Artificial intelligence is a disruptive technology that is fundamentally changing the way almost every industry operates. While AI already impacts many areas of our lives, the majority of businesses are only beginning to understand the potential of AI for their industries.
Like companies who were able to effectively harness the power of the web in the late 1990’s, companies who are able tap into the potential of AI will enjoy a significant advantage over their competitors. In a global survey of more than 3,000 executives, approximately 75 percent of participants said they believe AI holds the key to new business opportunities. Yet, despite this excitement over AI’s potential, only one in five companies currently use AI, and an astounding 60% of companies lack a strategy altogether.
Adopting AI and integrating it into business processes can be daunting, especially if your company is new to the technology. AI requires experienced talent, strategic data infrastructure and deep analytics expertise. Many executives are uncertain about the resources they’ll require to move forward with AI, and unclear about practical issues, such as budget and how to transition from legacy data infrastructure.
These uncertainties cause many leaders to get stuck on important early-stage AI decisions, or to attempt to adopt a “fast and easy” off-the-shelf solution from a large corporation, which may or may not provide adequate return on investment (ROI)—or even offer a viable solution.
Datasets and AI approaches
AI use is driven by two core elements: data and a clearly defined problem. What data are you currently collecting? How is it stored? What data is needed to get to where you need to be as a company? And then the problem: what are you trying to solve or accomplish?
It is vital that the data a company is collecting is relevant to the company’s targeted problem. AI involves specific data requirements, far beyond basic warehousing and business intelligence techniques. Data required for advanced AI solutions is often derived from multiple sources, entailing sophisticated integration and algorithms.
Many of the AI techniques used to solve today’s business problems are leveraging fixed or static datasets. Analytics on this can lead to insights into unknown patterns in data. These techniques can be used for image recognition, bioinformatics, or recommendation engines.
The application of static datasets for AI is useful, but it has its drawbacks. Frequently, companies who can leverage online learning for dynamic datasets – where new information becomes available over time - have a better opportunity to harness the competitive advantages AI can offer.
Time series data and the power of prediction
Time series forecasting is a promising area of AI that can be used to solve prediction problems that involve a time component. These are often some of the toughest, most perplexing problems in data science and AI. Many time series data problems are often left unresolved because the time component makes working with these datasets so much more difficult to solve. Companies frequently run into roadblocks in their attempts to work with time series data and create predictive models, often because they lack the in-house expertise required.
Time series techniques look at changes in the data over time, but can be far more sophisticated than just using a collection of data points looking at how one metric changes over time. Time series data flows in a continuous stream. Self-driving vehicles, machine failure detection, and patient health monitoring all involve a constant flow of data that may be collected from many sources and tracked over time, with time as a primary axis. In other words, every new observation in time becomes its own row in a dataset.
Rather than simply analyzing what happened in the past with static data, time series analysis allows for a full-spectrum, 360-degree view. Businesses can grasp past changes, actively monitor what’s happening in the present, and then predict how things may occur in the future.
A recent example of the power of time series data analysis can be seen as applied to the problem of short term hospital readmissions or re-admits. Hospitals around the United States struggle with preventing re-admits – when a patient has to return to the hospital within 30 days of being discharged for the same medical condition.
Re-admits are extremely costly for medical insurance carriers and patients. In fact, some insurance carriers charge hospitals a penalty for re-admits, and in 2015, hospitals across the United States were fined $420 million in readmission penalties. In 2017, a major university hospital group looked to data scientists to help them identify the variables causing re-admits and to predict the likelihood of a patient’s readmission risk during the initial visit.
Various data sets were taken into consideration, such as demographics or cause of original visit to create a risk score for patients. Staff assignment, performance, and work schedules were also built into the model. Additional data input variables included the total number of returning patients, total visits per year, age, ailments, triage code, facility, marital status, sex, type of accident, and various other factors. In total, over 62,000 patient visit records were included. The time series data based solution was built with full data transparency to elevate the visibility of key factors causing readmission and enable real-time decision making.
The AI model achieved 90 percent prediction accuracy for patient re-admits. Armed with this predictability and visibility into factors closely related to re-admits, the hospitals were able to make changes that resulted in a 30 percent reduction in readmission and savings of $20 million in fines for the university hospital group.
Time series analysis allows AI to look for patterns hidden within the data streams. Data is continuously being collected and analyzed, tracking changes over time and providing a unique advantage to predict future outcomes.
Fundamentals of AI success
Insights brought by AI using time series data can deliver transformative power to businesses, although the time and expertise required to build an effective AI program in-house can be daunting to many companies. That being said, there are a few guidelines businesses can follow to help them successfully bring AI into their business processes.
1. Make sure you have a clearly stated problem or target, the more specific the better. For example, “We want to reduce manufacturing equipment maintenance costs,” is not as useful as, “Manufacturing equipment failures cost the company $10m/year in down time and repairs. We want to reduce this by 50% by predicting failures and performing required maintenance before the asset breaks and has to be taken offline.”
2. Access clean data. Either you can track and store your own data streams, or you can acquire the data you need from a third party. Additionally, the data needs to be applicable to your problem statement.
3. Ensure the AI solution is right for your needs. Is a static dataset sufficient, or is time series forecasting a better approach?
4. Determine whether your AI needs require you to build an in-house team, or if you can partner with a third party company with AI expertise. Data scientists are in high demand and can be difficult to hire, and once on board, will need time and resources to be able to build an AI solution. You may be able to save time and money and achieve a better result by working with a company that already has AI models that can be adapted and customized to your needs.
AI offers a way for companies to dramatically streamline processes, make better decisions, reduce costs, or automate low-level tasks — empowering humans to make better decisions and enabling a clear competitive advantage. Forward thinking business leaders around the world are on the brink of unlocking the predictive power of AI. Those that do will have a clearer view of the past and the present, as well as a reliable tool for predicting the future, making the initial investment into AI well worth it.
