Coffee and a catch up: Don
We catch up with Don Liyanage, Research Engineer in AI about his role at Flexys.
What is your background?
I have been fascinated by the world of programming and computing for as long as I can remember – but funnily enough this firstly manifested itself through electronics! When I was just 13 and living in Sri Lanka, I built (and sold for a profit) a custom water pump cut-off circuit that prevented water wastage when pumping water from a borehole for domestic consumption. Not a typical teenage hobby I know…
A few years later I moved to the UK to complete my degree in Computing and Information Systems at London Metropolitan University before completing my Masters in Mobile Computer Systems at Staffordshire University in 2010. I now have over 12 years’ experience in Computer Science and Mobile Computing.
I am currently carrying out a PhD in “Artificial Neural Network Reasoning Using Case Based Reasoning System” alongside my work at Flexys, which I have been doing part-time since my Masters and I am due to complete in 2019. I am also a Professional Member of the British Computer Society, Associate Member of the Australian Computer Society and a Member of the Internet Society – so I guess you could say I’m keeping myself busy!
How did you hear about Flexys, and what does your role involve?
I joined the Flexys team in November 2017, having come across the company and its vision for the world of digital collections via a LinkedIn post about explainable AI. At the time I was working as a technical specialist/architect at Nationwide Building Society, where I spent a fantastic 17 months providing technical advice and consultancy to internal clients and management on matters of technical architecture solution design and development.
As a Research Engineer at Flexys, I work closely with our CEO Jon and Chief Research Officer Joe to conduct applied research as part of our widespread R&D work. In order to truly innovate, as a company we must interrogate and investigate the subject matter at hand, and that is the purpose of my job. In terms of how I set about achieving this, I think about actual business problems and try to develop AI based solutions, firstly by building a mathematical model, and then by inserting the model into a prototype environment. Here I can simulate most of the problems that could arise with the mathematical model which is a good gateway to figure out if it will work or not. If it all looks good at that stage, from there I work with the development team to implement the prototype into an actual product.
What is a typical working day for you?
I drive from my home in Gloucester to start work bright and early at around 7.30am, traffic dependent. I like to get into the office at this time to give myself time to plan out my work and focus on any particularly complex areas without any distractions. I am a multi-tasker, so I’m comfortable working on a variety of things at the same time, but naturally I base my work plan for the day around priority at any given time.
When I really need to concentrate I sometimes listen to music or take a walk around the building to aid my thinking, which might sound simple but really works! Everyone’s thought processes are different – walking is the trick for me. I then finish work at around 4:00pm and head home to work on my thesis. I am a great believer that sleeping helps ideas, so if I go to bed with a problem or task unsolved in my head, I find my brain works on it overnight and gives me an output to wake up with. I have a busy schedule, but I love what I do – if there’s a will, there’s a way!
How do you think that machine learning can help to enhance collections solutions?
I think it can be very influential in a few key use cases. Collections departments are normally considered cost centres. If we can minimise the cost and maximise the amount of collection, with happy customers throughout the process, that’s a great achievement.
In terms of methods to do this, a great focus must be on improving self-cures (the customers who genuinely forgot to pay a bill but will pay the balance willingly when they realise). An agent having to track this kind of customer is a cost to the collections department which can be avoided. If we can predict and classify those types of customers, the dial list will go down together with the workload of the collections agents, freeing up their time to service other customers.
A further use case is in identifying vulnerable debtors who for reasons of age, health or disability are unable to pay the outstanding amount – a hugely important capability and one that holds great moral value. Also, we can calculate the sentiment or emotions of customer responses using a sentiment analysis. An agent can then adjust their tone and the way that they engage with the customer to reach a better debt resolution.
How do you think future developments in machine learning could have even greater impact on the enhancement of collections solutions?
In terms of the Flexys roadmap, if we can further develop and fully implement our Multi-Level Engagement project – the purpose of which is to identify the sentiment variance and relationships between multiple engagements in a digital collections process – it will give us capabilities to automate the collections process further. This essentially consists of in-depth AI analysis that would enable us to provide reasons why the system took a specific path in the collections process, allowing organisations to be able to achieve an unprecedented level of transparency. If there is any kind of miscommunication or dispute throughout the course of events, the organisation will be able to refer to the AI-powered route and its justification. This would bring about huge change in the world of collections.
What do you most enjoy about your job?
When I have a hard nut to crack. For me there’s nothing better.