By Peter Simon, Financial Institutions Data Science Practice Lead, DataRobot
If banks want to survive and thrive in and after COVID-19, AI and machine learning are very important, but the current approach to making these technologies happen simply isn’t fit for purpose. Automating development, deployment and monitoring processes unblocks the bottleneck.
You’ve probably heard the one about COVID-19 doing more to change the way we work and to make global business digital in one year than the last thirty years of the “paperless office”. Something similar seems to be going on in financial institutions: the rapidly-shifting environment in the pandemic is driving changes to how banks work more urgently than a decade of competitive threats from fintechs and challenger banks ever did. At the same time, it’s unleashing a gale of creative destruction that could threaten the very fabric and existence of venerable financial institutions if not addressed.
Let’s cut through the hype and take the following premises as read:
- The COVID-19 pandemic is awful in many ways and devastating human lives and livelihoods.
- There are unprecedented (yeah, sorry, that word) downside risks to the global economy as a consequence of both the pandemic and policy reactions to it. These are affecting businesses across all industries and countries; the business of money is no exception.
- COVID-19’s effects are different from a typical recessionary downturn in a number of ways, and unpredictable. Yes, the pandemic has felt like a slow-moving train wreck this year, but business conditions have been moving fast, and one month’s certainties quickly become the next month’s history.
So what’s a bank to do? There are three challenges to address; let’s call them the three stages of ‘Rona Response.
First, survival. Banks’ short and medium-term survival will depend on their ability to identify and mitigate risks as the economy changes during and after the crisis. One would think that, almost a year into the pandemic, this stage would be complete, but arguably the worst of it will only arrive once fiscal policy ‘life support’ measures finally expire. As it is, the number of large corporate bankruptcies in the US this year already looks to be on track for an all-time high, with correspondingly high risks of collateral damage—and correspondingly difficult decisions to be made.
Next, resilience. Survivors will need to be proactive with identifying at-risk clients and help them before it’s too late. This will earn customer loyalty, better protect shareholders by reducing loan losses, and help banks and their communities to recover more quickly (thus being an investment in the institutions’ long-term viability as businesses). It will also shore up the odds of survival should the economy and/or the pandemic take another turn for the worse.
Finally, long-term adaptation and growth. The focus here is on preparing the post-COVID world, whatever shape it may take. It looks like we won’t see a return to “business as usual” anytime soon, but the ability to adapt quickly to whatever shape the “new normal” takes will ensure long-term profitability. There’s no time to wait and see how this new shape will pan out, as by then the most responsive institutions and fintechs will have stolen a march.
The common thread in addressing these challenges successfully is the ability to be nimble in the face of rapidly evolving conditions—effectively becoming an “agile bank”. It’s not going to help anyone if ensuring survival, by changing the way your institution does business, takes a 2-year big-bang transformation programme involving comprehensive re-engineering of all the business processes, hundreds of PowerPoint slides and thousands of consultancy hours; at the end of all of that, it could well be too late. Far better to address these challenges by a series of small, incremental, but immediately impactful adjustments which are fast to deliver. A good place to start is the way in which financial institutions make decisions; not the big, strategic decisions, but rather the hundreds of thousands of everyday operational decisions a typical bank makes, such as which customers to approve, how to target the marketing, what pricing to offer to new and existing customers, which transactions require supervision or investigation, and so on.
Example AI applications in financial institutions, by stage of ‘Rona Response
|● Rapid-response models (address specific short-notice needs and gauge speed and extent at which drivers are evolving)
● Anomalous activity detection for risk flagging (compare to pre-crisis activity patterns)
● Explanatory models of previous crises (what drove KPIs? what is different this time?)
|Survival use cases, plus:
● Customer forbearance models: where to extend maturities, reschedule payments, grant temporary limit increases, reduce rates, and/or waive late fees
● Regulatory capital and liquidity needs scenario modelling
● Sensitivity analysis5
|Resilience use cases, plus:
● Determining new channel usage patterns and preferences
● More efficient client outreach, to increase value, positive brand awareness, and loyalty
● Automated pricing and targeted marketing of new products
AI and machine learning have made inroads in this space, but banks have often struggled to realise these technologies’ promise. A recent McKinsey study found that reasons for this include, inter alia, a lack of clear strategy, inflexible infrastructure and outmoded working models that make it difficult for business and technology teams to collaborate. Let’s add slow and largely manual development processes, poor communication between technology and business silos, and the glacial pace of governance and implementation to the case for the prosecution. If banks want to become more responsive and agile, the current approach to making AI happen quite simply isn’t fit for purpose.
Modern automated approaches to machine learning and AI (“autoML”) help financial institutions achieve this, unblocking the bottleneck and sweeping away some of the problems described earlier, by:
- making the people who build the models to automate the decisions much more efficient and productive, automating their repetitive tasks and massively shortening the development cycle;
- exposing the ability to build powerful predictive models to a wider audience, such as business intelligence (BI) specialists who currently focus on building current-state or historical information—this allows the business to get much closer to the models than was previously the case; and
- making governance, deployment, ongoing monitoring and model refresh a much faster process, by enforcing a structured, repeatable approach to addressing the business problems at hand—even automating much of the documentation needed for model validation.
What can be automated?
Embracing these technologies can lead to some quite astounding results. One of DataRobot’s fintech customers has accelerated their credit model governance process to the point where credit scoring models can be built and deployed in a matter of hours. This allows the business—which approves over three thousand loans each day—to react quickly to the changing realities of their marketplace by ensuring that their decisions are based on the very latest data. Of course, not all banks will be able to achieve quite as drastic an improvement in process speed, but it’s realistic to expect substantial benefits; typically, we see the model development cycle shorten from three months to less than three weeks, with model validation, governance and implementation times also substantially reduced.
Around the world, we’re seeing increasing numbers of financial institutions of all shapes and sizes adopting these technologies. This newfound ability to finally “do AI right” is therefore becoming an increasingly important ingredient in the recipe for financial institutions’ long-term endurance, resilience and adaptability.
 Bank jumping over the candlestick is optional.
 The fully paperless office apparently being on track to arrive at around the same time as the fully paperless toilet.
 Perhaps even in the original, Schumpeterian sense, but that debate is well beyond the scope of this article.
 Special care is needed with this use case, as the unprecedented nature of current events results in an absence of historical data on which to train the machine learning models, and models trained on what is available may not extrapolate well, or at all. We list it nevertheless, as it can be very powerful when used wisely.
 Available at https://www.mckinsey.com/industries/financial-services/our-insights/ai-bank-of-the-future-can-banks-meet-the-ai-challenge
Global Banking & Finance Review
Why waste money on news and opinions when you can access them for free?
Take advantage of our newsletter subscription and stay informed on the go!
By submitting this form, you are consenting to receive marketing emails from: Global Banking & Finance Review │ Banking │ Finance │ Technology. You can revoke your consent to receive emails at any time by using the SafeUnsubscribe® link, found at the bottom of every email. Emails are serviced by Constant Contact
Top Stories1 day ago
Marketmind: Markets turn risk-averse after bumper month
Top Stories1 day ago
BoE’s Bailey says getting inflation to 2% will be ‘hard work’
Top Stories1 day ago
Baer says exposure to single group tops 600 million francs, as Signa crisis deepens
Top Stories1 day ago
UK’s FTSE 100 dips on miners, energy drag