Predictive analytics have had a big year. Incremental gains and engineering improvements have furnished marketers more access to data, storage, and computer power than ever before. According to many, this was also the year that predictive analytics became a key competitive differentiator and the age of “real-time” predictive analytics finally arrived. Companies like Mintigo and Versium offered real-time solutions for lead scoring that proved the transition is possible, if still a little complex. According to research from Dresner Advisory Services, 90% of businesses place at least some importance on advanced and predictive analytics.
But for Sriram Parthasarathy, senior director of product architecture and predictive analytics at Logi Analytics, changes in the predictive analytics market go beyond that. As Parthasarathy sees it, 2019 may be the year predictive analytics finally become a third intuitive voice in every decision people make, perhaps without decision makers even realizing it.
In this Q&A, we dig into what Parthasarathy envisions as the future of predictive analytics and what he believes still needs to happen before AI-enabled applications move into the mainstream.
Q. How have you seen predictive analytics evolve over the past year?
A. Over the past year, there are two important aspects of predictive analytics that are worth noting. One, what we started to see this year is customers wanting to utilize easy-to-use solutions to gain predictive insight from their historical data. In response, analytics providers are now offering solutions that don’t require data scientists or specialists with R or Python expertise to operate the tool. Businesses no longer need specialized tools with specialized skills to extract predictive insight from historical data.
Secondly, customers want to distribute these predictive insights to their end users throughout their organization. This has evolved the use cases for embedding analytics directly into everyday applications, allowing them to use these insights to take action within the context of the application.
Q. What surprised you the most about how the industry shifted in 2018?
A. In 2018, we saw predictive analytics start to become a key competitive differentiator from the get-go. An analytical startup is now able to compete for business against large industry players by incorporating predictive analytics into a solution as opposed to pure descriptive analytics.
When looking at the industry as a whole, we are seeing a shift in the market. For example, there are 100 players offering descriptive analytical solutions. However, once a predictive component is added, the competition is reduced down to just a handful. What we are starting to see is more and more startups rolling out analytical solutions with a key focus on predictive insights.
Q. What do you see as the future for predictive analytics?
A. Predictive analytics will start to become a third intuitive voice in every decision or action someone would take without them even realizing. By partnering human operations with predictive models, people are able to silently do their jobs better by having the technology validate the best possible action or outcome in response to a thorough analysis of historical data or past events. In other words, more business decisions will start to be guided through a recommendation made by predictive analytics.
For example, an auto body shop that utilizes predictive analytics could estimate the recommended price of a vehicle repair based on the condition of the car. This would allow an appraiser to double check his or her estimated cost, reducing human error.
Another example is looking at the healthcare industry. If a hospital’s scheduling application has a predictive analytics component, when the institution schedules a patient appointment, it can automatically recommend sending a follow-up email in one week, as there is a chance this patient (based on his or her demographic) may not show up on time.
Q. What do we still need before AI-ready applications move into the mainstream?
A. Today, there is a misunderstanding on interpreting the prediction accuracy with predictive analytics. In order for these applications to move into the mainstream, customers must embrace the probabilistic answer. This will help more companies incorporate predictive insights into everyday application faster.
For example, based on historical data, a human can report how many customers churned in the last month with 100% accuracy. However, when it comes to predicting how many customers will churn in the next month, there is probability involved in the prediction. Predictive analytics cannot provide an answer that is 100% accurate. However, using historical data to produce the prediction around how many customers will churn next month based on a trained model might be 70% accurate. And that 70% will almost always be more accurate than a random guess based on gut feeling.
Through predictive analytics, the accuracy improves as the system is updated with new data and trends. End users will need to gain trust in the process before many of these AI-ready applications become mainstream.
Q. What other forms of data analysis do you see organizations turning to in 2019?
A. Most organizations currently ignore unstructured data such as text, image, and video. This is where predictive analytics can play a key role. Analyzing unstructured data and such documents can look for automated insights to help companies make more strategic decisions.
When looking back at the customer churn example, analyzing unstructured data could allow a company to search for a customer and automatically pull information from all sources such as text documents, voicemails, internal call logs, and image catalogs. This could provide a sales rep with a 360-degree overview of the customer.
Q. What does data usability really mean, and why do you believe it will be a key topic of conversation in the coming years?
A. Data usability is useful data; useful data is the data that a user actually needs to input into a predictive model to determine his or her desired outcome. Having the useful data is the key ingredient to quality predictive models.
To put it simply, garbage in is garbage out. In order to predict an outcome through the use of predictive analytics, one needs historical data that has several input variables with good instances of the positive outcomes and good instances of the negative accounts. If some of the inputs that affect the outcome are missing or corrupt, it will impact the usability of the data and the ability to create a successful predictive model.
Stephanie Miles is a senior editor at Street Fight.