Global asset manager Schroders and Artificial Intelligence (AI) solutions firm Nexus FrontierTech today announced the successful development of a Proof-of-Concept (POC) for a data parsing and extraction solution that achieved over 97% extraction accuracy. The successful POC is the first step in developing and integrating an AI solution that will help Schroders’ Fund Accounting to accurately complete reporting validation checks in just half the current timeframe.
The collaboration between the two parties took root in March 2021 on the Infocomm Media Development Authority’s (IMDA) Open Innovation Platform (OIP), hosted by Investment Management Association of Singapore (IMAS)’s Digital Accelerator Programme, that connects and matches problem owners, consisting of small and medium enterprises (SMEs), large enterprises, and government agencies, to a pool of problem solvers with a range of expertise.
As a global active asset manager that provides a range of wealth management services for institutions and individuals, Schroders sought to address problems of highly manual data extraction workflows and gaps in data coverage faced by many of Schroders’ business teams, including its Fund Accounting team.
Nexus, an AI software and systems development firm that automates and accelerates business processes involving large amounts of fragmented and unstructured data, proposed the creation of a custom-built, industry-specific data parsing and extraction solution.
Using a combination of a new, multi-step engineering method combining Computer Vision and Machine Learning techniques, traditional Optical Character Recognition (“OCR”), and financial- industry specific Natural Language Processing (“NLP”) to detect domain-specific content, the solution’s POC aimed to achieve three main objectives:
- complete data extraction coverage of the three main tables[i] that the Schroders’ Fund Accounting team uses in their validation checks: Portfolio Statement, Statement of Total Return, and Balance Sheet;
- deliver an output accuracy of 80-85%[ii]; and
- develop a model that is scalable for production use.
The timeframe of the POC build was just over 3 months, beginning early September and successfully completed in December. Results far exceeded expectations, with Nexus’ Intelligent Document Processing (IDP) model:
- achieving over 90% accuracy across all three types of tables;
- establishing a systematic way to transform document structure into AI-readable logic;
- proving usability and scalability for future use; and
- proving the effectiveness of applying the Expert-In-The-Loop feature.
The POC successfully demonstrated the technical feasibility and potential business value of applying intelligent data parsing to address user pain points and improve user productivity.
Chwee Kan Chua, Global Head of Operations Innovation, Schroders, commented, “Next- gen NLP-as-a-service is fueled by the demands of accessible AI models to quickly and accurately extract and process complex data that was previously impossible or extremely labor-intensive. By leveraging such capabilities, Schroders can rapidly scale up operations to meet the increasing demands of the business without compromising our quality of service to our clients.”
“With Schroders’ deep domain knowledge of the asset management industry and our IDP capabilities, we’re confident that this partnership will yield numerous benefits for not only the two companies but for the asset management industry as a whole,” added Nexus FrontierTech Chief Operating Officer, Derrick Liao.
“This POC is an important milestone in an industry that is witnessing tremendous growth and under enormous pressure to enhance operational efficiency and satisfy customers. Asset managers are now more than ever depending on Machine Learning and AI technologies to stay competitive, and we couldn’t be more thrilled to join forces with Schroders in their move towards digitalization.”
Schroders and Nexus are moving forward to define the path to production, integrating the models built into the Schroders Fund Accounting team’s day-to-day operations, and will continue working together to achieve further breakthroughs for the asset management sector via intelligent data parsing and reap significantly improved operational efficiencies.
[i] These tables had high variation due to non-standardized structures featuring multiple pages and columns, increasing the level of difficulty in achieving high accuracy in typical data extraction projects
[ii] Estimated market standard for accuracy levels in artificial intelligence