Data-oriented working in nine steps

1. Make the business challenges transparent

“The first step you have to take as a company to start working data-oriented is to clarify where the challenges or problems of your business lie. It is important to have a picture of the deeper background. Suppose a manager notices that employee turnover is high. Then you start to wonder: when exactly do you think the turnover is high? Which roles within the company does this apply to? The more concrete the problem is presented, the more valuable the role of data science ultimately becomes.”

2. Define the business opportunities

"Once you've identified the problem, you start looking at the business implications of that problem. Why is that high staff turnover problematic? Does it take a lot of time to refill those vacancies? That way, you make it clear what the business implications of your problem are, and how you will benefit as an organization by solving this problem."

3. Look for the relevant data

“Before you can work on a solution, you look for data to analyze. That is a big step: you are going to consider all the factors that play a role. If you are going to investigate employee turnover, you can, for example, analyze exit interviews and appraisal interviews. But anything can play a role: salary, management style, career prospects, work pressure – there are all kinds of things to think of. And if the data is not immediately available, you ask yourself: how can we get it?”

4. Find causal relationships in your data

“From this step, the actual data science will play a role, especially with large amounts of data. You have collected all kinds of data and now it is important to discover causal connections. How you do that depends on the amount of information you have collected. If it concerns 25 exit interviews, you can simply apply statistics yourself. If you have many external contextual data sources or if you are doing research at a multinational and there are 1,500, then it becomes a more complicated story.

Data science offers all kinds of techniques to analyze enormous amounts of data. For example, you have the Random Forest method, in which algorithms themselves look for connections. You do this by first training the model via a data set and then verifying it with another set. During that analysis, a data scientist can find all kinds of connections. In your research into employee turnover, you can build a kind of profile around the circumstances of employees who have left or have not left the organization.”

5. Evaluate the business value of these connections

“Suppose you discover a causal relationship in step 4 that you can use to predict whether an employee is about to quit. Then you can also take action to prevent this. You then assess the added value of the model in this step. The reliability of the model is often the first point of attention, because 100 percent reliability is a utopia.

In doing so, it is also important to keep in mind that maintaining your analysis model costs time and money. So with this, you go back to the very first steps, where you identified the problem. You make the trade-off: do the benefits of my solution outweigh the costs?"

6. Put your model into practice

“Until now you have been busy putting together a model. In the fifth step you made a trade-off: does my model offer a valuable solution to the problem? If so, then you take action at this stage. The model outgrows the laboratory phase and is applied in practice after the consent of the stakeholders. You link the analysis model to your operational systems. This will be done step by step and under supervision, depending on the impact on business operations and technology, and on experience with previous implementations. For example, a DevOps approach is also a very obvious method for data science.”

7. Let the users experience for themselves what the model does

"The model is linked to, or implemented in, the operational systems. This can lead to new decisions in the workplace, hopefully resulting in a better outcome. In the research into employee turnover, it is therefore possible that the analysis model has now received data about the situation of an employee who, according to the model, may leave the organization. The model shows the reason for this diagnosis and provides options to improve this situation.”

8. Record the new activities and measure their business impact

“It is now also important to collect data about the actual influence of the analysis model on the work floor and its effects. After all, practice is more unruly than theory and without measurement data about the effects of your solution you will not gain further insights. You will monitor how the model and the organization manifest themselves from now on.”

9. Analyze your data for further steps

"In this last step you analyze the effects of the solution you have come up with: has the problem now been solved? Perhaps you notice that the signals and recommendations from the analysis model could have an even greater effect thanks to a particular intervention and you therefore start on the next improvement step. For example from step 3. It may also be that you encounter new problems and think: we should also take a look at this (step 1). Through such a methodical approach, data-oriented work leads to new, deeper insights that make a business challenge manageable, no matter what industry you are in."