Financial institutions are taking a considered approach to integrating unstructured data into their systems, exercising caution as they get to grips with the mushrooming data format and the technology that is enabling generation of it.
At the most recent A-Team Group Data Management Insight webinar, experts and audience members alike attested to the growing importance of the data format while recognising that it requires advanced technological capabilities and substantial resources to unlock value from it.
The data – which is being sourced from everything from PDFs and text documents to social media posts, videos and audio files – accounts for 80-90 per cent of all data ingested by financial institutions, the webinar heard, and ensuring that it is fit for use will be vitally important to finding patterns, insights held within it.
An audience poll that questioned the degree to which attendees’ companies were using unstructured data showed an even split between those that are using it widely, organisations that are deploying it for specific use cases and those that have yet to consider its utility.
The panellists – Victor Tewari, Senior Vice President, Wealth Management and Private Banking, Chief Data Office at Citi; Junaid Farooq, Senior Vice President at First Citizens Bank; Nicole M. Allen, Director, Text Analytics at LSEG; and, Vahe Andonians, Chief Executive and Founder at Cognaize – echoed that sentiment.
They said the results mirrored the current trajectory of utilisation but added that in time more companies would engage more fully with the data.
A Question of Definition
Commonly described as data lacking a predefined structure like relational databases, the panel noted the need for processing to make unstructured data machine-readable and added that failing to harness it meant “leaving opportunity on the table”.
The discussion underscored the critical importance to modern financial institutions of seizing on the value locked within unstructured data. The ability to access and analyse it offers a significant competitive advantage, the panellists said. As one speaker put it, “more data beats better models”, indicating that greater analytical value can be gained from larger and richer datasets.
With a poll showing audience attendees were more commonly using unstructured data to divine customer sentiment and for regulatory processing, the panellists reeled off many more use cases for the data. Among them were fraud detection enhancement by identifying behavioural patterns in unstructured datasets such as chat logs and calls. Negative news monitoring is another useful application, particularly in assessing credit risk. More forward-looking applications include the creation of “virtual twins” for operational efficiency in infrastructure management, and the integration of news sentiment into trading strategies.
Tough Challenges
Nevertheless, the panellists acknowledged considerable challenges in effectively extracting and utilising unstructured data. The process can be time consuming because it requires data collection, cleaning, pre-processing, analysis and visualisation. Scalability, searching, and sorting through vast datasets, ensuring data quality, and establishing standards across diverse data types present further hurdles, the webinar heard. Cost is another significant consideration when establishing an approach to using the data; doing so requires specialised professionals, storage infrastructure and the computational resources to power advanced analytics like large language models (LLMs).
The issue of context, especially within textual sources, was also discussed because the increasing use of LLMs to scour content posed questions about how well these technologies could parse meaning from and identify bias in different sources.
Data Management Approach
Panellists pointed out that current data management strategies often stumble in identifying the proper use cases for unstructured data and in effectively measuring and ensuring its quality. The crucial step of tying the right unstructured data to specific use cases, thereby making it searchable and queryable within a relevant context remains a significant challenge, the panellists agreed.
The speakers recommended that organisations take a cautious and exploratory approach to mining unstructured data. One suggestion was that practitioners embarking on a new capture strategy should cleave closely to their core business while examining where unstructured data could enhance those activities. They also urged organisations to carefully examine whether buying or building necessary capabilities would be the best approach.
Having a clear plan rooted in addressing specific business problems should guide planning, resisting the excitement surrounding new technologies like LLMs, the panel agreed.
Subscribe to our newsletter
