Exploration: Explore unknown transformative patterns in data
Data scientists should be encouraged to make "big data expeditions" where there is no clear objective other than to explore the data for previously undiscovered value. For example, data scientists at a Japanese maritime services provider realized that when providing their traditional services for ship classification, they were collecting a valuable store of data that had great potential in other areas. Applying the right analysis to this data meant that ship operators could reduce equipment failures and lifetime maintenance costs by 10%. This allowed the organization to quickly increase its market share by 20% when offering this value-added service to customers.
Prototyping: Challenge the status quo with radical new solutions
Human decision making is increasingly inadequate in a new digital world with an ever-expanding universe of data. Data science and especially machine learning excel in solving the kind of highly complex data-rich problems that overwhelm even the smartest person. The list of business or government challenges that data science can tackle is potentially endless.
Data science is already changing lives for the better — or even saving them
One example is a U.S.-based police department that needed an efficient automated way to pull actionable insights from a huge volume of crime data. The predictive analytics solution put in place generated crime "forecasts" that optimized deployment of police forces, reducing the murder rate by 35% and robberies by 20% year over year. The estimated ROI of these impacts was 863%. Automated analysis of various disease symptoms and medical test data is another common area where the application of data science is already changing lives for the better — or even saving them.
Refinement: Continuously improve existing processes and products
This is perhaps the most common application of data science. Most data scientists work in the production part of their business and have established models for refining processes and products according to the data their organization collects. Common examples would be marketing segmentation, retailers tweaking dynamic pricing models or banks adjusting their financial risk models.
A deeper dive by a data science team can uncover something interesting about what is really happening
One recent example is that of Zurich Insurance, which reduced the inefficiencies around handling injury claims by using an artificial intelligence (AI) solution to fully automate injury report assessments. It leveraged AI to fully automate the medical report evaluation so that human agents could focus on value-added activities such as negotiating with the counterparty. The time to assess a medical report was cut from one hour to just a few seconds, saving $5 million per year.
Firefighting: Identify the drivers of certain undesirable situations
This category is very similar to the exploration category in terms of its methods, but is applied in a different context. Sometimes organizations trigger a data science initiative in response to crises where the symptoms are obvious — for example, a rise in customer complaints or a rapid drop in profitability. In these narrow cases, the data science team has to identify only the cause, which limits the range of datasets it needs to analyze.
Sometimes basic data discovery or self-service business intelligence (BI) is enough, but often a deeper dive by a data science team can uncover something interesting about what is really happening. Common examples include online retailers investigating why customers return goods despite prices being unmatched, deliveries being on time and quality being good, or manufacturers running open investigations into quality fluctuations.