on individual or large volumes of documents. It finds not only exact matches, but is trained to also identify related concepts, ensuring nothing relevant escapes you.

The Need

What do contracts, terms & conditions, service level agreements, warranties, and many other legal documents have in common?  The devil is in the detail. Conditions and obligations are often buried – sometimes even intentionally obscured – within volumes of text. These details may be small, yet they can represent significant, latent risk. Multi-year insurance policies may implicitly cover minor risks  at inception, which have since grown substantially (for example, cyber risks). Or there could be a misalignment  between primary insurance and reinsurance coverage conditions. Business loans or credit lines may have a litany of covenants tied to them (generally well known) as well as exceptions (less well known). Manufacturers with complex supply chains contending with sanctions or embargos or auditors seeking clues – whatever the case, the search for relevant information can be daunting in the face of the sheer volume of documentation produced today. The wide range of wording and structural possibilities stymie simple automation through keyword  search strings. Experts rightly avoid such methods, which require exact matches to wording, spelling and punctuation – without which they systematically overlook important passages.

Here you can download the Fact Finder fact sheet:

Our Solution: Deloitte Fact Finder

Deloitte Fact Finder utilizes advanced Natural Language Processing (NLP) techniques to overcome these limitations. The simple, intuitive interface gives users access to powerful  text-mining functions without requiring any coding. Users follow a linear workflow: 

  1. specifying search terms (concepts)
  2. uploading (multiple) documents for review
  3. triggering (bulk) processing
  4. viewing and downloading results  

As inputs, Fact Finder requires only sample search terms. It automatically converts these into similar concepts – no additional input is required. It provides visual feedback of search results by highlighting relevant passages directly on rendered PDFs. Search results can be exported, either in the form of the marked-up PDFs or into a table citing the relevant passage, the associated document, and the page location within that document.


  • Speed – results in hours vs. days or weeks
  • Quality – less likely that critical details will be overlooked
  • Cost – time and quality gains  from manual reading, mark-up, data transfer
  • Simplicity – no training on documents required
  • Security – no dependence on Hyperscaler Cloud services

Example Use Cases

  • Search for contentious contract terms / liabilities / latent risks
  • Identify unfavorable counterparty risks (embargo or Know-Your-Customer)
  • Extract rules or conventions embedded in policy documents (e.g., minimum password syntax & length)
  • Quickly identify topical relevance to documents, even serving as a preliminary step to labeling a classifier

Ihr Ansprechpartner

David Thogmartin

David Thogmartin

aiStudio | AI & Data Analytics

David Thogmartin leads the aiStudio internationally and the “AI & Data Analytics” practice for Risk Advisory in Germany. He has 20 years of professional experience in Analytics and Digitization, large... Mehr