1 April 2022 - Over the past few years, asset servicing across the alternative investment universe has experienced significant growth in Assets Under Administration (“AUA”), with funds across the sector increasingly needing more support from their asset servicers as they manage rising volumes.
This, in turn, has translated into an explosion in the amount of textual data exchanged between funds and asset servicers.
As witnessed across many other industries, technology is being used more and more for the heavy lifting behind the scenes – especially when it comes to managing this data.
One technology making waves is Natural Language Processing (“NLP”). NLP concerns the interactions between computers and human language, with a focus on how to program computers to process and analyze large amounts of natural language data.
The aim is to get to a point where a computer is capable of "understanding" the contents of documents, including the contextual nuances of the language within them. The technology can then be used to accurately extract information and insights outlined in the documents, without a human having to manually do so.
NLP can have a variety of uses. In the world of risk management, for example, NLP can be used to extract significant contract terms from contract documents.
Automation can also help reduce manual errors, process forms faster, and be more consistent, while it can also scale quicker than administration which depends solely on manual processing.
All of the above can lead to a reduction in data entry for asset servicing staff, and leads to increased verification and focus on more crucial tasks, such as extracting significant contract terms to check parties are compliant.
Is NLP a must have for asset servicers?
Is NLP really needed in the fund servicing industry? It is a fair question, especially as the industry has managed to cope without it so far.
However, technological advancements can undoubtedly bring benefits to businesses and clients and, if integrated correctly, NLP could be the workforce multiplier that enables fund servicers to better handle textual data ingestion and processing efficiently and effectively.
Indeed, given the large number of documents processed in the industry, this technology could be critical for asset servicers and their clients in future.
There are multiple use cases where it can be implemented across the industry; when it comes to business operations, for example, it can help drive faster processing of paperwork in accordance with Service Level Agreements by reading pdf statements from third parties to extract Net Asset Values (“NAV”s) and speed up NAV finalization.
It can also improve the client experience by automating data extraction for processes such as capital call notices, corporate actions, and account statements via a variety of unstructured formats to speed up responses to clients.
The key with much of this is time. NLP, if implemented effectively, should reduce turnaround times. This could range from NAV processing that may switch from monthly to weekly (or even daily) cycles as more steps get automated, to its use in collateral management where clients may expect faster responses to their emails as the industry automates the process of “inferring” client emails thanks to NLP. If more asset servicers adopt NLP then investment funds operating across the alternative investment universe may well grow to expect NLP-based solutions as the norm, provided they are making life simpler for all parties.
In such a world, those asset servicers who don’t have cutting edge solutions may well see an impact on their ability to maintain and grow their market share.
Digitization at the heart of the industry
While tools like NLP could see wider uptake, it is worth remembering one other point; to help handle the ever-growing amounts of data, the real end game should be to build a better ecosystem for all participants in the industry.
More standardization of the various practices we carry out daily would make a big difference, and a number of participants across the industry (ourselves included) are pushing for this in order to help everyone communicate more effectively.
Although this won’t happen overnight (meaning tools such as NLP will be necessary in the near term) it should be the aspiration for all participants to largely eliminate non-standard data and paper in future and truly digitalize the space.
As ever though, this means getting competitors to cooperate by adopting the same technologies, and historically this has often been the sticking point.
However, it should remain the industry’s ultimate goal to digitize where possible, creating a more streamlined experience for all participants in the alternative investment universe.
By Tim Mietus, Senior Executive Vice President of Innovation, Citco Technology Management, Inc.
The evolution of natural language processing
The Citco group of companies (“Citco”) has been researching natural language processing for a number of years as part of its technology innovation program, and there are a wide array of approaches being utilized in this space.
Technologies that extract text from documents are not new, having emerged in some format or another in the last fifty years. The earliest versions are mostly based on Enhanced Optical Character Recognition – something Citco itself has utilized for several years.
These tools map fields in the document by position but they have drawbacks – for example, if things move on the page they usually have difficulty adjusting without user interaction.
Then there are deep learning models that understand text and its meaning, such as Named Entity Recognition (“NER”) models which can identify key text within documents.
NER models extract information from documents in the same way that a human would, using the structure of the document and specific key terms to identify information which they need to extract. In the investment world, this means looking for words like “final”, “maturity”, or “settlement”, for example, followed by a date (in the example above that would be in order to extract the maturity of the instrument).
This in itself is a form of NLP, but now models are emerging which are one step beyond this. The latest variants of NLP can now understand the text on the page and read it in a similar way that a human would, being able to recognize labels and structures, such as tables, as well as text. As a result, it can read similar documents once trained, and changes to documents don’t impact extraction as much as they do on other approaches.
These processes are being enhanced all the time and are becoming more effective, and we expect them to be used more and more across the industry.
By Albert Bauer, Chief Technology Officer, Citco Technology Management, Inc.