Thanks to the sheer amount and complexity of data that is generated each day, the office of the chief investment officer has effectively expanded into an intelligence unit in recent years.
Data is flowing in from multiple sources, including custodians, fund administrators, consultants and managers, to name just a few, and that data is flowing into inboxes and shared drives in tens of thousands of emails per year. All of that information could live in any number of functional and technology silos within a typical investment firm, while at the same time, complex assets often have life cycles that outlast investment staff, leading to issues around knowledge transfer and data continuity.
Machine learning and artificial intelligence (AI) can absolutely be valuable tools in collecting, analyzing, managing and putting that vast amount of information to use. They can generate more insights and better analysis, create smarter automation for business processes and continuity, drive down processing costs and generally yield more value from data.
However, they are not a magic bullet. The technology is only as smart as the data it is working with. This means that not enough data, poorly trained data or dirty data will render even the most advanced automation and decisioning algorithms ineffective. Data governance is the critical element that will determine the efficacy and success of any machine learning or AI initiative.
Defining Data Governance
By data governance, I am using an umbrella term that refers to defining the sources of data within an organization and the process for gathering data and making sure it is clean. In practice, this often involves data stewards in the firm implementing practices to account for compliance, security, privacy and quality of information.
All of this has to be in place starting from day one in order for machine learning to be effective, as these processes and the proper information architecture will ensure that data sources are being combined in sets to tell the right story.
The first step is understanding existing data flows across teams, as well as who owns what part of each flow to establish the source of truth. This will help firms avoid a "garbage in, garbage out" scenario when attempting to use machine learning. When AI gets it wrong, it's most often due to a data governance issue that hasn't been sorted out from the beginning.
It is equally important to be outcomes-focused in establishing data governance processes. An ultimate goal backed by executive support is what will ultimately lead to organizational change. Whatever the clarion call may be, there must be one that serves as the guiding principle to which everything ties back. This helps keep firms moving toward their ultimate goals (and avoid getting sidetracked by occurrences that don't have a direct link to the desired outcome).
Putting It Into Practice
Once sound data governance practices have been established, firms can begin to realize the maximum benefit from investments in machine learning and AI technology because they're starting from a place of clean data for entities and relationships of data sets for funds, accounts, investors and organizations.
As the CEO of a technology provider to institutional asset owners, some of my recommendations for leveraging machine learning and AI may include:
• Leverage automation to reduce manual touch time on processing incoming data, while simultaneously unlocking new insights from that data: Institutional allocators receive hundreds of documents related to their investments into private equity funds or hedge funds on a monthly basis. Some of these documents contain timely information that requires action, such as call notices. Others may contain information that needs to be extracted for analysis and summary, such as balances and transactions.
With sound data governance practices in place, allocators are able to automate the identification of these documents and the information extraction, removing manual errors and delays.
• Streamline knowledge transfer for enhanced business continuity: Institutional allocators regularly manage assets with life cycles that outlast the tenure of investment team members. When an individual leaves a firm, they can often take much of the valuable knowledge about the investments they've been managing with them. Standardized, automated data processes will ensure that all knowledge about an investment or asset is preserved correctly and comprehensively for the next manager to step in and provide seamless continuity of service.
• Organize vast amounts of unstructured data: Another key challenge for institutional allocators is consuming and organizing the vast amount of unstructured data received. Machine learning can address this challenge in a few ways. Data governance policies and processes should ensure AI models have been diversely trained with ample amounts of clean data. With this underlying information architecture in place, firms are able to implement systems of automated collecting in a structured form.
Another option is to use machine learning to process the data using natural language processing (NLP) and named entity recognition (NER) techniques to automatically categorize the information. Both will save time and boost efficiency.
These are just a few examples of the benefits that machine learning and AI can yield with proper data governance in place from the outset. Looking ahead to the coming decades, growth in the financial services sector will only continue to be fueled by the automation and productivity gains driven by technology innovations in machine learning and AI, especially as more and more data is generated each day.
In order to reap the full value of these technologies — such as the automation of manual tasks, seamless knowledge transfer, ordering unstructured data and more — institutional allocators must also have the right data foundation in place, and that starts with proper data governance. It is only with the processes in place to ensure data quality that machine learning and AI can be effective to drive real organizational gains.