Navigating the turbulent waters of Artificial Intelligence (AI) and Machine Learning (ML) can seem like a daunting task to the uninitiated. In fact even the question of how AI relates to ML is answered differently depending on who you ask, as evidenced by the numerous articles about on these topics. In this area, confusion abounds – for example with ML being linked to predictive analytics, including Monte Carlo Simulations, which have nothing to do with ML!
Some of this comes from the breadth of subjects that are related to these concepts. For example Natural Language Processing, Random Forest, Dimensionality Reduction, Neural Nets and Deep Learning do not fit into a nice grouping structure or hierarchy which can defined as just an instance of that overall class of technique. The next layer of complexity comes from the fact that any given use case can actually use a fairly arbitrary combination of these tools to achieve its aims. This general applicability of tools and techniques does however provide a glimpse of how to think about mapping out the AI and ML landscape.
Ultimately all of this is a consequence of historic research going back 30+ years, without ever having mainstream applications, therefore the techniques are having to be understood by a large number of ‘non-experts’.
If we have to assemble our solution from a non-defined collection of tools in the toolbox, how can we expect to understand or “get our heads around” the overarching landscape. One answer is to look at the business drivers behind each use case. The business drivers (or the “Why” of the project) are usually from the ‘benefits’ side of the business case. The benefits we are striving to realise are however often a neglected part of a project. Some methodologies (e.g. Managing Successful Programmes) place particular emphasis on them and are the critical and measurable link between implementing your system and the overall corporate strategy.
We could suggest that the relationship between these elements is as follows:
Implement an AI use case using these techniques so you can derive these benefits
So what are the benefits for an AI / ML project?
Looking solely at Capital Markets AI use cases, we have identified that each will fit into one of the following three categories of business benefits:
- Efficiency (faster and cheaper)
- Insight discovery and new services
- Cognitive (more convenient)
This business driver covers the cost reduction initiatives that are behind headlines such as “AI is going to replace all our jobs!”. The cost reduction is achieved through automation and thus freeing up expensive human headcount. Obviously Robotic Process Automation (RPA) falls into this space as well, as on one level this is just the next chapter of Business Process Optimisation which has been on the agenda for years. Spun round the other way RPA allows an organisation to dip its toe in the water of Robotics and automation, acknowledging that it is a journey that will at some point end up in AI / ML. The challenge as ever with cost reduction programmes is with the change management i.e. how do you make sure you see the reduction in cost and not just the same team with an additional system?
2) Insight discovery and new services
Traditionally these use cases start off with someone having a dataset that they feel should be able to yield some actionable insights. As such this is a classic data science project with an initial proof-of-concept or experimentation phase, based around a one-off cut of the data. Once the mature hypothesis is proven, then the project becomes one of creating a new service based on this new understanding. A good example of this are recommendations for more optimal hedging trades based on historic data.
3) Cognitive (more convenient)
Cognitive is currently a very fashionable term in the industry that is used to cover many parts of the AI / ML landscape (and beyond!). We define it as making the Human / Machine interaction more convenient. As such this covers ‘chat bots’ in the manner that you can get a response straight away and driverless cars, which allow you to do other things with your driving time. To improve the human’s experience Natural Language Processing and image / video interpretation are also heavily used. As we can probably all recognise, the challenge here is to make the system seamless to avoid user frustration.
However, the best use of AI is of course likely to be one that meets all three of these areas. For instance a chat system that allows a client to fix broken static trade data themselves, automatically freeing up capacity from the operations team!
As if staffing and delivering an AI / ML project is not hard enough, achieving the true benefits requires focus all the way through the implementation. Understanding from the start where your use case sits in the overall business model, and the challenges you are likely to encounter on the way from concept to go-live is essential.
It is clear that in all areas of business, but particularly in financial markets the benefits of AI and ML will be huge. The key however, is to ensure that the right tools and techniques are applied to the right problems to achieve the most beneficial transformational outcomes, without simply creating another bigger better mousetrap.