In conversation with ETCIO, Vishal Shah-VP Technology, Digit General Insurance talks about the newest renewal prediction algorithm, automated pre-inspection process and how he is striking the balance between automation and human interference in the projects.“Renewal prediction algorithm is basically a model which predicts the number of customers who are likely to renew their policy. Since it will help us focus on those customers who we know are not likely to renew. So those are the customers where we will spend our time and energy to do a more focused follow-up. When we know the customer will re-new we obviously will be in touch with him but the energy and focus will be more on the others. It will definitely improve our renewal rates,” said Shah.
“We have taken all the information about the vehicle. We have also enriched our customer information with external data. So we use a mix of internal data, external data, and behavioral information. We also look at all the customer transactional data, how did he reach us, did he make a call, what was it about,” he added.
For this project, Shah said that anything or everything that has to do with a customer was put into the system. His team did not do a manual check of what parameters to be used instead just threw in all the parameters to the system. Shah mentioned that there was a lot of cleaning and derivation but through this approach, the team let the system know about all the influencing factors and the important variables which will lead to increased conversion.
Shah’s team worked with 8-10 different models before it finalized one. And at last, they went ahead with the regression model which was on the basis of boosting which helped the company to predict which customers are likely to renew.
“It is too early to talk about the benefits but we are expecting this to increase our renewal ratios by 2-3 percent which is huge in the current scenario. For now, we have 60 percent renewal rates for motors. Even if we increase it by 2-3 percent, it will be on the base of 200 crores, it will be huge,” Shah said.
Pre-Inspection model
“The renewal in our business has to be a continuous one. If there is a break in between, a pre-inspection process has to be followed. Earlier we used to depot an engineer who used to go down by the customer’s place and inspect the vehicle. We tried to automate this and we asked the customer to take the photograph and upload it. We further enhanced it by deploying an AI model with guided navigation for the customer to take photos and videos. If the photograph was not proper, he was being asked to take it again. This helped us detect the damages on a real-time basis,” he explained.
This process and a few others for the same inspection and re-new process have helped the company cut down the inspection time to 1-2 hours which was earlier at 6-7hours. Further, Digit Insurance is trying to make this whole automatic and automated. As soon as the customer takes the photographs and uploads them, the system will, in real time, tell him if he has been granted the approval for renewal or not. This all will be in such a way that the whole process will come down to real-time from 7 hours.
Talking further about striking a balance between automation and human intervention in an algorithm, Shah said, “You need to achieve the maturity level of an algorithm to be able to do away with human intervention. So, till the time the model is not mature, there will always be an element of human intervention.”
“So, our current benchmarking is that in 60 percent of the cases, it is direct pass-through and for 40 percent of the cases, it comes to the agent. We further intend to increase the direct pass-through to up to 95 percent. Since AI is never 100 percent so there will always be a residual which will go to the human to take care of,” he concluded.
