If you say you’ve had a discussion with a law firm data science team, then in the UK that can currently only really mean less than a handful of law firms, the grandfather of which is BLM, an insurance specialist firm at the vanguard of using data to create client solutions.

Which is why we ended up at BLM’s London office to meet with head of analytics Andrew Dunkley, business improvement director Adrian Spencer and their growing team to talk about the trials and tribulations of learning the hard way how not to form a data science team.

Dunkley, respected by those who know him in the industry for having a ‘brain the size of a planet’, says that in forming the team they spent a lot of time at first working out how not to do things, including how to integrate the team within the law firm, what the governance was and how the team would work within it.

Spencer tells me: “We had a lack of appreciation of the hurdles – no-one had done it before, accessing data in this way. But our partners were saying ‘have you built it yet?’ There was an incongruity of expectation.”

For former director of IT Abby Ewen, who is now at Brown Jacobson, the challenge was trying to manage a startup culture within a regulated environment.

While clients want quick results, Dunkley says: “Our data was encrypted at rest right from the start, which slows things down when you are processing millions of documents.”

Another complication, inevitably, has been the challenges around hosting and Dunkley said: “We’re living in a world where a number of clients say their data needs to be on premises. I can understand that when we’re dealing with such sensitive information – such as records relating to children – we have to treat it with utmost respect and it’s completely understandable that they say ‘this is how they want you handle it.’ It just means lumpy processing: on premises you have to buy or rent space and keep it internally as opposed to spin up an AWS cluster.”

The great thing about working in this space? “There are a million and one things that people want you to do for them,” says Dunkley – and he means it in a good way, in case you’re wondering.

“Our clients are mostly insurers and we have a sophisticated client base but everyone wants you to get where you’re going really fast. Given the costs involved in hiring data

Scientists – they have an incredibly transferable skill set and you’re competing with everyone – you need to be really thoughtful in terms of how you choose the projects you’re going to go after. You need to be prepared to take risks and be aware that some won’t work.

“One of the best bits of our job is that people come to us with a problem and I don’t know how to fix it but I have some idea and come up with something powerful that suits the data.”

One of the challenges and attractions is the complexity: Dunkley says: “I read somewhere someone say, ‘we have decades worth of data we can analyse,’ and I thought ‘really?’ We distrust it. It gets old. Why do you even have in light of GDPR? We have to think ‘is this going to change in the next year and blow up our model – our AI – and not enough people are thinking about that. If you build a tool that relies on an algorithm that relies on tomorrow working the same way and if something significant changes in law and if you have built your business round a model designed around yesterday’s data and then you deskill based on that? You might face some real challenges.”

BLM has launched a series of client tools of which one is Foresight, its flagship AI system developed by the London School of Economics and BLM’s data science team to help claims handlers decide who is at fault in a traffic accident.

Dunkley says: “Why Foresight works well is that it just looks at data today and simulates judgment. It’s a really powerful way to attack a whole host of ‘yes no/black white’ propositions for recovery and its scaleable – takes best practice, takes out cognitive bias and always gives the answer that’s the most appropriate.”

Foresight supports the decision-maker and Dunkley adds: “We’re not saying blindly trust the model.” It’s not surprising that he says that given that the biggest inhibitor of change is people feeling threatened.

Within BLM things have got to a stage where the data science team has created a methodology for prioritisation and sourcing, to the point that they almost think of it as a product line. “I know what the next ideas are and we have two or three projects that are in design and build and two or three that are coming on to the assembly line,” says Spencer.

A steering committee decides those things, with ideas coming through from the business piecemeal, facilitated by key partner groups focussed on certain problems.

Dunkley says: “There is a balance between people bringing you ideas and then going out and eliciting the good ideas that are buried because people can’t articulate them.”

And Spencer adds: “It’s important to balance a level playing field and business-critical complexity – that’s the steering committee.”

What about the cost of all this – data scientists don’t come cheap, how did BLM take the early step of giving the green light to hiring a team?

Spencer says: “It was a conscious strategy decision driven by Michael Brown who made it very clear that we needed to invest in this sort of disruptive technology.”

Ewen added: “It was driven by the client – we’d gone through process of launching a portal for clients and giving them trends on their data and the feedback from them was we want you to be forward-looking and help us to make decisions better.”

While there have been numerous challenges – the team had to stop trying to punch holes in the firewall, for one, – Spencer says there are bigger risks. “The experience of taking risk is less common in law firms.”

He adds: “Our lawyers are rightly talking about making a difference and in data science that’s the case too – we’re having a real-world impact.”

The rest of the team:

David Elliott

Data scientist

Studied computer science at university

Was a quantitative strategist at hedge fund Maven Securities until February 2016.

Joined BLM in February 2017 to apply his skills to the insurance law sector.

Says: “A big motivating factor is that it’s all pretty groundbreaking stuff. A lot of the challenges haven’t been tackled before or even thought about too much.”

Urvi Parekh

Data scientist

BLM was Parekh’s first job after completing her Master of Science at City, University of London.

She says: “I was looking for something challenging and new – you hear a lot of financial services looking for a data scientist and what attracted me to BLM is that it’s still rare to be a data scientist in a law firm. I feel like the problems are more personal. It’s someone’s medical history rather than financial quantum.”