According to a 2018 survey by New Vantage Partners, 97% of firms are investing in big data and artificial intelligence (AI), and the primary goal for most is to deploy advanced analytics capabilities for business decision-making.
However, at least half of analytics results never make it into production.
The ultimate goal of digital transformation through data science is to improve the organization. Whether the focus is on increasing revenue, lowering cost, increasing productivity, or launching new businesses, organizations have to move beyond gathering data.
The analytics assets and AI models created in the discovery phase also are not the end goal. If you do not operationalize these assets by putting them into action to drive business outcomes and manage them continuously, data science cannot fulfill its potential.
Where to start with data science
If you plan to invest in a data science program, you should be well past the hypothesis and data collection phase of a project. You need to have a clear problem identified that you are ready to tackle with analytics, and you need to know what data you will use to solve that problem.
There are many barriers to success in data-driven initiatives. Chief among them is the difficulty many organizations face in operationalizing analytics: deploying, monitoring, and managing analytics and AI in business processes.
How can you overcome this challenge and move your idea from a science project to true data science?
- Start small, with a project that addresses a core competency of the business. Make sure all parties agree on the business value and technical feasibility of the project.
- Select a project that will offer a win within a year. Know going into the project what a win looks like and how it will be measured.
- Look for opportunities to automate and expand your use of analytics. Automation can multiply the results of your project exponentially.
To illustrate these concepts, let me describe two recent data science success stories. One is a health-care system using computer vision to help treat cancer patients. The other is a social services agency using machine learning to protect vulnerable children. Finally, you will see how these projects are moving out of the pilot phase to be deployed for long-term results.
Automating tumor detection in medical images
Amsterdam University Medical Center recently began using computer vision and predictive analytics to improve care for cancer patients. Its initial project uses object detection to identify and measure tumors in CT scans of livers.
Previously, radiologists measured the size of tumors manually in the scans before and after treatment. This work often takes up to a third of the radiologist’s workday, but it is critical work. If the patient’s tumors are responding to treatment, that also makes the patient a good candidate for surgery.
Leaders saw this as a perfect pilot project to test the capabilities of analytics and AI. The hospital built computer-vision models to analyze the medical images in a fraction of the time. Object detection recognizes tumors and tumor sizes almost instantaneously. Plus, the models are more objective and more accurate than the radiologists are.
This use of analytics not only frees up radiologists to do more hands-on work with patients but also saves lives. By finding results faster and more accurately, computer vision can help get more patients into lifesaving surgeries sooner.
How did this project move from science project to data science results?
- AI is being applied to real patient data to make decisions about patient care.
- The hospital selected a project with a lot of potential because the current method is manual, time-consuming, and somewhat subjective. Leaders clearly understood the benefits of automating with a more objective and accurate method.
- The initial project size was manageable because AI was applied to help with one aspect of treating one particular type of cancer.
- Automating the measurements of the tumors is a repeatable process that will continue to save time and improve patient care.
- The success of this project can be repeated to help treat other types of cancers and to read other types of medical images.
Alerting caseworkers when children are at risk
New Hanover County, North Carolina, is ground zero for the opioid epidemic that is ravaging the US. As a result, the Department of Social Services there has seen skyrocketing cases of abuse and neglect.
The number of children taken into custody because of opioids has doubled in the county since 2013. Opioids now account for nearly 30% of interventions by the DSS.
The agency knows the factors that put a child at risk. But when it can’t act on disconnected data at the right time, children can easily fall through the cracks.
Wanda Marino, assistant director of DSS at the time, knew the county could do better. When she heard about a way to address child abuse with predictive analytics, she applied for an endowment to help pilot the technology.
The new system brings together disparate data sources and generates rule-based alerts when a child’s risk level has increased. It might be 911 calls from the home, arrests of family members, new individuals in the home or new investigations. Regardless of the source, the visual presentation of the data makes it easy for caseworkers to see what triggered an alert, drill into the case for details, and determine what interventions may be necessary.
The results of faster interventions include reductions in child harm and increased rates of “permanency”—a permanent home for the child. Marino says at-risk children are the real beneficiaries. “This partnership has been monumental. It’s been the one thing that has helped us to move forward and prevent child abuse in a timely manner, and also to save lives of children.”
How did this project move from science project to data science results?
- The county selected a project that has the potential to save lives and make caseworkers more efficient.
- The project had a clear goal to help children in an area with overburdened resources.
- Machine learning is being applied to multiple data sources to help make decisions about children’s safety.
- Automating the alerts for caseworkers is a repeatable process that will continue to save time and improve lives of children.
- The success of this project can be repeated and expanded to help other counties and other personnel interacting within the social services system, like judges and police officers.
Moving beyond the science project
Analytics is not a science project, and it is not the domain of only statisticians and data scientists—not anymore. What we see in the examples in this article is how these principles manifest themselves:
- Enabling insights and decision-making based on data.
- Jobs made easier and more productive.
- Decisions made more reliably and faster by increasing access to analytics.
- End results that benefit patients, families, and the community.
- Analytics that moves from science projects into operations to help the organization, its employees, and the people it serves.
For both of these projects, initial results are leading to expanded use of analytics. At Amsterdam UMC, administrators tell us they hope to expand the object detection models for more types of cancer and more types of patients. In North Carolina, what works in one county can be expanded to other counties—and could lead to a statewide program.
For data science to work, you should be on a mission to remove barriers to producing and consuming analytics. How can you take a pilot project and turn it into something bigger? How can you take data you already have and turn it into positive change for your organization and your stakeholders?
Oliver Schabenberger is executive vice president, chief operating officer, and chief technology officer at SAS.