Making the most of machine learning

Head of Futures, BAE Systems Applied Intelligence Read time: 4 mins
How can machine learning strengthen the use of data and alert investigation? We’ve got some ideas but we want to hear what *you* think too. Matt Boyd explains
Making the most of machine learning image The current health emergency ricocheting around the world shows how challenging it can be to predict future events – but that doesn’t mean we shouldn’t do it. On the contrary, it could be argued that the escalating impact of the COVID-19 pandemic underlines just how important it is to be ready to tackle whatever might be lying over the horizon.
At BAE Systems we set up our Futures Team to do just that. Using our own formula and score boarding system based on lean start-up and design principles, my colleagues and I spend our time helping organisations face up to the challenges of the future. For example, we’re working with Digital Catapult to explore how start-ups and scale-ups can bring significant incremental benefits to our existing products, services and solutions in the area of machine learning. 
So, why are we doing this?  Well although we already widely use machine learning in our core “analysis” capabilities, we think machine learning can bring significant benefits to the processes which sit either side, namely data ingest and alert investigation.  And rather than turning our own data scientists to look at this, we’re keen to access the wider innovation ecosystem where they are already looking at these challenges, to bring a breadth of the most up-to-date techniques to our customers working with the most effective partners.

Accessing the data

Our products tend to be deployed to clients who operate in complex environments. We need to collect data from a range of the systems within those environments to allow us to, for example, detect financial crime. These source systems will have grown via internal development, as well as through acquisition, meaning that it is common for customers to have multiple systems with similar types of data.
But this isn’t ideal. To accurately detect activities such as financial crime, we need to ensure our systems are getting a complete view of the data available, as well as mapping each of those data sources into a common representation which we can use for our analysis.
Although every customer’s environment is different, there’s a lot in common – for financial crime, we’re dealing with customer records, bank transactions or insurance policies, and the same types of entries – names, addresses, phone numbers and so on.
Within this commonality, context is really important. For example, an insurance claim may have multiple addresses representing the policy holder, the driver, witnesses and so on. When we’re mapping data, identifying addresses is just the first step, we also need to ensure that they’re also going into the right places to ensure the results are valid.
So we think machine learning can have a really powerful impact on this space, for instance in accessing, organising and mapping the multiple data sets that are needed.
We’re excited to see what capabilities already exist out there around this space.

On alert

A number of our products are used by our customers to triage or investigate potential cases, for instance of financial crime, which are typically triggered by one or more alerts from an analytics system.
Efficiency improvements can offer significant cost savings or detection improvements. As such, there is currently significant interest in using robotic process automation to automate parts of the analysts’ workload. However, in many cases these processes have to be manually configured. There are also concerns about how “fixed” these automations are, and their ability to react to, and reflect, the individual nature of the cases involved.
Given the ever-changing patterns of financial crime, our goal is that our software maximises the time the human analysts have to understand patterns of behaviour and the risk it poses to the business while the computer handles the mundane or repetitive tasks for which it is well suited.
So this also feels like a really promising space for machine learning to accelerate these tasks in a flexible and non-constraining way and empower the humans to take better and faster decisions.

Challenge time

We see significant opportunities for machine learning in both these spaces.  For example, customising automation for different communities – in the extreme, customisation to individual preferences – or perhaps capturing knowledge from more senior and experienced analysts to feed into the activities and decisions of the more junior levels, and improving performance over time.  But these are just a couple of options out of many ways that machine learning could make a difference.
Working with Digital Catapult, we are inviting start-up and scale-up businesses to consider where machine learning can help in these areas. Following a rigorous selection process, the best will be pitching to a BAE Systems audience in mid-May. And after that we aim to open up collaboration relationships with the most promising organisations with a view to bringing joint capabilities to our customers.
We’re really excited to hear your ideas and suggestions, your thoughts and proposals. Let us know what *you* think – we’re looking forward to building some new relationships!

If you would like to get involved, please visit the Digital Catapult open call

About the author
Matt Boyd is Head of Futures at BAE Systems Applied Intelligence 

Read more from the Futures Team:

Matt Boyd Head of Futures, BAE Systems Applied Intelligence 30 March 2020