Over the past decade, the entire notion of what a trader does has changed radically. Core functions involving price discovery and execution have been transformed by automation and artificial intelligence. The race to zero latency, the rise of algorithmic trading and the development of sophisticated order routing products have turned significant aspects of the trading function into a question of who has the superior hardware, software and network infrastructure.
We can trace much of this change to intense competition among trading firms looking to both cut costs and boost revenues through better offerings. That has spurred waves of innovation, with automation and artificial intelligence (AI) at the core of that progress. But there is a major part of the front office that has not undergone such a wholesale revolution: the sales function.
Imagine a sovereign wealth fund or a large asset manager calling its investment bank and asking for an account overview as it gears up for a strategic review. Given the nature of sell side institutional structures, and the way that so much data ends up in silos, producing this kind of holistic analysis could take days, if not weeks.
Yet in so many other respects, we expect financial services to produce instant results. That, after all, is what we’re used to in our personal lives as we use apps to consolidate disparate streams of information.
Perhaps the best analogy is the Internet. When you use a search engine, you are asking it to hunt down relevant data from thousands and thousands of places and bring it back in some prioritised form. Text, images, video – whatever fits your search criteria. That then forms the starting point for you to do your own analysis.
Could automation and AI bridge the gap between what we experience as consumers and what a buy-side client experiences when it wants to get the big picture?
In a word, yes. The building blocks are all there. To extend our earlier Internet analogy, what the sales function needs is an enterprise-wide search engine that can find all of that data, in its various forms and locations, and then apply AI to it.
First, it’s worth noting what the sales function entails. The whole purpose is to help buy-side clients do their jobs, letting them know about products, services and analysis that will allow optimum portfolio management.
But to do that, you need data, and that’s one of the reasons why sales has been more stuck in the past than other functions. A sizeable chunk of a salesperson’s time is taken up with assembling key data, which is often spread out across the enterprise and sometimes captured in hard-to-use formats such as voice. Sales people become data managers rather than relationship managers.
The goal, then, is to make the process more efficient. So, for instance, a salesperson might want to look at his or her customers and see which ones are less profitable and then try to identify ways they could become more profitable. Conversely, if there are some clients who are performing well, could one understand what’s driving that performance and see if other customers could benefit?
This is the sort of task where AI can help comb through data and maximise a salesperson’s performance. The trick is in recognising what AI is good at. On the trading side, AI has proven to be exceptionally good at tasks like execution and pricing. On the sales side, the opportunity lies in recognising that AI excels at making sense out of large amounts of data – in all its forms – by identifying hidden patterns and correlations through trial-and-error analysis.
A prime candidate for reaping the benefits of AI would be voice data. A lot of work at institutions is underway to capture voice-to-text data and store it. For a range of both front and back office functions, this process alone holds promise. Now imagine overlaying that with AI in order to make sense of the vast amounts of voice data you’ve assembled.
What this boils down to is taking unstructured data, making it structured and then using machine-learning software to analyse it. Something that previously was not visible – or would have taken extraordinary amounts of time to become visible – is suddenly available and part of the relationship management process. For instance, using AI-based methods, a salesperson could identify and review the texts of all client interactions of a certain type over a period of time to identify common client problems or issues, or even trading opportunities. If a certain customer profile is finding success based on a product or market, could that be applied to other clients with similar characteristics? The trial-and-error approach of machine learning lends itself to finding this kind of information.
It’s hard to imagine life without the Internet these days. Similarly, there will come a day when it will be hard for people in the financial services industry to imagine a salesperson without instant access to a much broader and well-assembled pool of data, one that will make the business of servicing clients look radically different than it does now.
With the right data architecture and the right tools, it’s not hard to envision such a transformation for the sales function. On that note, it bears remembering that the actual technology for the Internet itself had been around in the form of ARPANET for a couple of decades before the web really took off. In the same vein, the building blocks for an AI-led sales function are there now. It’s just a question of time before savvy trading firms start to use them.