Avoiding Bias In AI And The Qualities That Make Strong Data Scientists

Tell us a little bit about yourself, your professional journey, and how you came to be where you are now.

I've been in various credit risk management roles over the last 23 years. I started at Capital One, where I worked in product strategy, credit strategy, and then a more general centralized credit risk management function. I left Capital One after about ten years and moved to London and then to Johannesburg, South Africa, where I was Head of Credit for Standard Bank's Auto Loan portfolio. I came back to the U.S about ten years ago and started working for smaller lenders and startups. Eventually, I connected with Scienaptic. I've gone away from big banks; I like working for smaller companies. I like the amount of impact and influence you can have at a smaller company. The cutting-edge work in the lending space that Scienaptic is doing is what drew me in and resonated with me. I could see myself adding value to their work and being successful here.

Data Science and AI are in such high demand, and more companies are investing in this space. How should companies think about extracting the full value from Data Science?

Data Science is all I know in terms of how to make strategic decisions. Starting at Capital One in the late 90s, we called it "Information-Based Strategy," a precursor to data science. The premise of Capital One's decision-making approach was to look at all available data to help you make an informed decision, whether that be in terms of market segmentation, product strategy, credit policy, etc.

I think so many have been willing to embrace data science more recently simply because of the plethora of data available. In terms of how companies can extract the most value, most people don't realize this until it's too late, but it starts with substantial data collection and retention. Hiring a team of data scientists is excellent, but they need strong data to work with. Data-informed decisions and collection must be part of the company's culture. We are a data-based decision-making company, and we use data science to make better decisions. When there is a lack of data, you obviously can't. Of course, you often have to decide when there is not enough data. I'm not saying data is the only way; it's a mindset. It gets built into the culture and ethos of a company.

What is Scienaptic doing that is driving the demand for data scientists in your organization?

We are a machine learning and AI-based credit decisioning engine. Our software platform functions as the brains of a lending company. When a lender takes a credit application from someone, they pull credit bureau information, call on other third-party data providers, or ask them to connect to a bank account or provide proof of income. They are collecting that data and incorporating it into a decisioning mechanism. Some of that is they have human beings underwriting. For example, lending in the small business space is still heavily human-powered decisioning. But what Scienaptic does is codify and systematize those credit underwriting decisions that a lender needs to make. By doing that, we enable lenders to make better decisions by empowering them with AI and machine learning tools that go above and beyond what lenders have traditionally used.

When we work with clients trying to move their credit decisioning engine onto our platform, we are not just giving them the keys to the platform and telling them to figure it out. We work very closely with them and help analyze their data. We often build them a custom credit risk scorecard that leverages machine learning techniques. It’s almost like a hybrid approach where we sell them software, but we also provide them advisory services. It is a very collaborative approach to getting the software deployed. In that, we’re using our growing team of data scientists to make informed recommendations to our clients so that they can get the most out of our platform and data they have in-house that they may not be fully utilizing.

Our client base runs the whole gamut from smaller credit unions to large banks. Large banks obviously have data science teams in-house, but many of our smaller clients don’t. So we help them supplement whatever needs they have to analyze their data and make data-based strategy recommendations. We are a startup, and startups either grow or die. We are growing very rapidly. As our business grows, we will continue to have lots of roles to fill for data scientists.

There has been a lot of industry spotlight on ensuring that the use of AI is “unbiased,” and this is even more important in areas such as lending. Are there any specific strategies and initiatives that you take to ensure minimal bias?

We have to ensure that our models are unbiased, not just because it’s the right thing to do, but because if they aren’t, regulators will be knocking on our door. Regulators require documentation from lenders to prove that models are not biased or exhibit bias in who gets approved. That goes as far as not just direct, overt bias but also implicit bias. It’s a concept called disparate impact, where even if your model doesn’t look at variables or attributes that would potentially indicate bias, like age, nationality, or race, it could be implicitly creating a bias. Regulators require lenders to prove that their models aren’t creating that disparate impact and disadvantaging certain groups.

Because they’re based on data, all models reflect history. Companies retain a lot of data over the years, and our models are based on the data that has been provided. What we bring to the table to help combat potential bias is our experience having worked with lenders across different spectrums, products, and lending approaches.

Because we have such a breadth of experience with different data sources, we can help make better decisions. We come from a standpoint where we know how to approach a model-building project so that it is less likely to provide bias than it would be if we just relied on historical data provided by that lender. That’s a value that we take great pride in. We approach every modeling project with a mindset to avoid bias, not just because it is a requirement but because it’s the right thing to do. Our CEO frequently talks about wanting to leverage machine learning and AI logic to democratize credit and bring credit much more broadly than it has been historically.

How do you hire? What do you look for in data science professionals?

We start with traditional tools like resumes, experience, and education. Then, what Scienaptic does differently is going through case study focused and behavioral-type interviews. What we are really looking for is problem-solving skills. I came up in an environment where data is used to inform better decisions. So what we’re looking for in an individual isn’t just someone who can apply tools they were taught in school or an algorithm, but can they identify the real-world problem or issue? Can they devise a solution that could be executed in the real world? It’s a real-world problem-solving approach versus a purely technical approach. We value that pragmatic problem-solving angle more than purely technical skills. With COVID, we are working in completely different locations. We have offices in two different countries, in India and the US. I’m in Dallas, but we have US offices in New York. We have people who worked from home, even pre-COVID. That is to say. We are a self-motivated company and value candidates who are independent workers who can get things done. Executing a task without someone constantly looking over your shoulder and checking in is an incredibly valuable skill for a startup like ours.

What advice would you give to younger people looking to build a career in this space?

First and foremost, we generally don’t hire fresh out of university, and most of the candidates I hire have some experience. Early in your career, it’s important to focus on technical skills. I know I was just talking about real-world problem solving, but you need a solid technical foundation, and the best time to cultivate that is earlier in your career. As your career progresses, you’ll have much less time to dedicate to building those technical skills. Building skills early on will greatly value you down the line.

My next piece of advice is to decide which specialty you want to work in. Data science is a very broad term, not all data scientists are created equal, and not all can do the same kind of thing. There are very narrow fields in Data Science, and in more general problem-solving, people leverage data to make better business decisions. Try to figure out what your interest is and what specialty you want to focus on. It could be data engineering, coding, client-facing or problem solving, etc. Figuring that early can help you immensely, especially in building industry-relevant skills and domain knowledge. That doesn’t mean that you can’t jump industries. It just means once you have experience in an industry, your opportunities are probably going to be the best there.

This is not unique to data science, but becoming more than just a technical expert in your field is key. Make an effort to try to understand business problems and understand why decisions were made. Data Science is so much more than just building a model. Ask your boss or end customer how they used your work, what they learned, and what issues they face. Make sure you are trying to see the holistic picture of how data science is being used for decisions. A piece of analysis in itself doesn’t do anything. How do you take information from the analysis and apply that to the decisions you make in the future? Try to see as much of the big picture as possible, even if that means being proactive and reaching out.

My last piece of advice is to network. Cultivate a network of peers, industry leaders, managers, etc. In data science, there are many new techniques, tools, and advancements in the field. It is as important to keep in touch with your peers as it is to build relationships with senior leaders and try to learn as much as possible from them. The beauty of the internet is that you can connect with people on LinkedIn and maintain connections through social media and tools designed for data scientists. So networking can keep you abreast of what’s happening in the world but can also lead to finding your next job.

What do you think companies can be doing to increase diversity in data science and technology as a whole?

I really think that as a society we need to do a better job encouraging diversity in data science and other technology fields. This isn’t to cut too much slack to hiring companies, but companies are really the last step. We as a society can do a better job of recruiting and retaining diverse candidates not just into companies but into schools, programs, clubs etc. From the company standpoint, there is a lot to be done. One of the things we have learned at Scienaptic in terms of retention is that flexibility is key. Having kids, I absolutely understand why women drop out of certain fields or workplaces. It isn’t easy to balance work and one kid, let alone two or three, without an employer that supports the need for flexibility. That puts many parents, especially women, in a position to make a difficult choice between running a family and furthering their career. I have lived abroad several times in my career and other countries do a much better job providing for maternity leaves, back to work programs, and other programs that support raising a family better. It’s an easy area where U.S-based companies can do better and have a huge impact.

In terms of hiring, for smaller companies it’s a lot different than larger companies, small companies hire heavily through networking, especially early on. Networking tends to reinforce biases. Companies need to call that out, recognize it and be conscious of it as early as possible. A new CEO isn’t going to hire a stranger just because of the need for diversity. When starting a company, leaders rely heavily on people they know. That means very quickly as you start to expand, you need to be aware enough to leverage networks outside your own. Companies are much more aware of these challenges than they were 5/10/20 years ago, so I think we’re trending in a positive direction. It's more than just hiring managers. This is a societal challenge that we need to keep bringing up and keep addressing.

Meytier interviews leaders of all backgrounds about their journeys, their work, and their outlook for the future. Know someone who we should interview? Reach out to us here.

Read recent leader speak interviews:

The future of cybersecurity and advice for aspiring security professionals, an interview with Terry Stevenson 

Changing industries and getting into technology leadership, an interview with Radha Sankaran of Capital One

A fearless approach to a career in innovation, an interview with Darlene Newman, DTCC