Propel@YH

Kevin Flanagan is a hybrid technology and finance professional having spent 25 years in finance and technology in roles as varied as lead programmer on the first networked bank systems in Ireland for a major Irish clearing bank and laterally as the first global head of Development at Barclays Capital. He is also serial technology entrepreneur with 2 exits from 6 startups and Co-Opts is his 7th company in 3 countries. He’s worked as a trader, risk manager and restarted Barclays Capitals Investment Bank in Japan. Kevin has designed and built several global systems that have run for more than 25 years. He brings experience in building artificial intelligence (AI), machine learning (ML) and natural language programming (NLP) products, and holds an MSc Intelligent Systems (AI) from UCL and an MBA from CASS. In his spare time he snowboards, reads and enjoys walking in the mountains with his family.

What was the idea borne out of/ what is the main challenge that the product is trying to solve?

The product was borne out of our experiences of the mental health systems whether as practitioners or as parents of patients. These were very unsatisfactory experiences and spurred us to investigate why that might be so for mental health. The research revealed some disturbing statistics. For example for depression and anxiety-related disorders, the number need to treat is 7 but twice that number recover spontaneously. That represents ⅔ of the 12m people who will have a mental health condition in the UK this year. NHS mental health services only gets to treat 2m of those people. An additional 4m typically get medication and computerised CBT but it is difficult to successfully treat complex mental health conditions in 12 minutes!

From this it became clear that mental health services don’t reach enough people and the services that do reach are not very efficient in terms of outcomes. We based this conclusion on our observation of a therapy called Open Dialogue which appears to be approximately 4 times better in terms of numbers needed to treat. There were subsidiary problems such as for clinicians in mental heath are in percentage terms, the most likely profession to use it’s own services and twice as likely to suffer burnout. In the NHS they suffer twice the sick leave rates and vacancy rates of any other clinical profession That is the problem we decided to try and address.

When did the idea first move into development?
It took almost 2 ½ years of work to get to the point where we felt we had enough of a handle on the complexities of the problem that we could start to work on the technology. Preliminary work had revealed that the technology wasn’t capable but a breakthrough in early 2020 lead us to start working on solving a specific problem which was being able to identify individual speakers in real time on a low powered edge computing device. By August 2020 we had solved this problem and then we applied for grant funding and were successful in getting an Innovate C-19 grant to build out a beta product.

Who is the main target audience for the platform?
Our main target for this is talking therapies, typically face to face but with the onset of C-19 we are developing a hybrid solution capable of interfacing to a video session and live speech from the therapists side.

Is the platform live? If not, where in the development stage is the project at?
No our platform isn’t live yet as we are still building the beta version of the product which is scheduled for April 2021. A key feature of our product is customised language models for combinations of sex and accent. This is to lower the word error rate and this is one o the reasons why voice to text hasn’t really gone mainstream yet because there is no such thing as an average accent.

How did you find the Propel@YH application process?
We found the overall process a fair bit of work. While the application asked for a lot of the same sort of information they asked it from unusual perspectives. I’m not sure we did as well with the video as we could have. Part of our problem is we are literally focused on filling a single field in a patient record. The implications of doing that well are potentially profound and perhaps even revolutionary. It’s hard to get the balance between hyperbole and grounded realism while selling a vision.

What have you found the most insightful or useful part of the programme so far?
The class on clinical safety by design has had a significant impact on us. Stuart Harrisson’s has helped us focus on the critical aspects of our products clinical safety in a field where the rules are still being created to some extent, the application of evolving AI models to healthcare. Given we are working in the mental health space it was fantastic to come across someone as knowledgeable and grounded and happy to advise us how to deal with navigating the regulatory and safety ecosystem around health.

What advice would you give to yourself if you were to apply again?
Having a rehearsed and canned video pitch would help a lot, good videos take an awful lot of work. Having a very clear value proposition is critical to tying all the elements of the application together. I don’t think we did that as well as we could have so worked more on that first would have helped make our case stronger.

What are you most looking forward to learning about in the programme’s masterclasses?
None of us have ever designed a clinical study so the masterclasses around this is where we think we will get the most out of the masterclasses.