One of the fundamental of Data Science is the team. Whether you prefer your teams more A- Team or more regional chess team the experience of being propelled and enhanced by your teammates is something anyone who has been in a successful team cannot deny is a benefit. So how do successful teams operate, form and how to be in one or form one. We all know successful teams whether the one were in or ones we are fans of.
Forming that team is a skills not often recognised. How to form and who to include are sometimes the most difficult to do. There are memorable strategies one of which is entitled forming, storming, norming, performing and adjourning. The forming is similar to a canopy reception you do forming through collecting your name badge and meeting a few attendees around a table. This is before the storming begins. Then the canopies arrive a few make polite indication they are listing while reaching out to grab the parsing canopy other chase the canopies around the room. Other rush over to the canopies. Other quietly wait until halfway until the caviar is there. This is the storming where egos occur, teams iron out their conflicts and teams start to gel. The norming is the conversation that the egg pastry was the one canopy to get. It is similar to being not being able to get on the metro and complaining to the others that the metro is decrepit. It is the gelling of team members over something. This bonds teams. The next is the performing when you someone else joins your table and you instantly perform in team to ask questions. The last is the adjourning when the networking session ends.
So we know doing data science needs a team and a team with a variety of skills. So how do you setup a strategy to form this team or be in one. There are some unifying things that you can isolate in most teams. These are, suggested by an MIT report, A shared understanding of the mission and commitment to goals, clearly defined responsibilities and roles, agreed ground rules, a decision making model, effective communication, mutual and self evaluation.
When you have all of this and how do form Data Science a team to come up with insights that propel your team to legend status.
To give you an example we are joined by a strategist and leader of a new data Science within the NHS. Ken Nicholson.
Where you do you find yourself in the NHS or what is your role and department
My title is Principle Information Analyst within the information services division of the NHS known by NHS NSS.
It is reporting information and insight to NHS boards.
What does your organisation do for the NHS and what does it provide
We provide management information in terms of Health Analytics and strategic management insights.
Mainly statistics to determine strategy anything that allows the NHS to more optimally manage its operations.
We provide information to NHS health board and NHS staff. We started with statistics then on to logistic regression and decision trees 10 years ago to Visualisation through Tableau to data Science.
Is there any difference in the local and larger organisation information
Yes although it is basically the same structure with the main difference being communication. How you influence and present differs in how much gets recognised or understood. Local areas get sensitive information and larger organisations get more strategic insights. It's important to remember that it's a public service what that means is there are multiple customers and they all want something different. Still one of the most influential articles that influences us is "Management misinformation systems" by R L. Ackoff of giving the information and insight to those that understand it.
It is about managing expectation. Managing how customers expect the insights be deliver so they are relevant to them and in turn can managing expectations of the customers of the NHS. It is something we are expert in and this doesn't change. We are just changing how we get to the insight using more and more advanced data Science and understanding of the problems.
What is your experience with data Science
It's through Tableau that we have been able transform what we do using descriptive statistics and techniques of which logistic regression is something we have been doing for multiple years.
We have had sophisticated models for decades one is the Scottish Patient at risk readmission and admission SPARRA which predicts for 4.2 million people their risk of admission to AE. These are general statistics that require large accuracy. It is important to remember the size which is often in the millions and significance which often is an individual health risk. It is this size which is difficult to get away from because every user of the NHS is so different and has a different health risk. For a daily admission of 300 to AE in one Hospital there maybe 100 different health reasons were every individual is different. The direction is understanding these experiences more specific to the admission. We understand that not every value of a statistic is similar or have similar consequences so we should better represent this in our insights. We see a huge opportunity in the increase speed testing data Science model, opening or increasing available datasets and collision of data Science skills in a team.
My own experience has been through online courses to an intensive course led by the Datalab and the Data incubator in the US. I have then taken multiple online courses. I have led on Tableau and it's role out with our division.
We bring an understanding of statistic with how to present these having real time descriptive or diagnostic presentations and having predictive analytics to manage the NHS. We want the ability to align these more to the heterogeneous experiences in the NHS through data Science teams.
My role is an overseer to understand the benefit and limitation of these data Science team projects to give practical actions.
How do you see it benefiting what you deliver
For our SPARRA product we only have three age division old, middle and chaotic young. With data Science project teams we can test what happens when we use demographics, other datasets or more inference. This gives us the ability do more than predict but ask why these predictions are given.
We have had the opportunity to test this through hackathons which we recently had an Open data hackathon. This has proved that it's possible to manage. We have began to implement it by being involved in the CivTech project from the Scottish Government to help change our approach.
We are looking to prove data Science or it's advancements are do able in the NHS and become a data savvy organisation. It is about changing an approach and we maybe have the skills internally or training can be done. Although it's a change of what we deliver and an ability to try new approaches. This is why we are forming specific data Science team projects and considering external recruiting. It is for this reason we have become more agile and with a fail fast then learn approach to manage it.
We are essentially expecting 20 failures to get one success to give more understanding of what's happening in the organisation.
How are you going to form the data Science team
It is most about having the roles within the NHS and having the right amount for a project. The project, the understanding the problem and what's needed form the team. For this we foresee multiple data Science teams for individual projects. These are on demand teams for specific projects.
It important to remember that it is all data Science, all roles are data scientists and that there are many different roles. We have the approach that we don't want a lot do data scientist we want experts in just one role. So a team member is just an expert in data wrangling and another is expert in machine learning. These specific roles are although not limited by data metrics, data pipelining, data querying, hypothesis generators, data science modellers and machine learning modellers, informing and influencing experts, legal experts and data journalists.
Forming the team
Myself and my colleague Andy Gasiorowski, similar to the Company Valve, suggest a flat hierarchy where there are projects suggested and a selection of experts in individual expert roles. The teams form around projects in teams of size no more than 7 team members.
To form these teams relies on all team member agreeing on the opportunity and the importance of the problem. When you have the chance to maximise the number of operations done per month it gives teams an incentive. I agree with the forming, norming, storming, performing adjourning and have strategies for these. The performing is a trial and when there is one successful project then we can try to role this out on the whole organisation.
How is this a different approach
Before we didn't have dedicated teams for data Science, dataset had to be requested and by they time they arrived the team had adjourned. There was not the approach or direction to try alternatives. It is this change of approach to specify the project, let those data scientists specify these projects and form team with expert roles to do this. With management through KPI's, stakeholder involvements and management of direction against practical actions to improve our products.
How it's different to other DS Teams
When doing Data Science in a factory the range of data Science is small although our range is large because it has so many heterogeneous experiences and variables that changes these. It is mostly about trying many alternative directions. We going towards modelling individual experiences and how we can understand these. To do this we have defined specific projects with roles that are needed. The main thing that differs is the size of the dataset with the significance of the insights which mean you can make a difference and change people lives.
To read a longer script of the interview visit by blog.
Some of this is about using the same statistical analysis but just specifying it to individual heterogeneous experiences so the stream of data that is available mean something to those that use the insights from it. Some of this is about just trying models to get additional insights. It is mainly advancing descriptive, diagnostic and predictive analytics with a potential to do prescriptive analytics of modelling the consequences of change much further done the line. It is apparent that the size of datasets and significance of the insight halts data Science in the NHS and some of the solution is to have dedicated expert roles and well defined projects.
This main insight from this of the importance of defining experts role for a specific project highlights a failing of the data Science recruitment. That a role of data scientist or data engineer is too broad and that more specific roles are the ones that bring the most benefit. To help know which role are which there is an kdnuggets article and from the S2DS cohort I was in there was a project to identify roles from specifications. Once these roles are more clearly defined much more effective teams can be formed.
One of the other insights that was profound was that there was no mention of smart optimisation. It is not a concern to monitor every instance and instantly optimise to it. It is more of a concern to know more about the data they have. The NHS is run by people, experts, for people and they need to know what the data means. This is why this ability to try a variety of projects is a large opportunity to get insight that describe and predict more accurately. Once this is done it could become smart. The large difference in this data Science team is the impact and the potential is profound.
To find out more go to the website http://www.isdscotland.org/
By Gordon Rates - founder of AirNode - firstname.lastname@example.org - @air_node - alumni of S2DS virtual
MIT recommendation to forming a team