The Value Chain of Data Science

«We’ve collected a lot of data about our customers. How can we generate value from it?» – is this the right approach? 

Data Science, Data Analytics, Big Data, Machine Learning – These topics play an increasingly important role for many companies. But how can you decide where to start? What project will deliver value to your company?  In this article we provide you with a framework for thinking about the value you can deliver with data. We will clarify this framework and explain it using a real-life example.

Examples for the use of data  

Data is used everywhere around us, for a variety of purposes diverse set of goals:

• Creating better products for customers
• Improving cross selling and conversion of customers 
• Predicting customer behavior to proactively service their needs 
• Sell data to other companies 
• Improve internal processes

Some examples of data science applications we have seen:

• Offers from telecommunication companies to existing customers to avoid churn 
• Suggestions from streaming services, which movie a customer might be interested in watching 
• Tailored products from insurance companies or banks for each customer segment 
• Identification of maintenance needs and prediction of machine or system failures

Many companies are struggling to generate value from their data. They either don’t know what to do with the data they collect, or they have ideas for generating value, but don’t have the necessary data and capabilities to implement their ideas. To know where to start, we have to think about how value can be extracted from data. For that purpose, we introduce the value chain of data science.

The value chain of data science  

To create business value from data, the following process is ideal:

The steps in this process are:

• The starting point is high quality, relevant data (preferably lots of it). 
• Learning algorithms are unleashed on this data to make predictions and thus gain predictive insights
• The insights are transformed into suitable actions using mathematical optimization or the application of business rules. 
• We then measure the outcomes of the actions, the value they generate and in addition the efficacy of the analytics solutions. The results of these measurements are used to improve the solution.  

But what data and techniques should you start working on? To answer that question, you must start at the end of the value chain. Because a good solution often means working backwards through the chain, by starting with the problem to be solved or opportunity to generate business value.

From the value to be generated you define potential actions that could be taken to solve the problem or realize the opportunity. From there you can determine what insights are needed to decide what specific actions will lead to the best result. This leads to the data required for the solution.

How many sandwiches are needed on a flight?

We’ll explain how this works with an example. An airline realizes, that on some flights too many fresh sandwiches are not sold and must be thrown away. On other flights, there are too few sandwiches available. This leads to dissatisfied passengers and stressed flight attendants. Therefore, having the right number of sandwiches can add value on several levels:

• Lower costs
• Higher sales
• Higher customer satisfaction
• Higher employee satisfaction

The desired action is to order an optimal number of sandwiches from the caterer for each flight. The optimal number depends on the flight route, the day of the week, the departure time of the flight and the number of passengers. The airline needs to know, how many sandwiches are sold on each flight and what demand cannot be met. Data on sales is, of course, available. To analyze the unmet demand, additional data is needed. Therefore, the app that flight attendants used for sales was extended to capture this information as well. Based on this data, a model was used to predict the expected demand. This is then optimized to the best ordering quantity with minimal loss of sales, minimal waste etc.

Lots of data doesn’t mean the right data 

Back to the opening question: relying on the data you’ve already collected isn’t always effective. Instead, you should start with the problem or opportunity and consider what data and insights are needed. Sometimes there is missing data that you need to collect first in order to generate the most value. Or sometimes you have all the data you would need but are missing the right talent and infrastructure to generate the insights you need.

We’ll get into the details of Data Science in further blog posts. In the meantime, if you want to develop your knowledge and skills in Data Science, GoDataDriven’s (online) trainings are a great way to do so. GoDataDriven, like SwissQ, is a part of Xebia. Co-author of this article is Herbert van Leeuwen, Analytics Translator Trainer at GoDataDriven.