credX

credX is a digital financing platform that facilitates commercial real estate lending by connecting qualified borrowers with lenders.

Challenge

For an efficient lending process, the financing parameters must be presented in a trustworthy and standardized manner to the lenders. This requires significant effort from Real Estate Analysts - making it both expensive and challenging to scale the platform.

• Complexity: Each transaction requires analysis of 50+ documents with 1,000+ pages in different formats resulting in 200+ data fields and 10+ individual explanations
• Accuracy demands: Analysts needed a nearly 100% accurate tool with human oversight—even advanced systems achieve less than 80% on similar complex tasks
• Assembly line, not a toolbox: The team needed an end-to-end workflow that handles complex data rooms and generates complete memos
• Analytical skills: Connecting the dots across disparate real-estate documents requires deep domain expertise

‍

credX found traditional intelligent document processing (IDP) tools didn't meet their standards. They required a custom Gen AI solution that could deliver the correctness and completeness required for production-grade use cases.

Solution

credX partnered with deepset, leveraging their AI expertise and platform to build an IDP solution tailored to their business needs:

• Developed a multi-stage multimodal document processing pipeline that combines Optical Character Recognition (OCR) with foundation models (language and vision)
• Enhanced performance through modular integration of AI models and components using Haystack, deepset's open-source framework for AI Orchestration
• Transformed manual document review processes by expediting the analysis of 1,000+ page document sets, answering 100+ standardized questions, and delivering a complete data sheet with annotations and references

Impact

credX successfully launched an AI solution in a fraction of the time it would have taken to build in-house, while achieving significant cost savings. The system provides exact references for each memo – a level of detail that was not cost-effective before. Key results include:

• Efficiency: 40% of analysts’ hours saved for complete memo generation
• Speed: Approx. 80% time reduction for due diligence document screening and preliminary analysis (from 1 day to 1-2 hours)
• Quality: 99.5% correct answers across analyses with correct references attached
• Relevance: Detects 95% of all relevant documents. Advanced prompting techniques with overlapping questions pushed it to 99.75+%
• Revenue: credX has created a new SaaS revenue stream by licensing the tool to commercial real estate players and to banks/credit funds

"At first, our analysts were skeptical," says Ralf Kauther, CEO of credX. “But they soon realized AI can relieve them from reading and memorizing tons of documents, leaving them free to apply their deep expertise where it matters most.”