First steps in creating a #DataAnalytics function

I recently answered a post on Quora: I was tasked to build the first data analysis unit in my company, what should I do?

So I’d first be asking why. What is being expected of you and the team you will be building? Which decisions will your insights be informing? What will be done differently as a result of the analysis you provide.

For example, you work for a Cable TV Provider that supplies TV, Internet and Phone to the home. You’re expected to build the analytics that will reduce customer churn, that will determine which customers are ready to upgrade their service, that will find the next home to cold-call or direct mail. You might be expected to look at how to save money on customer service – the data might reveal that there are set-up problems with certain set top boxes.

So speak to as many people in the business (perhaps outside of the division you are working in) to understand what their critical business decisions are and what is the one additional piece of information that would help them improve the quality of their decisions. When you add all of this up, you’re going to have a great brief and business justification for your team and your budget.

Then what data do you have at your disposal? Where is it coming from? What quality is it? Who owns it? What are the privacy laws governing how that data is being used. Consider not only the data you know you have, but also the data you don’t have and if you can go there, the data you don’t know you have and the data you don’t know about and don’t have.

At the Cable TV provider, you will have all the customer names and addresses, of current, past and prospective customers. For current and past you have viewing history, calling history and internet surfing history, you have bill payment history. You can cross the household data with postcode based profiles, you can combine individual email addresses with social data.

Go back to the types of insights the business wants, then look at the data you have and see what of those insights you could deliver based on the data you can get hold of. It might also be useful to see what Enterprise reporting provides and where it falls short and what is being done departmentally on the desktop and via external consultants.

Figure out who will sponsor your projects, you need a business sponsor that will do two things for you: 1) Champion getting data for you – this is more difficult than it sounds and 2) Own the change that comes from the insight.

At the Cable TV company again – so the customer support and billing data is in their respective systems, so your champion will need to go to both of those departments and get you granular data. Let say that the insight you come out with is a predictive model that can predict a customer who will churn based on the support call data and billing data. Your champion is the one that will take your churn model and implement some recovery strategy for those customers that are about to cancel their service. Your champion will make sure that this insight creates an action that can be measured and ultimately an ROI built

This is where your group gains traction and you will need to find champions who own the business change.

I have deliberately stayed away from technology – and how you recruit and manage data analysts and data scientists. There are plenty of resources out there for that and every vendor under the sun will bang at your door to give you technology and consultants.

It’s really important to get those first steps right – that way you will be able to set up your group in the right context within your organization. Don’t start selecting tools or recruiting data analysts until you really understand what you need to achieve.

Think of a production line: what is the output – and who needs it, where and when. Then what are your raw materials, how can you guarantee a regular and high quality supply of them. Then go about building the line that turns raw data into actionable insight.