How fintech wins by adding a little UX in their ML
The rise of mobile banking apps provided the world of fintech with an enormous stack of data. Valuable data ready to assist your customer to make better-informed decisions. In that context, Machine Learning presented itself as the go-to technology to fulfil that mission. But the sheer focus on ML as a tool to automate didn’t always lead to better-informed decisions. Why? We believe it is because you can’t talk about ML without adding UX in the mix.
The true power of ML in fintech
When we talk about Machine Learning, we use it as a collective name for modern data-driven AI solutions. It’s often synonymous with recommender systems, natural language processing and computer vision. But ML also has an essential role to play in transforming a customer’s raw banking data into financial insights.
Perhaps not in the way you might think, though. ML in fintech should not only focus on creating hyper-personalised products and recommendations for customers. It’s more about helping users spot patterns and make better, informed financial decisions.
Users don’t want the help of a chatbot assistant on every screen of their app. They want help when they need it. When there’s value for them. It’s not about creating an ML-based tool just for the sake of Machine Learning. It’s all about finding the sweet spot where ML can bring real value to your customers.
"ML in fintech should not only focus on creating hyper-personalised products and recommendations for customers. It’s more about helping users spot patterns and make better, informed financial decisions."
ML has trust issues
Let’s face facts. ML still hasn’t won the complete trust of your user. And we need to acknowledge that. Even if ML solutions will fulfil your users’ expectations 95% of the time, it’s that 5% of failure that creates a breach of trust. Especially in the context of banking apps.
Just think of a recommendation for pension saving in your banking app. What if you already have one, only with another bank? Or you just don’t need it? It isn’t possible to notify that this isn’t relevant to you in almost every banking app. And the recommendation will continue to appear, which leaves your user irritated and powerless to do anything against it.
Choices sometimes seem made by the roll of a dice or by a system that your user can’t challenge. Unexpected results, provided by your ML solution, have an impact on the trust of your customers. And there’s nothing they can do about it. That’s why it’s time to add UX into the mix.
Not a hard truth but an opinion
MLUX responds to the looming AI product winter - in which the interest in AI decreases because it’s simply too hard to integrate well. The goal of MLUX is twofold. First, to create an effective interface between AI and the user. Second, to build up trust and manage a user’s expectations.
The key takeaway from MLUX is that we should not look at the output of an ML algorithm as a truth. But rather as an opinion. An opinion that’s biased by data and processes used to train the algorithm, as well as the context it is deployed in. Your user should always be able to challenge this ‘opinion’ and remain in control over the results.
“The key takeaway from MLUX is that we should not look at the output of an AI algorithm as a truth. But rather as an opinion.”
Finding a balance between automation and control
Naturally, ML is all about automation. But beware of this pitfall. We tend to think of automation as a characteristic of a system. It either automates a task or doesn’t. But automation is more like a spectrum. As we move from zero automation to augmentation and further onto partial automation until we reach full automation, we gradually shift the responsibility from the user to the system.
You may think full automation is the holy grail of every ML task. But that’s not correct. In a fully automated system, your user is never the truly responsible person unless we give them the tools for feedback and control. For every ML use case, it’s about finding the right balance between automation and control.
Make your business process about your user
If full automation is not the way to go, then what is? In essence, you need to enter a co-creation mode where you leverage the power of your AI system in a way that gives your customers superpowers. In other words: augment the user experience with data insights. Luckily, you don’t have to start from scratch. A lot of the business processes for these use cases are already in place. They just have to be repurposed to put the user at the centre of the value. Let’s take a look at some examples.
MLUX in practice: financial simulations
Financial simulations are also a form of ML. For instance, when you take out a loan in order to buy a house, personal information like financial stability, BMI and instalment period is combined with external information like the health of the housing market & financial market, to predict the interest rate you will pay.
This semi-automated business process is currently owned by the bank clerk. But in some banking apps, this ownership is shifted to the end-user who is able to make simulations on their own. When the UX supports this, users can fully personalise their offer. But they can also ask “what if” questions and explore different scenarios. The result is a user who’s confident about the final configuration of their simulation and who trusts they got an optimal solution, tailored to their specific needs and context.
Face recognition
In fintech, we tend to only use and accept technologies which have a low cost when it goes wrong. Or where the performance is high enough to allow for a high level of automation.
Take fingerprint scanning or face recognition, for example. It might not seem obvious but the ML needed to perform the pattern recognition is a real work of art. It’s not trivial to detect faces in multiple positions with only a few training examples. But its true power comes from how the algorithm handles uncertainty and mistakes. It balances value and the cost of being wrong by putting its emphasis on authentication.
When it’s uncertain, the algorithm will prefer to keep a person out and ask to retry rather than granting access. Next to that, when the authentication fails multiple times, a passcode is still an effective fallback to avoid locking out a valid user.
Bank transfers via OCR
Another technology with a similar balance between value and cost of being wrong is OCR - used for digitising bill processing to perform banking transactions. The value resides in speeding up the process by using OCR to automatically fill in all relevant fields. There’s a clear advantage of using the feature when everything goes right. But when a mistake is made, the user can still adapt the result in a fraction of the time of doing the task manually.
It's not about ML. It's about your user
The value of a data-driven approach is to automate dull tasks with a low cost of being wrong while augmenting the thought process of more creative and context-rich environments. The end goal is to help users feel comfortable using ML solutions in their banking experience and set the right expectations. It’s not about reinventing the wheel here. It’s all about putting your user front and centre. About finding a balance between automation and control.
