Artificial intelligence: to the rescue of tedious compliance tasks has been saved
Artificial intelligence: to the rescue of tedious compliance tasks
UCITS eligibility validation with AI
Performance Magazine - Issue 36 ⬤ Published on 9 September 2021
Artificial intelligence: to the rescue
of tedious compliance tasks
UCITS eligibility validation with AI
Partner, Artificial Intelligence & Data, Deloitte
Fabian De Keyn
Director, FIS – Capital Market, Deloitte
Manager, Artificial Intelligence & Data, Deloitte
Senior Consultant, FIS – Capital Market, Deloitte
To the point
The Commission de Surveillance du Secteur Financier (CSSF) requires fund managers to ensure the eligibility of funds relating to undertakings for collective investment. To meet this requirement, fund managers must review these funds’ prospectus documents to ensure they explicitly report certain criteria. As this tedious task is currently done manually, it is both time-consuming and error-prone.
Automating this task with artificial intelligence (AI) and natural language processing (NLP) shrinks this workload, provides significant time gains, and mitigates operational risk. And, it allows fund managers to stay in the driving seat by validating the algorithm’s results. This AI-driven product also enables traceability through an integrated database and metadata tracking process, supporting compliance and audit-check communication.
Undertakings for collective investment in transferable securities (UCITS) are collective investment schemes established and authorized under a harmonized EU legal framework. These schemes allow fund managers to operate freely throughout the European Union through a single authorization from one member state.
The Commission de Surveillance du Secteur Financier (CSSF) requires UCITS to ensure their eligibility. This requires fund managers to check each fund’s prospectus to identify phrases that confirms its accordance with UCITS eligibility requirements. As fund managers have many prospectuses to review, and each averaging around 150 pages, this task is time-consuming and can be error-prone. Generally, fund managers are performing this task manually, generating high operational costs and deviating resources from more added-value activities.
To tackle this challenge, Deloitte explored the field of artificial intelligence (AI), with a particular focus on natural language processing (NLP), an AI subfield combining linguistics and computer science that allows machines to understand, process and perform human-language tasks.
Fund managers review UCITS eligibility by checking if these two necessary but not sufficient conditions are covered in its prospectus:
- Compliance with the Luxembourg law of 17 December 2010 implementing the EU Directive 2009/65/EC; and
- That the fund shall not invest more than 10% of its assets in transferable securities or money market instruments.
Without a dedicated tool, this is mainly a manual task. It consists of reading through the prospectus page by page, documenting the location of sentences that justify one of the UCITS eligibility criteria in a review sheet, and finally checking if both eligibility criteria have been met.
Identifying these relevant sentences from a 150-page prospectus is tantamount to looking for a needle in a haystack. And, if the prospectus is only available in hard copy, keywords cannot be automatically searched. Furthermore, prospectuses from different sources have different layouts, increasing the difficulty of a human review and potentially leading to classification errors.
Figure 1 shows how an AI automated solution can reduce fund managers’ current burden. It inputs the PDF prospectus, preprocesses and parses the text, and then separates and organizes the content into a standard format ready to be analyzed by the algorithm. Through NLP and rule enhancement techniques that recognize patterns, the algorithm can identify relevant UCITS compliance sentences. Based on the sentence extraction, the model can identify the prospectuses that comply with the stated criteria. The model’s classification proposal is evidenced through the algorithm returning relevant sentences, page numbers and paragraphs to provide the fund manager with the full context.
FIGURE 1: AI-based solution
An example of the model output:
- Criteria 1: compliance with the Luxembourg law of 17 December 2010
Sentence extracted: The investment fund is an investment company incorporated in the Grand Duchy of Luxembourg and qualifies as an Undertaking for Collective Investment in Transferable Securities (UCITS) complying with the provisions of Part I of the 2010 Law.
Model output: Compliant
- Criteria 2: less than 10% investment in transferable securities or money market instrument
Sentence extracted: The sub-fund may invest a maximum of 10% of its assets in MMFs.
Model output: Compliant
The involvement of field specialists in the model’s training and output validation is the foundation of the product and optimized its accuracy. More specifically, the algorithm is trained on a dataset of prospectus labeled by fund managers. Fund managers remain the final decision-makers on a prospectus’ eligibility by assessing the batch of relevant sentences returned by the algorithm. Furthermore, feedback logged during the validation process is used to enhance its accuracy.
By virtue of the AI algorithm, the UCITS eligibility check can be performed in five minutes as the analyst just has to review and validate the outputs, which is statistically six times faster than the traditional manual approach.
Automating UCITS compliance tasks with AI techniques like NLP overcomes the limitations of traditional approaches and offers a multitude of benefits.
Not only does it significantly trim operational costs but it also optimizes resource allocation. As the manual processing of financial documents takes up professionals’ time, outsourcing this process to AI frees up talent to focus on the core business.
It can also help reduce operational risks and bias compared with manual processing. In particular, a dedicated platform allows for comprehensive documentation and better traceability regarding potential errors.
Last but not least, the solution provides sustainable data capacity. In stark contrast to conventional methodologies, AI works better with scale—the more data it is fed, the smarter and more agile it gets. This enables organizations to keep up with explosive data growth.
Investment funds and management companies are responsible for checking that investments in hybrid instruments embedding derivatives comply with UCITS requirements.
The traditional manual approach generates high operational costs, especially regarding the large amount of prospectuses that require assessing.
NLP and AI techniques can significantly accelerate this task, while also mitigating operational risks, reducing bias and increasing traceability.