Do I need to be a data scientist in an AI-enabled world?

Updated: Sep 17



So, the short answer to the headline question is ‘no, absolutely not.’ There is both hype and real changes being made in the banking industry. “AI Enablement” is a good description of where we are on our industry journey. This article series is meant to both inform and provide a sense of emerging roles. Some roles are very different, some are variations of existing roles.


1) Introduction - to set the table.

2) Trainer - Machines need to be trained.

3) Sustainer: Machines need to be governed and helped along.

4) Explainer: Machines need to have a voice: Executives and Regulators.

Introduction - to set the table


I get this question in a number of different contexts, like students coming out of college and from existing clients and coworkers currently in the banking field. It is an excellent question and it is good to prepare and anticipate changing roles. Here are a couple of preparation suggestions for our evolving relationship with work and machines:

Be a lifetime learner, be curious….the speed of change is only accelerating. Being open to new experiences and learning opportunities is critical. This is a beneficial mind set.
For college students, consider an academic focus of…. “Major in the ‘what,’ minor in the ‘how.’” That is, employers are increasingly looking for graduates with “how” skills. Those skills should complement the content or “what” skills. In the business context, “how” skills can relate to data science.

“Well, what if I do not want to be a data scientist, then what?” Before answering that question, first, a little background:


Based on my experience, if I was a car, I would be driving near the intersection of the banking process boulevard, risk management road, and data science drive. I have built and led the implementation of Machine Learning algorithms, I have led RPA (Robotic Process Automation) development and integrations, I have led mortgage and consumer lending business units and related consulting practices, and have been responsible for 2nd line risk management organizations. Over decades, I have participated in tremendous banking automation imperative. In fact, some would say the financial crisis was a speed bump/wake-up call. Contributing to the crisis were business and secondary market participants ambitiously and somewhat blindly following algorithms. Some participants lacked a full understanding of their meaning and power.


It is important to understand Machine Learning algorithms are not new. (e.g., Neural Networks / Deep Learning, Decision Trees, Logistic Regression, Bayesian Mathematics / Markov Chains / conditional probability) In fact, they are quite old. Rene Descartes lived around 400 years ago and theorized the linear equations related to matrix algebra and common regression techniques. Linear regression is a bedrock modeling approach used in bank demand and credit risk algorithms. Thomas Bayes lived over 250 years ago and Bayesian math-based conditional probability is at the core of “next product” algorithms. Neural Networks are the “babies” of only about 30 years old.


So what is new? Simply, it is the ability to house, transport, and process data, in a world exploding with data production.

How can non-data scientists be successful in an AI-enabled world? First off, I believe we have some data science in all of us. Think of data science along a continuum of understanding, experience, and willingness to learn. Our relationship with data science is not a binary “Yes or No” engagement. All of us can certainly play a role.


This article uses the “Trainer, Sustainer, and Explainer” AI role engagement framework. Think of these as broad role categories, with roles that need a conceptual understanding of AI or Robotics, but are generally “Data Science lite” or “Data Science free.” Thanks to Tom Davenport (Babson College) and Jim Wilson (Accenture Consulting) for their framework thinking. In the articles that follow, I will provide examples specific to the banking world.




Trainer: Machines need to be trained


In the context of algorithm development, this is commonly done via data sets that are modeled and validated against out-of-sample and out-of-time data. Typically, the learning is “manual” in that data is provided by an analyst at each iteration. Increasingly, “automated” machine learning is becoming available, though not as common today. In the highly regulated banking world, trainers need to understand the regulations. That is Fair Lending, FCRA, HMDA, RESPA, and others. These are rules that need to be categorized, interpreted, and implemented for each bank. This is still done by people today, though, some automation is available for managing the rule sets. Companies such as ComplianceEase provide technology and rules management capability to help with the complex and ever-changing regulatory rule environment.

Tim Sullivan, a ComplianceEase leader says, “The ComplianceEase platform is focused on the quality of our rule sets and speed and ease of use. It is a good fit as banks move up their automation curve.”

(Full disclosure, I have led ComplianceEase integrations) Also, the algorithms do not have an emotional context or develop interpersonal relationships. To the extent algorithms interact with customers, which could certainly happen, the algorithms need to be trained with emotional intelligence. Today, customer-facing bankers are increasingly armed with AI-based suggestions. Mortgage companies like Quicken Loans are using automation to help drive their business.

Dave Schroeder, a Quicken Loans leader mentioned “The Quicken Loans approach is to allow our team members to do what they do best- create relationships, provide counsel, and create customer delight. Machines allow our people more time to focus on customer needs.”

In the context of robotic process automation, robots need to be modeled. For example, if documents need to be read and interpreted, a qualified person needs to show the robot how to perform those tasks. In the case of RPA, some automatic machine learning is starting to occur, but in the main, when documents or processes change, the changes need to be manually remodeled and re-interpreted. As an example, I had a new KPMG associate trained to model the documents for an RPA solution. We used software provided by IBML, a document capture automation company. (Full disclosure, I have worked with IBML and licensed the software)

After a week of training and on her first assignment, she was able to model an auto loan document set and demonstrated a 90% time to complete improvement over the manual process it was augmenting.

It was both a significant money savings and liberated teammates from performing pretty boring work. By the way, the KPMG associate has a humanities-based college degree (not a drop of formal data science academic training!) She does possess a positive attitude and the desire to try new roles.




Sustainer: Machines need to be governed and helped along.


At the core of Machine Learning is the learning process. Think of this learning as an asset that needs care and feeding. Over time, as the algorithms or robots get smarter, the customer experience improves, costs decrease, and a defensible competitive advantage is created. In many ways, this “learning as an appreciating asset” is at the core of the profitability of big consumer AI platform companies like Facebook, Amazon, and others.

Susheel John, an executive with IBML said, “A key part of our automation software is our Accelerated Mortgage Capture Library. This allows our clients a jump start on the document machine learning process, delivering faster implementation and quicker return on investment.”

Learning is tricky. We must be careful about documenting and centralizing the learning. Learning needs to be leveraged by other parts of the organization. Also, people must oversee the learning. Think of AI as both high leverage and high risk. If you teach a robot to do something wrong, it will be really, really good at doing something wrong and in high volume. In some cases, we have intentionally dialed back the machine learning capability to manage this risk.

Machines are not perfect. For example, a robot could get it right half the time, and could still be considered a success by dramatically improving operational efficiency. The key is identifying the potential issue (e.g., a challenge in reading a document) and dynamically queueing the exception to people experts.

Daryl Grant, KPMG's Digital Lending Solutions Lead, said “A critical component of KPMG's Automated Loan Review Platform is the enablement of frictionless interaction between people, process, and technology. Fully automating complex business services is not the goal. My expectation is that intelligent automation enhances the KPMG professional's ability to drive unprecedented levels of quality, speed, and accuracy to delight our clients.”

Similarly in algorithms, identifying and managing false positives and false negatives are critical. In the case of a loan decision, a credit false negative (a loan predicted to be low risk but is actually high risk) could lead to an unexpected loan default. People underwriters specializing in credit evaluation can review high-risk or exception-based loans, leaving the machine to handle the lower risk and more straightforward decisions. In my experience, this general approach has been around for decades. The difference now is the table stakes are much higher!


In today’s banking world, significant resources are needed to manage data and documents. This is traditionally managed at each bank or within bank divisions and product groups. Think of older but still very relevant technologies like SQL Server and SFTP (Secure File Transfer Protocols) to house and move data and documents. As the process of reading, interpreting, and legally executing documents or related agreements becomes more automated, so will the need for better data and document management across organizations. This is the promise of Blockchain or Distributed Ledger Technology (DLT). While DLT “v1.0” got off to a touchy start with the Bitcoin use case, development to “close the gaps” is moving forward at a rapid pace.

Marlon Muller, founder and CEO of ACEx, a Blockchain and DLT enterprise solutions company, says: “The latest generation of our DLT capability literally closes the gap on past generation security issues. It’s just a matter of time before this technology becomes a core banking capability.”



Explainer: Machines need to have a voice: Executives and Regulators.


Comfort level using AI / Automation based capabilities varies greatly, by industry, product/solution type, background, etc. It often depends on direct experience. In areas of the bank with higher unit volume, relatively homogenous bank product types (like credit cards), comfort levels can be higher. Executives with experience in lower unit volume, more heterogeneous “expert” products (like commercial loans) tend to be a little less comfortable. Of course, this is a generalization. KPMG, under Clay Gaitskill, has created a commercial loan review automation solution. So, the general comfort level does tend to be rising.


The point is, education is critical!


Whether selling as a third party or communicating internally, the education process can take time. Also, finding quick wins is often helpful to generate momentum and credibility. As related to the sustainer role mentioned earlier, communicating and evangelizing the company’s AI / Automation priorities is critical. The explainer can have leadership, strategy, consultant, regulatory, or PR-related roles to generate excitement and communicate wins, strategies, risks, and the “art of the possible.”


In the highly regulated banking world, transparency is a regulatory must. A regulatory challenge may occur from a hard-to-explain algorithm. For example, if a loan is declined because of an algorithm prediction (e.g., score cut off) the loan decline reason needs to be communicated to the applicant. A traditional logistic regression scorecard is pretty easy to interpret and identify the independent variable driving the decision. In neural networks, the algorithm complexity makes this very difficult. For the time being, a human interpretable algorithm is required for loan-based scoring.


Allow me to leave you a few parting thoughts. First, thank you for reading. I do hope you found the article helpful. Via the Trainer, Sustainer, and Explainer framework, we have explored evolving banking environment roles. These roles are necessary as AI-enabled capabilities are integrated. In the introduction, we discussed 2 mindsets that should help you, namely:

  • Be a lifetime learner; and

  • Major in the “what,” minor in the “how.”

Finally, you need to adapt to what it means to have a machine as a teammate. This is a tough adjustment!


It is like the machine is getting a promotion, from “tool” to “teammate.”


To dive into a little neuroscience, keep in mind, our brain is divided into 2 selves. (See Kahneman or McGilchrist for more insight) The evolutionary older emotional and instinctive side of our brain (the Limbic system) and the evolutionary younger information processing/executive functioning side of our brain (the Cerebral Cortex). These 2 selves are highly integrated and dependent on each other to be successful. Machines are solidly teaming with our prefrontal cortex self. Machines are faster at processing large amounts of information and providing predictive suggestions based on the insights derived. However, the Limbic system-based emotional self can get in the way. Fear is certainly normal but not a very productive response. Machines can be very useful, but only if “both of” you take the time to really get to know your new teammate!


Your Personal Finance Journey Guide:


Core Concepts

1. Our Brain Model

2. Curiosity Exploration - An evolutionary approach to lifelong learning

3. Changing Our Mind

4. Information curation in a world drowning in data noise


Making the money!

5. Career choices - They kept asking about what I wanted to do with my life, but what if I don't know? - Part 1

6. Career choices - They kept asking about what I wanted to do with my life, but what if I don't know? - Part 2

7. Career success - Success Pillars - Maximizing luck with an adaptable mindset to reach your goals!

8. Career choices - Do I need to be a Data Scientist in an AI-enabled world?

9. Career choices - Diamonds In The Rough - A perspective on making high impact college hires


Spending the money!

10. Budgeting - Budgeting like a stoic

11. Car Buying - Auto buying and financing thoughts from a Behavioral Economist, a Banker, and a Dad

12. College choice - The College Decision - Framework and tools for investing in your future

13. College choice - College Success!

14. College choice - How to make money in Student Lending

15. Event spending - Wedding and event planning guiding principle


Investing the money!

16. Investment thoughts for my children

17. Using the Stoic's Arbitrage to choose a great investment advisor

18. Anatomy of a "pump and dump" scheme

19. The Time Value of Money Benefits the Young

20. How Would You Short The Internet?


Pulling it together!

21. Capstone - The Stoic’s Arbitrage: A survival guide for modern consumer finance products

3 views0 comments