“I think it’s safe to say that we all understand that in the brokerage world there are numerous strategies at work and they all work around the notion of either A-booking or B-booking clients – or a combination of the two” says Paul Foley when looking at the application of AI in the electronic trading industry
Paul Foley is a seasoned CIO working in Financial Services. Based in Cyprus, he is a senior executive at TCG Europe, which acts as a shareholder for companies working in IT, real estate, legal services, commodities and portfolio management.
I’m going to start by asking a question – Do you know what AI is and do you really need it?
Or alternatively is it a buzz word that makes you think of star trek, robots and teleportation devices?
For the millennials this is probably enough to start the ‘tutting’ and head shaking – what’s next, “Medicine, do we really need it?”.
I’m quite fortunate in that my degree includes AI and I’ve had the opportunity to work on global AI projects and lots of automation projects.
So why the question?
In the SME environment (small to medium sized enterprise) we typically have a very dynamic IT function with some really enthusiastic people.
This enthusiasm and desire to strive for greatness is a doubled edged sword – on one hand it ensures that we have access to a source of innovation and positive company culture but on the other hand it guarantees that we will have people in the office who want to build the death star when a shopping trolley would do the trick.
Perhaps a better question to ask before we go into technology would be what are the functions within an SME that are the most time consuming, have an associated benefit, have an associated risk and also make a direct impact on the company’s profitability?
Within the context of brokerages, we have the following functions:
Of the above list we know that the support function is probably the most time intense function and is the one place where it is possible to destroy a company’s reputation whilst also having one of the most diverse ranges of skills within the company.
If we look at the typical brokerage we see that the support function has a live chat option and a telephone number for clients to call – so two channels through which the client might get support. There’s a problem with this support model though – inconsistency. Depending on the time of day support may be busy and the skill level of the available staff might vary.
There’s also no way of your HNW worth clients being identified and getting the appropriate attention (they have to queue with everyone else at least to start with).
The smarter broker will have augmented these two channels with an automated online presence – this then allows the brokerage to prioritise clients, use channel switching as a strategy to manage the incoming clients based on predetermined criteria and to provide an overall better client support function.
We know that there will always be the need for humans in the support function but how can you make the channel switching paradigm better? Chatbots.
At the time of writing there are several different types of chatbots on the market and there are also a number of frameworks available which allow for the creation of your own – this is where we come back to the initial question about AI.
As I mention chatbots use machine learning to achieve NLP (Natural Language Processing), what I didn’t mention is that machine learning comes in a number of different flavours – these are supervised learning, unsupervised learning, semi-supervised learning and reinforcement learning. Each of these methods has it’s own benefits and drawbacks (I’m not going to go into them now though as I’d like you to keep reading).
In the chatbot arena the current best of breed are the systems that use supervised learning. The way that these systems work is that when they are installed the administrator is asked to provide both successful and unsuccessful conversations (ranging from one liners to multiple chained questions).
This then allows the chatbot to learn from the existing data and to build its own rules for future conversations. The admin gets to customise the bot in a number of other ways as well – so specific vocab, language/cultural norms, etc. (the degree of customisation depends on the specific bot provider).
Most importantly the administrator gets to define ‘off ramps’. In Chatbot terms an off ramp is a place where the bot has to redirect to a human operator – an example may be that the client has asked the same question 3 times or the bot can’t classify the clients request. This then allows the company to ensure that the bot doesn’t destroy the company’s reputation (the human operator can still do that). The use of AI with NLP ensures that the system learns and can infer meaning.
So what does the company gain from using a bot?
The company instantly has a channel switching strategy in place, can deal with clients in a more consistent manner (the knowledge developed in each chat is applied to subsequent chats by the bot), the number of clients that the bot can deal with is scalable – so a sudden influx of clients can be dealt with immediately thus providing the clients with a very positive experience (compared to waiting), the definition of multiple off ramps allows the company to manage it’s risk and also allows the company a method for dealing with HNW clients (straight to human operator for balances over 200k for example).
The chatbot idea could also be taken further with the likes of Facebook ‘M’ and incorporated into the marketing effort – but that’s something for another day.
In this example we can see that using a bot is a good way to replace the human in the function (to a certain extent) but what about the other areas?
The sales team are effectively the face of the company to the majority of clients. They may never have the need to talk to support or dealing, they wont speak with anyone from IT or marketing so the sales team are the only face they know.
The sales team need to be dynamic in their approach – what works for one client wont necessarily work for another. So how do you make your sales budget go further?
We’ve seen that the support function can use a bot as an entry point to the company so what about sales – is there a bot for that?
In the old days (pronounced 2016 and before) if you wanted financial advice you’d either talk to your dad, a financial advisor or have an off the record conversation with someone from sales (options 1 and 3 are not to be recommended unless you have an independent financial advisor in the family).
The IFA (independent financial advisor) role has now been automated into something called a Robo Advisor – they are still subject to the same limitations and regulation as an IFA but are a bot. The early robo advisors were little more than a set of reports with a simple questionnaire but now the later generation of bots are AI based and whilst they provide a simple front end for the client to converse with the analytics in the background are becoming increasingly complex.
From a sales perspective the bot is a tool that is capable of standing on it’s own for a large client demographic or being used as a support tool for the sales staff for other demographics. If your company is acting as an investment manager or promoting the activities of several IFAs (or trading strategies via a mamm for example) then the robo advisor can be connected to the trading platform and show individual strategies or portfolios of strategies in near real time.
I should mention that a robo advisor is again something that you could quite happily build inhouse if you wanted to (although you probably wouldn’t have the same level of sophistication as a commercially available product on your initial release).
We’ve now seen that a bot can be used to triage all incoming clients for support and that a bot could be used either on it’s own or as a support tool for sales but what about high risk areas such as dealing?
I think it’s safe to say that we all understand that in the brokerage world there are numerous strategies at work and they all work around the notion of either A-booking or B-booking clients – or a combination of the two (A book – client’s orders go straight to market, B book – client’s orders are taken internally and never reach the actual market).
If a brokerage sends all of its clients orders to market then it has no risk and has no reason to care if the client wins or loses – it’s acting purely as a broker and taking it’s cut of the transaction in fees.
If the brokerage is B booking it’s clients (or a portion of them) then the brokerage does care if the client wins or loses – primarily because if the client wins then the profit comes out of the brokerage’s pocket and conversely if they lose then the brokerage pockets the loss.
B booking opens the brokerage to risk and there are many different strategies that the dealing room might use to offset this.
So how can the dealing room take advantage of AI in order to better manage risk?
One of the things that AI in general is very good at is identifying patterns, in this case that could be patterns based on currency fluctuations and user behaviour.
If we imagine that our brokerage has a an exposure to USD to the tune of 3 million, it’s Friday morning, mid month then the dealer who is responsible for USD based pairs might employ any number of strategies to offset the current exposure (such as manipulating prices on sell prices, changing different client groups to A book instead of B book or even B booking additional groups who are currently trading against the current exposure – I’m not a dealer, this is just a simple example of what they ‘might’ do for illustration purposes).
There is another thing that the dealer could do and that’s use an AI based tool to identify trading patterns. There are a number of tools on the market that will look back at trading activity over a given time frame and then try to predict trading trends – so the tools aren’t trying to predict market movement but rather trader behaviour.
This information can then be used by the dealer to decide which strategy to use in order to best manage the current exposure (the tool might identify that on Friday afternoons for the last month traders have tended to go short USD and so any attempt to offset the current exposure will be reversed within a few hours).
In the case of high risk tasks we’re seeing a trend across financial services in general toward support interactions rather than replacement interactions.
One of the key considerations that I think is worth emphasising when we consider AI is that of data. The more data you have (clean data) the better AI solutions are going to be able to perform – smaller data sets or dirty data sets will ensure that an AI system does not deliver or actually makes things worse.
So to go back to the initial question of do you need AI I think it’s clear that whilst you could argue that you don’t ‘need it’ there is the possibility that AI could help you grow your business and reduce your risk.
It’s also possible that you could develop your own systems using any number of available frameworks but if you’re an SME it’s probably better to buy an off the shelf product (there’s still integration work to do). There’s also nothing to stop you putting in place a hybrid strategy of AI augmenting human operations.
In case anyone’s wondering why I wrote this piece it’s because I’ve spent time talking with people over the last few months and whenever I mentioned AI I either got a “we’re too small for that” or a “can it do everything?” style response.
One of the key points that I’ve mentioned before is that the financial services market is a great place to operate BUT it’s a competitive environment – embracing new technologies will allow you to grow your inhouse capabilities and manage your risk in a more effective manner whilst also allowing you to increase your service offering.
The subject matter and the content of this article are solely the views of the author. FinanceFeeds does not bear any legal responsibility for the content of this article and they do not reflect the viewpoint of FinanceFeeds or its editorial staff.