The pressures paving the path to AI 

Digital Banking is one of the hottest things on the market, and everyone out there is racing to prove that they can do it best. The competition is getting tighter as digital banks converge on the same technology and sets of features. In 2020, it seems normal to have a snappy and intuitive app, pop-up notifications every time some money is spent and a ream of analytics to boot. Behind the scenes, it seems like a no-brainer to build an army of microservices, process everything in real-time, collect large swathes of customer data and run everything in the cloud.

With broad technical conformity on the rise, banks will inevitably look towards the next differentiator – Artificial Intelligence, or rather, the proper application of it. Getting AI right can do a lot for a digital bank, from slashing its operating costs to accelerating the discovery of new opportunities to personalising the digital services that underpin the proposition.

Be good at process, not just algorithms 

With dropping computation costs and the rise of new model-building technologies like AutoML, the process of building a good and comprehensive model is becoming increasingly commoditised. Over the longer term, choosing and tuning the right algorithms will not remain as strong a differentiator. Instead, we’ll see robust development processes and good judgement become the lynchpins for making the right choices about AI and delivering on them.

Getting AI right is only part of the problem – a winning organisation must change its attitude, have a strong Data Engineering foundation, a robust Data Architecture, and a clear Data Strategy.  

AI behind the scenes 

This section is dedicated to the AI-wary (or AI-weary) executive. Behind the scenes, AI is not click-of-a-button pretty, and it certainly isn’t a magic box that’s fed data in exchange for miracles. It’s a tricky maze fraught with dead ends and requires a lot of thinking to get right. This leads me to my big reveal (and well-known industry secret) - the “fake it till you make it” approach.

That’s right. Numerous companies often use “placeholders” for components when developing complex AI systems, such as approximation rules and/or a human-in-the-loop approach. Approximation rules consist of one or more simple if-this-then-that style statistical heuristics. They require some data exploration upfront but can be a cheap way of papering over a problem. Human-in-the-loop is a fancy term for getting people to do the tasks expected of AI, with the goal of prototyping part of or all of a system. Does it sound like cheating? Yes, it does. But it’s better to “cheat” for a short while to stagger the complexities of building deep AI systems and remain competitive in the market.

Oh, and Picasso’s quote “Good artists copy; great artists steal” also applies to building AI systems. The really clever companies don’t build everything from scratch. It is important to know when to buy and when to build. Uber does not do any of its routing or mapping; instead, it pays a subscription fee to Google Maps (a very big one, by the way). In this context, leveraging pre-trained models offered by Google Cloud and Amazon Web Services can help to shortcut the path to delivering new features.

Finally, don’t sweat the small stuff. Chasing impossibly high accuracy in each AI model is not always worth the resources. People sometimes have an unfair obsession with model accuracy and treat it as they would an exam score. What they don’t often recognise is that they too make mistakes, meaning that their accuracy on a given task is far from 100%. With AI, you are thinking in bets. While guessing correctly is important, using those guesses wisely is more important.

What we’re doing with our Alpha Platform

Alpha Platform is our digital banking accelerator, built to help you design, assemble and deliver a new banking proposition to your customers within three months.

In Alpha, we are developing a host of exciting use cases, such as models that monitor each customer’s usage and detect friction in real-time, models that understand spending patterns and offer up recommendations, and even a model that offers tailored flash loans to keep business operating comfortably.

For us, the strong differentiators are our approach to data engineering and our user-centric AI design. We're geared up to collect user behaviour and their financial information in real-time and run it past a series of models to determine whether any intervention should take place. You can find out more here if you are interested.

(Before you ask, we have no humans in the loop.)

Pick your direction 

The use of AI in Digital Banking is still growing, and there are plenty of opportunities to outshine the competition if you are able to act quickly and be smart about the way you do it. Remember, respected tech-savvy organisations such as Netflix do not differentiate themselves by using superior technology, but by using technology in a superior way.

If any of this has resonated with you and piqued your interest (or stoked uncertainty), please reach out and speak to us. We have helped build digital banks, we’ve developed solutions like TrueVoice and Controls Intelligence that are designed to streamline back-office operations and manage risk, and we would be happy to work with you.