The world’s leading publication for data science, AI, and ML professionals.

Why Bankers Will Not be Replaced by Artificial Intelligence Any Time Soon

The challenges and career opportunities for all flavours of data expert

Image by Author: Generated with Midjourney
Image by Author: Generated with Midjourney

Robo-bankers are still science fiction

Many science fiction films depicted a world where AI & Machine learning systems would replace us, but in large banks this is far from the reality. What’s actually happening is that ML revolution is opening doors nobody would have believed were there a decade ago. I am fortunate enough to be at forefront of this, so I’d like to give you a view of what’s happening in the industry right now.

While it might seem like the pace of innovation at the large banks is slow compared with FinTechs, I would argue that the complexity of the challenges being faced are greater. The savvy businessperson in you will sense that where challenges lie, opportunity follows…And there are many at the large banks right now.

The disruption sparked by the rise of Fintechs has forced incumbents to start taking machine learning seriously, with many now having funding specifically for ML innovation. This is exciting news for all flavours of data and ML experts worldwide.

The relative immaturity of large banks

Any ML expert will tell you that data is the "magic" ingredient that makes it work. Large banks are entrusted with tonnes of valuable data on their customers, but making use of this data is proving difficult. Right now, In comparison with the likes of Facebook, Google and Amazon the data ecosystem for machine learning is immature. I see the main challenges as infrastructure, regulation and machine learning expertise.


Cloud platforms & legacy infrastructure

Cloud platforms are quickly becoming the easiest way to manage the storage and computational requirements for applying machine learning at scale. Many incumbents understand this and are slowly transitioning onto cloud. Now cloud is hardly new, some tech companies have been operating in cloud platforms for the good part of a decade. So, what’s the issue, why is it so challenging for the large banks? To understand the challenge, we need to have a brief look at history.

Most incumbent banks have existed since the dawn of computing in financial services.

In 1959 Martin’s bank (UK) installed the first computer to reproduce all current account operations, in parallel with standard operations.

Since then, data infrastructure of banks has grown into a complex mesh of legacy platforms and confusion. Naturally, this has led some in the industry to proceed with due caution before adding even more layers. The main fear here is technical debt, which some banks are paying dearly for right now.

Opportunities: Bank’s want to transition to cloud in a way that incurs the minimum technical debt and is sustainable long term. The greatest opportunities here are for technical experts that can understand the complex layers of legacy data infrastructure and how that interacts with cloud platforms. If you can map out data architecture, write rigorous technical documentation, understand data at the system level your skill set is in demand.

Career Paths: Data & Cloud Architects, Cloud Platform Engineers, Data Engineers

Regulation is a big deal

Regulation is often the elephant in the room, ignore it at your peril. Banking is a heavily regulated industry which has only grown in complexity since the financial crisis of 2008. All companies are under increased scrutiny on how they are managing data with the introduction of GDPR in 2018. In 2021 regulators dished out massive fines for companies failing to comply.

Amazon was fined $877 million for the way it collects and shares personal data via cookies*.

WhatsApp was fine $225 million for not properly explaining it’s data processing practices*.

Caixabank was fined $7.2 million for violating GDPR transparency requirements and not establishing a "legal basis" for using consumers’ personal data*.

*Source:https://www.tessian.com/blog/biggest-gdpr-fines-2020/

You might have noticed here that some of the largest fines have been issued to companies that have reached machine learning maturity (relatively speaking). This is the danger on the horizon for big banks, many of which are still early in the adoption phase for ML and big Data infrastructure. As the data ecosystem matures, regulatory challenges for banks will grow in complexity and the risk of colossal fines will increase.

There is also the question of fairness in AI, which is a ticking time bomb. There are numerous machine learning models out there that have been trained on biased data sets that are due to face regulatory crack down.

If you get the chance read Weapons of Math destruction (Cathy O’Neil), it serves as a warning of some of the issues of AI and fairness that are on the horizon.

Opportunities: Large banks need experts that not only understand data regulation but understand machine learning too. In banking specifically, explainable AI is key. Most models will not reach production because there are regulatory requirements for decisions to be explainable. Skilled operators that can build high performance, explainable models will be key here too.

If you’re not convinced about the size of this opportunity just have a look once again at the fine Amazon received in 2021. What’s not stated here is the reputational damage, which for bank’s is often greater than the immediate financial impact of a fine.

Career Paths: Legal & Compliance specialists, Statisticians and Data Scientists

The right machine learning expertise is rare

I believe the two sides to this challenge lie in prototyping and scaling machine learning. Let’s look at prototyping first. It turns out that it’s not so easy to spot viable machine learning use cases in the wild. In incumbent banks, the challenge here is in the size of the organisation and domain knowledge required.

The second head is about scaling. Large incumbents serve customers in the order of tens of millions with the total volume of customer transactions in the billions or even trillions. Delivering machine learning models that operate at this scale presents engineering challenges. Scale here goes beyond volume as often data is being generated across a fractured banking ecosystem.

Opportunities: For the prototyping phase, there is a huge opportunity for skilled data scientists and analysts. You will need the ability to "cut through the mess" quickly. You should have a framework for formulating suitable machine learning problems and the ability to rapidly develop prototypes. Data scientists with some domain knowledge in banking and finance will win here (this is how I landed my first Data Science job).

For the scaling phase, it’s more of an engineering challenge. A skilled machine learning engineer will understand how to take the prototype built by the data scientist and develop it to operate at scale.

In practice there is an overlap between prototyping and scaling. Imagine this scenario: you have hundreds of models in operation and now some of them begin to drift in performance, what are you going to do? Well, a skilled data scientist should understand why those models are drifting and direct training appropriately. Or one might discover that the drift is being caused by a corrupted data pipeline requiring an engineering skillset to resolve.

Delivering effective Machine Learning at scale requires constant collaboration between engineering and data science for model delivery, performance and monitoring. The overlap here is an area now being commonly referred to as machine learning operations (MLOps for short).

Career Paths: Data Scientists, Software Engineers, DevOps, Data Engineers, Machine Learning Engineers, Business Analysts


Final thoughts

The ML revolution has brought with it new challenges to banking and opportunities for all flavours of data expert. Although there is a lot of hype around data science right now, in actuality all flavours of data experts are needed. So, play to your strengths, if you’re highly analytical and understand the scientific method, pursue data science. If you’re more interested in systems and software, pursue engineering. For regulatory challenges and fairness in AI, choose the legal and compliance route. If you have domain expertise in banking, consider taking a course in machine learning to increase your market value.

I’d like to leave you with this thought, right now machine learning systems require a human in the loop. Instead of replacing us, ML systems have created jobs where we must understand them.


⭐️ I love helping people by sharing my experiences of data science in the wild. If you’re not already a member, consider subscribing to Medium to receive more helpful content from me.

Join Medium with my referral link – John Ade-Ojo


Related Articles