THANK YOU FOR SUBSCRIBING
Artificial intelligence and machine learning (AI/ML) is a game-changing technology, primarily driven by the tech giants like Google, Amazon, Apple, Facebook, and Microsoft. It is rooted in statistical methods to predict the probability of an event happening. Instead of formulating algorithms based on a hypothesis and coding the algorithm to process data, as in traditional programming, machine learning uses historical data to develop the hypothesis and create a program (the “model”) to predict outcomes when processing new data.
Proofs of concepts are emerging from the shadows of laboratories and being operational in production deployments.
“Data is the critical resource that the machines use to build the AI/ML models”
There are several drivers for this change, the main ones being
- Low interest rate environments and the resulting margin pressures causing banks to look at ways to look at ways to increase revenue
- Competitive pressure from fintechs encroaching into the more profitable aspects of banks’ value chain (credit origination)
- Changing attitudes of consumers towards tech giants as providers of financial services
- Softening stance of regulators on use of these technologies that benefit consumers.
- Pandemic causing transient hits to credit worthiness
In loan underwriting, AI/ML models better predict a good credit risk based on features beyond the traditional FICO scores and debt to income (DTI) ratios and help approve more applications without increasing risk. The models help navigate past the pandemic triggered events on the credit profile.
Customer attrition is another area banks can benefit from using the technology, with the models providing strong signals of an impending departure and potential actions that can be taken to prevent it.
In fraud detection, models help distinguish between genuine customer transactions that appear as anomalies and indications of a compromise on the account, and recommend and take the right action, thereby reducing “false positives” and improving customer experience.
More recently, AI/ML models have enabled banks to tailor products and services to suit individual needs, creating a “segment of one” not just at the point of sale but through the customer journey from sale through life events, providing a unique individualized experience through “mass customization.”
Self-service channels, mobile apps, chatbots, and websites can leverage AI technologies like Natural Language Processing and sentiment analysis to assess a customer’s mood and trigger a transfer to an agent to respond with the required level of empathy.
Data is the critical resource that the machines use to build the AI/ML models. From a data engineering perspective, three aspects of data need to be addressed: Quantity, because the larger the dataset, the better it is for model training; Quality, ensuring completeness with missing data replaced with appropriate substitutes; and data devoid of bias, being representative of the general population being modeled to ensure the model created is not skewed creating biases against the unrepresented category.
Shortage of data science skills is a real issue. Besides continuing to hire data scientists, banks should offer “data science internships” leading to full-time employment. Banks also need to encourage existing staff with a flair for mathematics/statistics to take courses offered by platforms like Coursera and Udemy at a nominal cost and flexible timelines, and try their hand at solving problems local to their field of work, under the mentorship of a “Citizen Data Science” program.
Resistance from the practitioners in trusting the model’s decisions requires engaging the practitioners, early on, in the modelling process.
On closing, there is an Orwellian school of thought that the machines are taking control. But the reality is that AI is just a prediction tool that can be applied to make better decisions for the benefit of customers, employees, and organization. It can be seen as evolution of the data driven enterprise into a machine learning driven enterprise, moving from a rear view mirror driven approach (business intelligence) to a predictive approach of machine learning and AI.
Finally, there’s the impact of automation in general and AI/ML in particular, displacing jobs as the nature of work changes. While there will new jobs created, as in “bot supervisors” in RPA, and “trainers,” “validators” and “explainers” of the model, these will not offset the jobs that have been displaced. Organizations need to retrain staff for the new roles as well as help place the displaced workers in other jobs both internally and externally.
These are the key takeaways for organizations to succeed in their AI journey:
- Get the data in order; a clean, complete, and representative of the population being modeled
- Get the required data science skills, seeded with data scientists from other areas of the bank and/or external hires supplemented by internships and citizen programs
- Create an explorer group, that looks for opportunities in different lines of business
- Partner with a solutions provider that can bring technical and domain skills to jumpstart the program
- Establish a competency center, that provides best practices and encourages reuse.