18-03-2020

How AI is helping process insurance claims at ICICI Lombard GIC

Insurance Alertss
|
18-03-2020
|

How AI is helping process insurance claims at ICICI Lombard GIC

The conventional methodology when a patient is admitted to a hospital involves a process as such: A doctor checks on you, writes out a diagnosis and a course of treatment, and post that, you are at a Third Party Administrator desk along with insurance papers for cashless facility approval.

Post the submission of documents, the entire process takes two to four hours, longer in case of queries. This is because the TPA sends it back to the insurer, the insurer looks at the policy, the doctor checks the diagnosis that is made by the physician who is treating and then also the recommended course of action, if it fits into policy, terms and conditions, etc. and then give the approval or the authorization for admitting.

AI intervention for faster claim approval

ICICI Lombard GIC wanted to change this long drawn process to ensure claims are processed faster. To begin with, the general insurance company started scanning all the submitted documents to be read automatically by the system. “The doctor’s diagnosis and the required course of action is something which is in free text. That’s where the AI comes in. It is able to understand the doctor recommendation along with the doctor’s diagnosis, and match that with the policy terms and conditions, and come out with a ‘whether this is a case which is admissible under the policy terms and conditions or not’,” Girish Nayak, Chief-Customer Service, Technology and Operations, ICICI Lombard GIC said.

Nayak while explaining the intervention of AI said, “When the doctor writes out a diagnosis, he usually writes few sentences using medical terminology, including the course of treatment. In India there is no structured code for any particular treatment.”

The AI program at ICICI Lombard is able to understand the recommendation based on all the past records, what this could mean, and what is the course of treatment which would be appropriate for this kind of admission. “And based on that, it also checks the policy terms and conditions which are in our system itself. And then it checks whether this matches with what is allowed, and based on that it’s able to arrive at whether the case is medically admissible or not,” he added. Additionally, there’s a machine learning algorithm which decides the amount that has to be approved for this authorization. “The learning algorithm is again based on past claims from that location. So it is able to throw up an amount which is then sent back to the TPA along with the approval,” said Nayak.

“Based on that, within about 90 seconds we are able to return to the TPA desk with the approval and the amount of authorization. So this speeds up the process dramatically. From a couple of hours it comes down to less than a couple of minutes. Almost about 40-45% of our corporate health claims are processed using this method,” he added.

Infrastructure changes to accommodate new age technologies

Nayak explained that most of the AI-ML models are run on the cloud, owing to data centre limitations. “Especially while training the algorithms, it requires a huge amount of computing, that you definitely need to do on the cloud,” he added.

But Nayak’s vision is practical. While the model is run on the cloud, through APIs it is integrated with the legacy systems and process end to end. For instance, for the claims authorization part, Nayak said, “The posting still happens through the legacy system, but there is an API to the cloud engine which is actually running the AI program, and it sends the output back to the legacy system for it to process the claim further.”

Nayak doesn’t consider it to be much of a challenge. “We integrate AI into legacy systems. As our volume increases, it’s helping the claims managers or the doctors actually accessing the claims, it helps them preempt the time from regular routine cases and helps them concentrate on larger cases where they can add significant value,” he said.

Source: The Economic Times