Digitalisation and Artificial Intelligence (AI) are accelerating transformation in the fund industry and have the potential to bring about comprehensive changes to processes in modern fund management. Against this background, it is strategically important that we at Deka identify specific areas of application.
AI Use Case for Efficient Data Structuring
Our current practical example, “Optimising Product Selection for Real Estate Fund of Funds Management” is another building block in our overarching digitalisation and AI development process. We use it to reduce routine tasks, provide consistent data sets, and thus strengthen our foundation for well-founded investment decisions. In this context, the AI use case focused on increasing efficiency in fund of funds management. Together with the Technical University of Applied Sciences Mannheim (TH Mannheim), we investigated how AI can be used to evaluate fund prospectuses more quickly in fund of funds management. The data we obtained needed to be provided in a uniformly structured manner.
The most challenging aspect of the project was in the details. In Germany alone, there are several hundred real estate funds from various different providers that are currently approved for distribution. Each issuers’ prospectus uses its own structures, terminology, and presentation methods. For well-founded investment decisions, however, our fund of funds managers need consistent and comparable information about the funds offered. Previously, this data had to be manually extracted and transferred over to internal systems.
To ensure comparability, relevant information such as distribution yield, term, or leverage ratio had to be painstakingly sought out and manually placed into tables. The time required for all of this per prospectus was between 30 and 60 minutes.
Bernhard KastHead of Deka Immobilien Compass
The Deka Immobilien and IT project teams together with students from TH Mannheim.
Prof Kirstin Kohler and Prof Dr Peter Knauber (right) from the Faculty of Information Technology at TH Mannheim in conversation with Jörg Theopold, Deka IT project manager.
AI Reads, Filters, and Compares
Based on this starting point, it was clear what Deka’s and the TH Mannheim students’ joint goal should be: the AI needed to automatically filter, extract, and validate product information from fund prospectuses.
Over two semesters, twelve business IT master’s students developed three prototypes. During the design phase, the Deka Immobilien team provided specialist input which the students then transferred into the technical solution. Data quality was particularly important for the Deka fund management experts during the development of the prototype. Uncertain values needed to be marked by the AI so that they could be checked manually afterwards. The solution also needed to fit seamlessly into our existing IT architecture.
The results underscore the enormous added value that AI provides during routine tasks. It frees up valuable time for our fund-of-funds managers, so that they can redirect their resources into the efficient management of the funds instead.
Jörg TheopoldDeka IT Project Manager
At the heart of the solution lies a suite of AI-powered processes that automatically identify, extract, and structure product information from PDF documents such as prospectuses and exposés.
The technology leverages several advanced methods such as optical character recognition (OCR), natural language processing (NLP), and large language models (LLM). The final prototype impressed our fund of funds managers with its flexibility, functionality, robust data architecture, and user-friendly interface.
From 60 Minutes down to 5: The New Solution Streamlines the Entire Data Workflow
Upload:
Fund documents are uploaded or forwarded to the AI by email.
Recognition:
The AI reads the documents using OCR and NLP technologies and processes the information in accordance with our high internal data protection standards and, of course, in compliance with the GDPR.
Extraction:
Key figures such as distribution yield, risk profile, and regional distribution are recorded.
Validation:
Uncertain values are marked and can be checked.
Display:
A single interface displays all data in a comparable format, and strategies can be applied as filters.
The prototype proves that there are substantial time savings to be had for our fund of funds managers: the AI tool has reduced what previously required a full hour of manual searching and data entry per prospectus to just five minutes.
