By Abhijit Akerkar
Managing risk is crucial in both protecting business value and improving risk-adjusted returns. Unsurprisingly, banks have made the greatest strides in using AI to manage risk.
Traditionally, risk departments have focused on eliminating undesirable operational risks. In the world of AI, risk departments have the potential to create much more value by partnering with other departments. Working with marketing departments, they can expand the universe of potential clients and offer dynamic risk-adjusted personalised pricing. Working with finance departments, they can reduce bad debt by generating early warnings about customers likely to get into financial difficulty. And working with behavioural scientists, they can devise customised collection strategies.
Seeing around the corner
No CFO wants to miss a profit target. Machine learning can track trends, generate increasingly accurate forecasts and spotlight early warning signs of problems that can then be acted upon. AI can:
- Monitor trends in economic growth, commodity prices, inflation, foreign exchange rates, demographics, regulation, geopolitical events and technology.
- Help forecast how such moves will affect a company’s competitive position, its input costs, customer demand and the health of distribution channels; and
- Predict how the company’s performance will be impacted.
A typical organisation loses 5 percent of its revenues to fraud – equivalent to a whopping $4 trillion a year worldwide.
Until now, companies have typically relied on rules-based models to fight fraud – for example, blocking credit card transactions in a different country until they can be verified. But fraudsters can change their methods more quickly than companies can update their rulebooks. Also, current systems are designed to detect fraud rather than predict it.
But forward-looking companies are using machine learning to connect millions of data points across different products, locations, devices, users, customers and employees to spot signs that are likely to indicate future fraudulent activity. Using a machine learning system devised by Featurespace, a Cambridge University spin-out, Capital One UK increased its fraud detection by 35 percent while reducing blocked card transactions by nearly half. Some banks are deploying auto-rule generation algorithms that allow detection systems to adapt quickly to new patterns of fraud.
In some frauds, customers are tricked into handing over the confidential information needed to gain access to an account. This type of fraud is clearly difficult to spot in real time because the customer accessing an account appears genuine. But machine learning can spot subtle changes in customer behaviour – for example, the sequence and timing of actions that the customer normally uses to authenticate their identity.
Protecting against cyberattacks
Machine learning can sift through millions of incidents to identify anomalies: this is emerging as the central feature of cybersecurity efforts.
An attacker is reckoned to be in a network for 101 days on average before real damage is done. Machine learning can establish what “normal” looks like for a network and identify deviations and anomalies in real time: these may indicate that a cyberattacker is lurking, and a company can take action before an attack actually happens.
In their efforts to step up protecting sensitive data, National intelligence agencies such as CIA and UK’s GCHQ are setting a high bar for cyber defences. This will, in turn, help corporate cybersecurity efforts as companies serving intelligence agencies and other government clientele deploy some of that learning towards corporate customers.
Combatting bad debt
No business is immune to the risk of its customers getting into financial trouble. In the US alone, more than 789,000 businesses filed for bankruptcy last year . Companies can tackle the threat of being left with a bad debt by devising ways of spotting early signs that a customer may be getting into financial difficulty – and then coming up with tailor-made strategies for collecting money due.
- Banks have already implemented early warning systems to identify “at risk” customers six to nine months before a problem becomes obvious. They analyse signals such as increasing overdrafts, overdue instalments, the use of credit lines, change of ownership and other negative news. McKinsey estimates that machine learning can improve the predictive powers of credit early-warning systems by up to 25 percent.
- Strategies for collecting debts can be more accurately tailored to individual cases. They aim to increase recoveries by making the right offer at the right time and through the right channel. Credit card companies have already used this approach to recover some of the money owned to them before selling their balances. A new breed of collection agencies such as Hamburg-based CollectAI and San Francisco-based TrueAccord have shown how more money can be recovered by using an approach appropriate to each debtor. CollectAI says its customers have seen an average 33 percent increase in collection and a 41 percent reduction in processing costs.
Improving credit risk assessment
Companies’ ability to assess risk is being greatly enhanced by machine learning: it can more accurately quantify the risk of default and reduce judgement errors. And AI allows them to scour non-traditional data to create individual credit histories from scratch where a would-be borrower has no conventional credit record.
Some companies are going further. They are using improved risk assessment to adjust their pricing to reflect not only the risk of default but also the prices paid by customers with similar buying patterns, customer lifetime value and the share of that customer’s spend that is likely to captured. Unsurprisingly, financial institutions are furthest ahead in benefiting from this.
Making it happen
To take the next steps you first need to consider the 3 ‘R’s: Relevance – Resolve – Readiness:
- Relevance: The examples in these articles may help you identify the full spectrum of opportunities for your business. However, the degree of their relevance and therefore suitability to your business will vary in terms of ROI and side-effects because, in the final analysis, AI is relatively less about technology than it is about people, mindsets, and ways of working. Pragmatic prioritisation of opportunities based on the feasibility, size of the prize and time to capture value is a must.
- Resolve: The enterprise-wide deployment of any game-changing catalyst, such as AI, requires steely leadership resolve because of the correlated down-stream and collateral impacts – especially on people and processes. The business world is an ever-growing graveyard of dead initiatives and investments because of inadequate or unsustained leadership resolve.
- Readiness: You’ll need a clear and precise line of sight relating to the infrastructure, data ecosystem, data-driven culture and controls that need to be developed along with building the buy-in of all key stakeholders and the effort and time required to upskill people and reimagine processes.
However, getting your hands dirty on AI applications through an agile test and learn approach does not require a significant upfront investment. Many machine-learning models, tools and datasets are available as open source and through application programming interfaces (APIs) from established companies. Computational power can be accessed through cloud-based solutions. Collaborating with external partners will bridge the capability gap in the initial stages.
So, CEOs, if you are merely delegating AI, you are doing so at your own peril. You need to take on the AI bull by its horns, tame it by understanding what it can and cannot do for you, and add it to your toolkit to transform your business and hence your P&L.
Abhijit Akerkar (MBA2008) is Head of Applied Sciences, Business Integration, Lloyds Banking Group.
Stay updated with all the insights.
Navigate news, 1 email day.
Subscribe to Qrius