Growth of Machine Learning in Asia and its Drivers

By Sanjna Parasrampuria, Head - Applied Innovation, Asia, Refinitiv

Sanjna Parasrampuria, Head - Applied Innovation, Asia, Refinitiv

Refinitiv is a USD 6.2 billion dollar annual revenue financial data and technology business. As the Applied Innovation lead, my role largely is to identify strategic and scalable opportunities that leverage data and emerging tech and to orchestrate the execution using lean, iterative and agile approaches. I lead an extremely talented team in the Labs and constantly work alongside global banks/asset managers/hedge funds, regulators, fintechs and bigtech to identify and synthesize real industry problems. Much of our work involves breaking new ground for alpha generation, risk mitigation, financial crime prevention, event correlation and sentiment detection - many of which make their way into augmenting our existing products and new product pipeline. Refinitiv Labs is a network of data scientists, research scientists, engineers, design specialists and subject matter experts spread across the global fintech hubs in New York, San Francisco, London & Singapore. We leverage the petabyte per day data we collect and the terabytes of historical data we have and many times combine those with alternative data sources and other proprietary data.

It depends what you use as an indicator. If benchmarked against North America, Asia of course still has ground to cover. That said, Asian institutions are more advanced in some areas than those in Europe, such as in machine learning (ML) being core to business strategy and the projected growth in numbers of data scientists.

At an operational level, the reasons for ML adoption are quite telling. Unlike popular belief that saving money is the primary motivation, Asia (like its counterparts) is driven by the desire to extract better quality information (69 percent), generating trading investment ideas (65 percent), and the desire to increase speed and productivity (65 percent).

Although at a macro level, we see that the high regional growth rate, the availability of open source ML technologies and importantly, the growing population with rapid mobile/digital adoption are leading factors making ML become core to strategy in the region.

Maturity Drivers of AI and ML Initiatives

A push to maturity is determined by the business’ sense of urgency, data availability, and scientific expertise. Given Asia’s lagging technology adoption, as compared to North America, only when the business environment’s competition can be won using an AI/ML strategy, will the technology gain industry-wide adoption. This was seen in the case of Grab vs. Uber, and can be seen in the battle between ecommerce leaders every day.

Though even if the business sees the value in implementing ML, there are a few barriers to adoption that are often missed: Data availability, data quality and data analysis capabilities. Asia has a wealth of yet untapped data. For instance, only 53 percent of the Asian respondents are using alternative data in their ML work as compared to 93 percent in the Americas.

In our lab, we developed a Data Science Accelerator for financial institutions to increase the speed at which data can be turned into usable input for a data scientist. Unsurprisingly, the traction from Asia was unprecedented as it provides data science ready data saving stress on the existing talent pool.

Lastly, the gap between the C-level leaders and the data scientists needs bridging. A common vocabulary around what AI/ML can and cannot do needs to be understood. Business leaders don’t have true visibility on how AI/ML are used to solve their business problems. Therefore, they lack clear articulation of problem statements in a way that can be readily turned into a useful AI/ML application by a scientist, an area that will have to mature over time.

AI and ML as Part of Core Business Strategy in Asia

Knowing that 76 percent of the parties interviewed in our machine learning survey are implementing ML in their core business strategy, 8 percent ahead of Europe, is encouraging. Though if you dig deeper, only 29 percent have actually deployed ML in their core business.

This is likely based on the fact that we are yet to mature in ML adoption and while ML has been accepted at a strategy level it still needs to percolate down to the ground for adoption at scale.

Real value from ML efforts need the right ingredients and real understanding of what is possible and what is needed to make it happen

Some of this is also on account of wanting to appear as an innovation company in the perception battle. The hedge funds where innovation and investment has been traditionally focused are certainly ahead although the greater availability of tools is likely to materially level the playing field—on both the buy and sell sides.

Another observation is that senior leadership need to appropriately incentivize their business managers to support ML adoption with their business lines. Real value from ML efforts needs the right ingredients and real understanding of what is possible and what is needed to make it happen. Educating this powerful segment of employees, identifying their communication gaps, and strengthening their ML/AI understanding over time is one approach. Elevating the data science team to the C-Suite level, creating a cross functional body responsible for the execution of the organization’s AI and ML projects, is another.

Prioritization of the Application of ML and AI in Asia

Beyond data availability and quality, talent is and remains the ultimate deciding factor. Although, nearly half of our survey respondents informed us that their companies will be increasing the number of scientists on staff in the next 12 months, actually recruiting, training, and retaining these scientists is a different story.

The Singapore government for example has set up multiple training resources to address the dearth of professionals trained in the field which addresses one part of the equation. A good data scientist is not just a whiz with numbers but also brings a certain level of relevant domain expertise to the table. That’s why the more tech centric and savvy departments enjoy the benefit of data science sooner, think payment processing, than those who are still stuck with largely manual processes, such as procurement and operations.

Let’s assume all the ingredients are on hand: Skilled talent, connected data, and a department ready to transform, then the final hurdle is proving the potential ROI of this investment to the executive leadership. Will you indeed contribute to the bottom line in one year’s time? If the answer is unknown then you better rethink your strategy or realign your expectation.

Reason for Risk Mitigation and Wealth Management as Key Focus Areas in AI and ML in Asia

The reason is practical more than anything.

1. In order to get buy-in from your executive leadership team to make an investment in emerging technology in a region that is new to this field, the potential ROI must be substantial to offset the cost.

2. Areas like the wealth management industry in Asia are in a very competitive space where MNC and local banks compete intensely for the fast growing Asian HNIs share of wallet. Given that the wealth in Asia is changing guard to the next generation who are extremely digitally savvy and far more self-directed than their predecessors, there is an obvious need to invest in tech for banks to compete and differentiate.

3. Lastly risk avoidance, potentially mitigating millions of dollars in losses in a regulator driven region is an easier sell. Throwing human resources at a problem instead of technology when it comes to reporting, reconciliation, and data entry is a problem with a defined bottom line. It is at the analysis stage where rubber hits the road. Finding a needle in a haystack as commonly required in AML and KYC, is simply an impossibility if done manually.

Just as correlating 100 or more variables to derive a leading indicator for an investment idea. That’s why most resources invested in ML in Asia, 69 percent according to our survey, are spent on wealth and risk management.

Dealing with Data Quality Issues

Data quality is at the heart of the impact your data science team can produce. The problem lies in a few core areas: Data availability determines whether there is even enough data to warrant a data science initiative; data quality addresses duplicates and inconsistencies; and data reliability, whether you will be able to access more of the same data in the future or if it is only a static data set for one-time use, which would require a system restructuring to create a reliable stream of data to feed your models.

The truth is, all three are often a problem. Luckily, there are a few solutions in the market. Data cleaning is a service, which can be procured by various vendors, so is data enrichment and supplementation with global data sets. One solid option of course is acquiring market data from trusted sources such as Refinitiv where data integrity is paramount. As I mentioned before, alternative data remains underutilized in Asia and rightly so, as before on boarding any new data source, financial institutions need to verify the provenance, the usability of the information, and all the above parameters and how far back the data is available for. Alpha seekers often ask, “Can I back test a model I have built to actually use in production?” In a world of post GDPR, not just data quality but also data privacy have taken centre stage.

Relevant Structured vs. Unstructured Data

There are several reasons for this ranging from maturity to talent skills to legacy architecture. Firstly, there is a lot of untapped potential in unstructured data—the power of generating signal from news, social media, company filing, broker research, transcripts, and more are plain to see. So I’m not surprised with this estimate.

Although, another very practical reason could be that in banking for instance, data science teams find it easier to access open source data as they use the cloud-based ML technologies. Getting access to company data in these cloud environments is an onerous task on account of security and privacy concerns. A lot of the open source information tends to be unstructured in nature and thus you find a lot of work with it, especially in companies that have not yet formulated a cloud-based sandbox environment to allow for efficient use of company data which tends to be largely structured.