Leading on Data Capabilities to Deliver Growth Strategies

Leading on Data Capabilities to Deliver Growth Strategies

I recently caught up with Andrew McMurtrie Head of Data Science & Customer Analytics – Green Flag at Direct Line Group.

1. What are you currently up to?

I’m currently heading up the Data Science team with particular focus on customer analytics, for our Green Flag business at Direct Line Group.  My role is like a “T” shape, the cross of the T means I’m leading on our wider data capabilities to deliver our growth strategies for the next few years.  Then the team go deep specifically on the Customer Analytics side of things.  Which basically looks at the funnel: how and who we acquire, how and who we retain, and finally who is churning.

2. How did you get into the world of data?

I fell into it, as the best roles are.  It goes back to my days at Accenture, where I was for over six and a half years.  The first year or so, was predominantly focused on business consulting, for example strategy, process redesign and operations improvement.  At the time, there was a lack of people in the technology consulting area, specifically as a business analyst, so I decided to move into the technology consulting team.  With the ever-increasing importance of data, the use of exploitation of data to drive commercial business decision making was the key.

For example: the growth of all the areas such as data warehousing, business intelligence, I was fortunate enough to move into few client projects, related to business intelligence and data warehousing.  Interesting for me, as I was in the business analyst role, being the interlock between what the business wanted, and how the technology could support that.  That was the catalyst for me, and I’ve stayed in this area ever since.  It’s a fascinating area, in which I’ve seen a real change in the technological developments that support business intelligence, and now we are seeing a real surge in the areas of advanced analytics, data science, machine learning – which have made this area super interesting and challenging at the same time.

3. Looking back on your career and where you are now, what would you say are the biggest data challenges businesses face today?

The biggest challenges are the accessibility of data, the legacy fragmentation siloed way that data has evolved, which typically has been for specific needs in particular areas, therefore, it has evolved naturally in those silos.  Alongside, this the quality of data: how it was designed and sourced, and the sheer lack of governance.  The lumping of requirements on requirements and not thinking how to manage this in a more robust, scalable way, applying the right governance. So, what do I mean by that?

An example would be where there are a number of data variables to answer a particular question / KPI from a specific source system, where an area of the business are looking at this as a key part of their decision-making process.  In another area of the business, another team is looking at similar data from a different source system, and are reporting on a very similar KPI.  You end up with two different teams, almost looking at a similar or the same KPI, where both are producing two different numbers.  I’ve had experience of this all the way up to very senior levels, of leadership being presented with two different numbers for the same thing.

Those are the types of legacy challenges that I’m certainly dealing with.  Which leads to how do we transition from those processes and systems to something that is governed, higher quality, coherent, and all in one place, and therefore very accessible. This brings me onto other strands of thinking about timeliness, and data we know degrades very quickly, if you don’t use it in a timely fashion for those things you need an answer too.  So, when data is tied up in inaccessible legacy systems, and the time it takes to get the data out to produce an answer it’s often far too late and the business value / outcome has been lost.

4. Do you think business people are set-up or trained enough in the use data?

Hah!  No, not enough! It’s a wide spectrum, and there will be some business people, who know a few things about the technology, and it’s that classic cliché of “a little bit of knowledge can be a dangerous thing”, therefore, they aren’t necessarily making the right decisions.

5. So, on that vein of thinking you as a data leader, how do you support the business to use data more effectively?

Understanding up front what the business problems are, even before you start thinking about data, solutions and technology.  What are the key business problems, coming back to the customer use case i.e. we are losing too many customers.  Now there will be a historical more traditional way of managing that, backward looking type of management information, seeing trends and patterns.  Then putting in place an intervention or initiative to stop the churn.  We want to understand what the business questions are that we are attempting to resolve i.e. why are the customers churning, where are they going, to which competitors, and is it something to do with price or products?

If we don’t do this, the tendency has been for users to ask for data, and they will go away and work out what their problems are.  If we take the user through the journey of asking what issues they are having, we can be more focused around what we are asking the data.  Having a clear plan / strategy about what we are addressing, and not getting hung up on where the data is coming from etc.  From there, go through a series of iterations to generate the insights based on the questions, and I think this is where agile works well.

Through this approach you start to get an appreciation of data quality, is the data sufficient and complete, to answer the question and what other insights does it highlight.  This helps the business work through a series of iterations, and the more technical teams have a real understanding and purpose of what they are developing and why.  Normally, starting with a very high-level business question, and then branching off into sub-questions focuses on the priority of delivery sprints.

None of this is done in seclusion, as through my learnings and failings (the latter not being a bad thing), is where we have done this in isolation or attempted to infer the business need, and then disappear down a rabbit hole, and you are way off the mark, as you don’t have the business knowledge.  The benefit of doing this in close collaboration with the business, gives them exposure to the approach, and when you want to produce new ways of using or visualising that data, they are onboard with that journey.

In the past the reason things were probably done in isolation, was about getting access to the business people.  If you aren’t getting time from your business stakeholders it’s probably a signal that it’s not important to them, and you need to look for those people who are interested in these challenges and support them.

6. What are the top 3 challenges have you faced now or previously in implementing data science projects?

• The hype factor or the belief that hiring a smart person with programming skills, good at maths and statistics, means that they will find the golden nugget and make you a billion pounds! Perhaps, that will only work if you have good data to start with.  I’ve found that data science teams as they have exposed inadequacies in data, can’t actually find those nuggets as the data isn’t good enough.  Data science if you are going to be a purest about it, is about advanced analytics i.e. prediction.  Whereas, the term often gets used with doing the more data engineering activities i.e. stitching together different data sets, which comes back to the data quality aspect.  True data science is more about machine learning, prediction.  I would say the preparing and structuring of data is the bulk of the activity that data scientists are currently doing more of, as opposed to the writing of algorithms etc.
• If you are hiring a data scientist, they probably do want to do predictive and machine learning. However, the business problem you are dealing with doesn’t necessarily require a machine learning model, they just need to bring some data together and visualise it for insights.  I’ve seen data scientists because of their background and skills, over engineer the solution and apply machine leaning methods or models to something that should be simple and straight forward.  The problem is the time to do this and the usability factor i.e. how useful is a regression model when we are attempting to answer a simple business question.  So, what is this person going to do with as logistic regression model.  It’s not always the use case that needs a data scientist, and not using the persons time effectively across the team.
• Developing models can be quite simple, but the challenge arises when we start to look at productionising the model and automating it in a business environment. What does this mean?  I will use an example to illustrate. If we take a customer churn model, you can develop the model and how are you going to deploy it into your marketing team or CRM system, so that when you come to run a marketing campaign and how are we going to embed that.  In most cases, in legacy systems, they aren’t designed to deploy a python prediction model.  So you need to factor in the production deployment side of things, so that you can deploy it and use it.  On some occasions, we have developed models and not really thought about how it will be deployed.  In some cases, we have had to reengineer the model into a SQL database to get the same answer, but you must recode the algorithm.  You have the right attributes and you know what the parameters are to predict customers, but you have to develop it in a different program for it to be useable and translatable.  Or do you it off-line with a list of customers, so you have a manual approach.  So, you have to work out how you will deploy it upfront, and that will make it more feasible to implement.  You could have the best artificial intelligence algorithm in the world, but if you can’t use it, what’s the point.

7. As the head of data science for Direct Line Group, from a leadership position and becoming a data scientist – what single piece of advice would you give someone starting up in this field?

I’m coming at this from two angles: if you are in a leadership position, the one piece of advice I would give is to: make sure you get the senior stakeholder sponsorship for the initiative, identified use cases that have got a good backing.  I’ve seen instances, where you have a proxy to the director and it becomes a vanity project.  If you haven’t got buy-in from a senior leader / C-Suite level individual, I wouldn’t bother.

If you want to become a data scientist: be pragmatic!  For example, if you are going into less mature environments, be prepared to do some of the leg work i.e. data wrangling and demonstrate value through some quick wins, which may not be very sophisticated.  You will get buy in, and then show the path to doing more of the advanced analytics, and/or if you only want to do really complex projects, go and knock on the doors of Google or Facebook.  We are doing some machine learning, I would say there isn’t an abundance of it right now, but it is emerging.  Most organisations are still at the very early stages of development in this area. Aside from some of the more greenfield companies.

8. We hear a lot about Artificial Intelligence, Machine learning, Internet of Things do you think these are now a reality?

I think these are a reality and I can use a good example.  It all comes back to needs and the use case.  An example of IoT and AI, in my world, where we do a lot of stuff to get insights into cars and vehicles etc.  Some of the things we are looking at are: when will they break down; when accidents might occur, when you are in an insurance / rescue business and our biggest cost is to respond to a claim or a breakdown.

If we can harness vehicle in-car tech better to prevent something happening, or protecting someone better.  Are we doing this as of today, a very small amount, but growth isn’t so far off in the future.  We are using telematics data to look at how people are driving etc.  There are other examples, if you are a heavy call centre based business, the use of chat bots is more productive and better in the long run, and I’m sure there will be customers that won’t be happy talking to a robot.  But, I think this is more imminent than what this was a couple of years ago.

9. Could you share any memorable or humorous moments in your career where data has been involved?

Crikey!  Yes, there is one example that I can give.  At another customer, we were developing a prediction model based on customer churn.  We took attributes in the data where it would indicate that customers are going to churn, often these data points when you are putting them into a model, aren’t in a normal human readable code.  So, in this example, there was a code that was established as the most probable reason someone is likely to churn, and when we uncovered what that was, it was the code for someone who was deceased!

10. Finally, can you recommend any of your favourite data books to our readers?

One that has jumped into my head is called “What would Google do?” by Jeff Jarvis.  It’s more about creating platforms, and ultimately it is about data and how it’s disrupting business models.  There is another one by O’Reilly called “Data Science for Business – What You Need to Know about Data Mining and Data-Analytic Thinking” by Foster Provost and Tom Fawcett.  It’s aimed at non-data scientists who are leading data scientist teams.