Artificial Intelligence (AI) and Machine Learning (ML) adoption continues at a slow pace in financial services, although the opportunities for organisations to enhance data capabilities are not lost. The latest LSEG Labs AI/ML Study highlights how companies can bridge that gap to embrace rich data.
- More companies in finance are making use of alternative data, uncovering new challenges in how they integrate and leverage their data.
- With only 45 percent of companies deploying AI and Machine Learning (ML) in more than one area, the opportunity for businesses to become industry leaders in AI/ML has become more pronounced.
- Institutions are turning to a core-satellite hybrid model in the adoption of AI/ML.
For more data-driven insights in your Inbox, subscribe to the Refinitiv Perspectives weekly newsletter.
Increasingly, a chasm is opening between businesses able to understand the world with data, and those who cannot.
Companies with better data, and a better way to integrate that data into their operations, have a clearer view of the opportunities available and are more able to act on those opportunities.
COVID-19 only pushed the boat further, leading to a surge of interest in using real-time data about the world to make decisions. Insights from consumer behaviour, supply chains and company responses to the crisis have raised the urgency of having a data-driven strategy.
We released our 2021 AI and Machine Learning study, which is an annual in-depth survey of hundreds of data scientists within financial services. Of its many striking findings, one stayed with me: 55 percent of firms we surveyed are only using ML in one area or not at all.
The other 45 percent are deploying in more than one area. They are hiring larger teams and solving a breadth of problems from the front, middle and back office.
So what are they doing and is it possible for others to catch up?
How to join the AI/ML party in 2022
Fortunately, it’s not too late – and you’re not alone. In fact, in many ways AI/ML has become easier to get right, compared with just two or three years ago.
1. Where are you starting from?
To get started, it’s important to be honest with your current maturity and capability.
We found that, as far as AI/ML capabilities go, companies have a particularly favourable view of themselves: On average, 63 percent of respondents see their company as either an AI/ML industry leader or challenger, and only 2 percent consider their company an industry laggard.
In practice, few companies have yet to emerge as leaders or challengers. Everything is still up for grabs.
2. Get help
Previously, companies believed they needed AI/ML researchers to develop bespoke models. But increasingly, we’re seeing companies outsource AI/ML services, with demand greatest for third-party vendors that integrate with internal systems.
All but 1 percent of companies surveyed use external cloud providers.
This can be successful when the tasks are generic across industries, so that general tools provided by vendors to help with areas such as customer service, sales analysis and digital marketing are more mature.
Out-of-the-box tools have never been better, but it’s also likely that many of your business specific problems are not out-of-the-box-problems. So how do you build a bespoke solution that meets your needs?
3. Embrace the process
As head of innovation at the London Stock Exchange Group, I’ve seen through many projects implementing AI/ML in different parts of large financial businesses.
It strikes me that, when it comes to predicting whether or not a team is able to achieve an AI goal, the top predictive factor is the team’s willingness to embrace ML as an iterative process.
Getting a repeatable process, where results are continuously delivered, monitored and adjusted is far more likely to lead to success than a single ‘silver bullet’.
Sure, it’d be great for your very first project to be the best alpha-generating model known to man. But it’s an unlikely outcome. Instead, remember: end-to-end processes underpin achievable AI/ML visions.
Later, your initial experimentations can help inform other parts of the business. It’s useful to include your data scientists, data engineers and MLOps practitioners into early conversations on trying out AI/ML in these other areas.
The growth in the discipline of machine learning operations is an indicator that this iterative process is being embraced.
4. Partner with the experts
Subject matter experts and data now matter more than the technology.
Companies are turning to a hybrid model for their data science teams: a core centralised team, and many smaller ones embedded within business units.
Last year, only 8 percent of respondents reported their companies operating in this model. This year, that number shot up to 25 percent.
It’s no accident that investment banking tends to lead on ML deployment.
We see top use-cases and gains are in data-native departments, most notably in operational risk, reporting and compliance, and portfolio management. These are all business areas that have traditionally relied on data-centric employees, such as quants and senior analytics professionals.
it’s better to identify “quick wins” by starting with the parts of the business most competent at managing internal data and information flow.
5. It is still all about the data
Data quality is a discipline in its own right, not a problem for your data scientists to fix. If you have an ML strategy but no data quality strategy, stop and rethink.
Systematic data problems are a source of consternation for companies deploying ML. Linking datasets, crucial for finding new insights, is also a particularly acute pain point (reported by 44 percent or respondents). All the more so with the keen embrace of alternative data.
I hope you find our new research useful and enjoyable to read as you plan for 2022.