You cannot boil the ocean with AI – Matthews

Financial services is no exception, with predictive and generative AI already delivering transformation, efficiency and disruption. As a sector, we are seeing fantastic use cases, whether it’s intelligent chatbots, vulnerability detection, data analysis or decision-making. The promise and potential is real, but so is the disappointment – especially for firms that focus on the wrong areas, or fail to put the right foundations in place first.

At a recent conference, the AI narrative was all about redoing your website at the ‘touch of a button’, giving it a quick prompt or some data and it will give you all the answers. It’s like the predictions of Tomorrow’s World have finally been answered – the future is here and it’s exciting. When you apply this to a business need, it can be really compelling, but without the necessary work around it, it all feels a little superficial and doesn’t offer a true return.

The problem is that AI is seen as this silver bullet – the answer to all problems. In reality, it’s a toolkit, and one that relies on having all the right building blocks in place first. Without this integral step, you’re really building houses on sand.

Doing the housekeeping

Before any meaningful integration, you absolutely need to do the housekeeping first and do all the below-the-line work – the unexciting but highly essential work that gets the house in order. In my view, this all starts with data. You need to get your data structure and data strategy in place to ensure it’s well-categorised, it’s consistent and ultimately coherent.

A clear approach to taxonomy, integration and internal governance prevents cluttered data and poor processes. If your team is using terms interchangeably or relying on multiple versions of the same data set, a large language model (LLM) is flawed at the outset. Once you get this right, you can then start to look at the above-the-line work and put some clever stuff over the top.

The doomed retrofit

It’s also critical to consider where your data is stored, the systems it flows through and the platforms and tech stack you plan to use AI with. This is particularly key where firms are still relying on outdated legacy systems as their backbone. Trying to retrofit cutting-edge AI solutions into these archaic infrastructures is simply incompatible. It’s like trying to open TikTok on a Nokia brick.

I will not apply AI across any legacy systems. It would be an absolute car crash and would send back so many crazy results that we would get absolutely no benefit from it. We can’t just take what exists in the ecosystem today and hope it will all just happen – it has to be much better thought through.

The same excitement that comes from playing with AI must be applied to digital transformation to really unlock the potential. We need to identify shortfalls in our data strategy and weaknesses in our tech stack that are stifling the ability to innovate and to scale at speed. We can then start to look at where upgrades are needed, what can be consolidated or what is just more efficient to outsource to an expert partner.

The right approach

While it is clearly making waves, we cannot think that we can boil the ocean with AI. We cannot view it as the answer to all our problems – especially when a business is ill-equipped. The risk becomes that we do too much, too fast, and we end up making things worse.

It needs to be targeted to particular use cases where it can demonstrate genuine value to the business, colleagues and eventually to clients. Don’t go all guns blazing on a full roll-out until you can prove it works in one part first.

Fundamentally, this is the process we have adopted at Target. We first looked at how we can leverage AI internally to solve operational and business problems. Our first solution was chatbots to streamline our internal processes and best support colleague queries. We then harnessed those learnings to create further proof of concepts and also leveraged AI innovation among current vendors and our existing platforms to make improvements to our service.

We are seeing fantastic examples of AI delivering genuine value in financial services – most notably for us, through the telephony system in our contact centre. Not only are call notes now automatically transcribed, but we are able to monitor 100% of customer calls and interactions, and provide agents with live guidance – particularly useful for identifying vulnerable customers – and instant feedback.

There are countless other examples of where AI is changing the game. We are seeing it ourselves as it enhances the service we can offer to our clients. But rather than diving straight in, we need to get our house in order, do the groundwork, make the necessary changes and investments and then apply AI where it can deliver real value.