How do you bring artificial intelligence into an organization that’s functioned perfectly well without it for decades? That was the challenge faced by Humana Chief Data and Analytics Officer Slawek Kierner when he joined the Fortune 42 healthcare insurance provider in December 2018.
Kierner, who had served as Microsoft’s PowerBI, Dynamics, Cloud and AI Chief Data and Analytics Officer at Microsoft before joining Humana, recounted his experience of bringing AI to Humana during the AI Summit in New York this month.
In a session at the event, Kierner presented the process of introducing AI to the organization, and Adam Wenchel, co-founder and CEO of machine learning monitoring startup Arthur, introduced Kierner and facilitated a question-and-answer session afterward with his own questions plus questions from the audience.
Wenchel pointed out how quickly AI in the enterprise has evolved. Pointing to Wall Street Journal headlines, he noted that in 2016, people were still trying to figure out what it really was. In 2017, the headlines were about how AI would take everyone’s jobs. Then headlines focused on the fact that more qualified professionals were needed to facilitate AI. The last few years headlines have focused on why AI projects fail, Wenchel said.
“In 2022, we are starting to see this come together,” he said. “We see the era of the AI-native enterprise.” But this is mainly for big companies, not mid-sized or small ones.
Large Fortune 500 enterprises are already getting AI models deployed. One of those is Humana.
What AI looks like at Humana
Humana operates traditional medical insurance and Medicare, provides medical insurance to US military members, and has also expanded to providing home healthcare, sending 50,000 nurses to attend to people in their homes and provide care, according to Humana’s Kierner. It also operates a value-based care model.
“This generates a lot of data,” Kierner said. “An important factor that attracted me to Humana from Microsoft is that there is so much data. That creates a lot of opportunities for someone interested in analytics and AI.”
But the new role was not without challenges. For instance, most of the compute was on-premises, he said, without access to some of the more modern technologies.
At the beginning of the transformation, Kierner said he started with people working on projects and areas that needed modernization.
“We knew we needed to migrate from on-premises to get all the potential the cloud could provide,” he said. That was one of the ways Humana’s journey followed the typical enterprise digital transformation journey over the last few years. The process also took lessons from successful IT project tactics of years past.
For instance, one of the top ways Kierner approached the AI transformation was to choose important transformational use cases as the initial projects so that he could demonstrate to stakeholders their value.
The organization is a hub-and-spoke model, he explained, with data governance, data analytics, and data science groups building horizontal platforms. But there were also a number of vertical units, he said, with integrators and product managers. Amid this team structure, some of the truly key team members were those professionals who understood the business process and could translate customer needs to the data scientists and engineers.
Kierner noted that it was important to measure the projects’ successes and report the value of them, but it was not a simple task. He took it in a few different stages.
First he measured how many users were moving to the new platforms. Next, the team measured how many times a member touched an AI-empowered “touch point.”
Finally, there was a point where finance and strategy teams took an interest in the projects, analyzing value in dollars. For instance, they would look at the number of hospital admissions avoided because better processes were followed.
But don’t expect to get to this point overnight, if your organization is pursuing a new AI program. Answering an audience question, Kierner said that organizations should expect results to take at least a year.
“It depends on the use case,” he said. “In our case, small use cases have to generate monetary impact in that time frame.”
Finally, for those who are still worried that AI will take their jobs, Kierner said his organization worked to create a lot of training for top executives and one to two levels down from that.
“We are still coming back with another wave,” he said. “We still find areas of friction. But we find the more we deploy AI, there is quite a lot of job creation.”
What to Read Next: