AI and Machine Learning in Banking: An Inside Look

Scaling AI Lynn Heidmann

We released a white paper in mid-July about the challenges, solutions, and steps to get started with AI and machine learning in banking, an industry that (like many others) has a lot of the essential ingredients already but is missing some traction. I sat down with the paper’s authors — and finance industry experts — to discuss some of the key points and get additional insights on what makes AI and machine learning in banking unique.

wall street black and white sign

ai in banking: get the white paper

AI in Banking - About the Authors:


Jason Moolenaar (JM): Jason has more than 20 years of experience in financial services and technology. After beginning his career as an equity sales trader with Deutsche Bank, Jason continued on as a Junior Portfolio Manager for a boutique asset management firm. He subsequently spent the majority of a nearly nine-year career at Bloomberg in the electronic trading group. He currently leads financial services sales across Eastern United States and Canada at Dataiku.


Hursh Rughani (HR): Starting his career off in consulting focusing on financial services, Hursh spent most of his time working with investment management companies. At M&G Investments (Prudential), he worked closely with the company’s Analytics Centre of Excellence as well as projects within risk, finance (AUM and FUM analysis), and operations. Today, he leads financial services sales in the United Kingdom and Ireland at Dataiku.

Pierre Ménard (PM): Pierre has 14 years of experience in technology, mainly dealing with financial services customers. He started his career at a French IT consulting firm in its dedicated Banking & Insurance unit. Today, Pierre leads the financial services team in France at Dataiku. He has been at Dataiku for more than five years, which means he worked with its very first customers and has been able to see the expansion and deployment of more advanced use cases over time.


Lynn Heidmann (LH): The white paper talks a lot about the challenges that banks face in increasing their AI maturity. What would you say is the single biggest challenge banks face in leveraging AI today?

PM: The biggest challenge I see for leveraging AI and machine learning in banking is to scale and systematize in a way so that projects can really impact business operations and processes up to the expected promises of AI. For the customers and prospects I follow, they had their first success — with more or less “blood and tears” to obtain it — in deploying one or a few AI use cases, usually in a limited division or within a subdivision.

Thus, the step is very high to have the capacity to deliver tens and then hundreds (if not thousands) of AI projects in a smooth way. Add heterogeneous scopes, different divisions, and different maturities with data, and the step starts to look even higher.

JM: I come across several challenges with regards to collaboration and explainability:

  1. The key people who are developing the AI project cannot easily work together. Sharing code notebooks and data across a team is a challenge, making sure that everyone’s environments are setup correctly so that they can run their colleagues’ code is a common problem. There are numerous problems I see teams face when trying to collaborate on a project.
  2. Business stakeholders are not involved in the development process and do not receive regular updates on the projects progress during the development phase. These business stakeholders only see an update after significant amounts of work have been done on the project. Often the project is not aligned with the business’s expectations, which leads to the data scientists having to make significant changes to the project. If the business stakeholders were part of the development process, adjustments to the project would have been made much earlier in the development cycle.
  3. Getting buy-in from business stakeholders can be challenging because they do not understand what went in to the project; they only see the results. Having tools, graphs and visuals which show the data used, how models performed, which variables in the data were most impactful, and other statistics help business stakeholders to understand how a result was achieved. All of this supporting data helps demystify AI project results and increases buy-in from the business.
  4. Explainability to regulators is key for banks. Being able to reproduce and show all of the steps, data, and results from a project allow regulators to easily see understand every aspect of a project and the results.

HR: Lots are doing AI proofs of concept (POCs) with new, disruptive teams within banks (data innovation labs, ML CoE’s, Digital Transformation teams, etc.). Not as many are pushing their work into production to see the real business value (this seems to be improving, but it is still a challenge). I think this comes down to a few reasons:

  1. Not always having a business use case from the offset.
  2. Not being able to tie back the results to actual monetary value.
  3. Not having the right technology that allows for operationalization of the work, monitoring, reiteration, and then adaption accordingly to get the best results.

LH: What are the use cases for AI and machine learning in banking that you personally are most excited about?

JM: I am most excited to see areas of financial institutions which historically were not very data driven starting to look at alternative data to enhance their business. For example, investment banking traditionally focused on financial statements for their analysis, along with some market data points. I am speaking to more and more investment banking groups that want to leverage alternative data to drive M&A target analysis, comparable company analysis, and much more to better service their clients. Early adopters in I-banking who incorporate advanced analytics into their client pitches will definitely stand out from their competition.

PM: For the past two years, we’ve seen emerging new types of use cases relying on new capabilities offered by computer vision, natural language processing, and/or speech recognition. In financial services, it has opened new exciting fields to complement the “traditional” data science initiatives (that relied mainly on internal, structured data that is widespread in the numerous silos of the organizations). It has opened, for instance, new frontiers in the AI-based automation of back-office and compliance processes (e.g., advanced information extraction in PDF files, etc.) as well as new sources of improvements in customer operations (automated routing of demands, proactive treatment of complaints, etc.).

HR: The ones that allow you to get big financial returns fastest. For example, a bank I have worked with uses ML to predict when transactions will fail during the settlement process. This has saved them several millions from the fines they would incur. It sounds boring, but the impact is high!

LH: What are the most common use cases that banks generally tackle first to get started in their AI journey?

PM: Usually, retail banks get started with customer analytics use cases like redefining segmentation, churn detection, or recommendation engines. In corporate and investment banks, the first use cases are typically more in fraud detection, fixing of compliance issues, or improvement in regulatory reporting.

HR: Fraud detection and AML.

LH: What are some of the biggest mistakes you see banks make along the way when trying to leverage AI and machine learning?

HR: Not having clear business challenges that AI can solve. When banks create AI or machine learning teams and then chase down use cases, it never works. You need business use cases first and then AI or machine learning may be the solution (it also may not be).

PM: To not sufficiently leverage the business end users and management to validate the operational goals and the way the AI will be concretely used. It may seem obvious, but with the strong organizational silos that exist in banks, this implication is far from natural for many reasons.

For instance, I’ve seen several churn detection projects fail at an advanced stage for this exact reason, even though:

  • Retention is usually within the top five priorities of top management.
  • Churn probability scores have existed for decade.
  • Data lake capabilities and machine learning approaches usually lead to a strong improvement of the churn detection models.

So, why do these efforts fail in that auspicious context?

Because often, the analytics team in charge of the models thinks that improvement of the churn probability score will just naturally contribute to better retention, and that it is not necessary to involve the business line, since the score already exists. So when deploying the project, the business is finally involved, and discussion with them reveals that the branch’s direction does not actually have the budget to retain the new potential churners detected or — worse — that the existing score was not really used so much. An initial discussion may have revealed other routes to more concretely enhance retention through an AI project.

LH: Any other advice or words of wisdom for banks out there looking to transform their business around AI?

JM: To succeed at machine learning in banking, make sure your data is organized properly and that the right people can easily access it before embarking on developing AI projects. Having an efficient way to do data prep and enrichment then being able to seamlessly transition into developing AI projects will drastically reduce your time to market.

PM: Understand that there is a huge heterogeneity of AI and machine learning use cases with very different levels of execution complexity, sensitivity of data, or business impact in case of failure. So be prepared to adapt the governance and operationalization of projects based on these different cases.

For example, enforcing IT governance too stringently for low-sensitivity projects might mean losing the confidence and support of key contributors (e.g., data miners, quants, marketing analysts, etc.). On the other hand, letting a data science team be responsible for deployment, monitoring, and maintenance of sensitive, real-time use cases instead of a dedicated It team or specialized DataOps / DevOps is also a mistake.

HR: Banks have been doing advanced analytics, data science, machine learning, and even AI for years and years. There has been predictive models used for decades, so I’d say banks are the most advanced in AI (despite an image that can sometimes be the opposite). My advice is to balance the power of the open source community with the power of select technologies that provide governance and stability while not locking you into one path or technology (since this field is changing fast).

You May Also Like

Building a Culture of Experimentation

Read More

Building a Modern AI Platform Strategy

Read More

Want to Kickstart AI Innovation? Try a Hackathon!

Read More

MLOps Basics: Why Is Everyone Talking About It?

Read More