How AI predicts time zone with highest contact probability to optimize telephone calls to demand payment of National Health insurance premiums
June 11, 2019
Japan has insurance programs that are aimed at reducing the burden of medical costs for sickness or injury, such as the National Health Insurance program for private business owners and unemployed people and the Advanced Elderly Medical Service System for elderly people 75 years or older. Aiming to lower delinquency in payment of insurance premiums, Kawasaki City in Kanagawa Prefecture has started using NEC’s white-box Artificial Intelligence (AI) in its payment-demand telephone operations in November 2018. How does AI contribute to increasing the health insurance payment rates?
38,000 payment-demand telephone calls each month
Kawasaki City, which lies at the eastern end of Kanagawa Prefecture between Tokyo and Yokohama, developed as a commuter town for Tokyo. It has a population of 1.51 million (as of February 1, 2019), which makes it the 7th largest city next to Kobe in terms of population despite its small area. Its coastal side is home to many large factories that are part of the Keihin Industrial Region, and although environmental pollution had been a major social problem before; today, night viewing of the beautifully lighted factories is becoming a tourist attraction in its own right.
”Although it’s a common trend among urban areas, due to the high number of working households, the income level of enrollees is high, and the city has one of the highest insurance premium collection rates. We used to have a low collection rate, but the various measures we have implemented to improve collection have paid off,” explained Section Chief Ueno Masaru, Collection Management Section, Health Insurance Department, Health and Social Welfare Bureau, Kawasaki City.
One of the factors that led to the increase in collection rate is the Call Center. The Call Center was launched in June 2015 as a central help desk for handling inquiries from residents about the National Health Insurance or the Advanced Elderly Medical Service System. The service was subcontracted to a private agency, which also handles making telephone calls to remind residents with delinquent payments and visiting them to collect payments.
The center makes around 38,000 telephone calls to demand insurance payments every month, but the very low successful contact rate had been a major challenge. To address this issue, the city decided to implement a full-scale AI-based system in November 2018.
Predicting time zone with high contact probability
”Hello. Is this the 〇〇’s? I’m calling about your payment for the National Health Insurance.”
At the Call Center located in Nakahara Ward, Kawasaki City, several operators wearing headsets and sitting in front of computer terminals speak with residents to remind them about their health insurance payments. Calls are usually made using an automated phone system, but operators directly speak with elderly people and other residents who need detailed explanations.
Majority of delinquent payments are due to “honest mistakes” such as failure to maintain enough balance for automatic bank transfers and missing the deadline for payment. Primarily, the goal is to remind residents who have forgotten to pay the premiums. What was done, however, was to call people sequentially based on a printed list, without particularly thinking about the order for making the calls. Thus, in many cases, nobody answered the call, and the process was very inefficient.
In 2017, Kawasaki City looked for a new subcontractor for its Call Center operations. And the operations were awarded to NEC, who proposed the deployment of AI to improve successful contact rate.
So, how does the use of AI improve the probability of contact by telephone?
Yoshinori Tetsuno, Head of Public Solutions Sales Division at the NEC Kanagawa Branch gave the following explanation about the project awarded to NEC.
”Households were classified based on the number, age, gender, etc. of family members, and previous data collected by the city on transactions and negotiations with residents related to insurance payments were inputted into the AI system. These data include information about the time zones wherein somebody answered the calls made in the past. AI then creates a prediction model based on these data to determine the time zones with the highest probability of contact with the delinquent residents.”
For the project, they used the white-box AI developed by NEC, which automatically identifies a number of regularities from diverse kinds of big data. Usually, it is difficult to precisely analyze data with variable conditions, but NEC’s AI automatically selects regularities based on data patterns and creates the best prediction model according to the situation, based on those regularities. Moreover, NEC’s white-box AI also presents the corroborative evidence for the model, enabling users to proceed with confidence even when humanly unexpected predictions come out. It is also very accurate.
After establishing a predication model using AI, it was found that one-person households, family households, and elderly households have different time zones with high contact probability in accordance with their unique characteristics.
”For example, households with children in elementary school were predicted to have high probability for having the mother at home from 3 p.m. onwards. Applying this prediction result, we make the telephone calls around evening time for such households,” explained Tomita Yoshinori, System Subsection Chief, Collection Management Section/Health Insurance and Pension Division (concurrent posts), Health Insurance Department, Health and Social Welfare Bureau, Kawasaki City.
Likewise, advanced elderly households were found to be more accessible by telephone early in the morning or between 2 p.m. to 3 p.m. in the afternoon. This is because many of them go to the hospital for regular treatment, so there is higher contact probability at time zones before and after their visit to the hospital. Meanwhile, households with single residents in their 20’s are usually difficult to contact by phone during the morning. Even for unemployed people, they usually do not answer the phone because they sleep through the morning, and were found to be more likely to answer the calls at later times of the day.
”Generally it was assumed that there is high probability for calls to get through during weekends, holidays, and at nighttime, but the AI prediction model and its supporting evidence showed that calls can get through even during daytime weekday hours for some households,” according to Call Center Deputy Supervisor Yuuta Kuroda. Calls to demand payment are made again on a different day when they do not get through at first. Up to a maximum of four calls can be made for a certain period based on rules. Before the introduction of AI, it was common to make four calls for a recipient. After its introduction however, cases when calls get through the first time have increased, and operators have reported higher instances of being able to thoroughly talk with recipients.
According to Kawasaki City, the rate of successful contact with delinquent residents by telephone has increased by 5.98% (as of February 2019), and the city will continue to further improve this rate based on the results for a fixed period of time. ”Learning of a higher volume of data further improves the accuracy of prediction by AI. Going forward, we will aim to further improve the rate of successful contact by leveraging more useful data and revising the prediction model, in collaboration with Kawasaki City,” said NEC Sales Division Head Yoshinori Tetsuno.
Direct effect on work optimization
A higher successful contact rate means that more target recipients can be reached by telephone, and that more specific support can be provided to them. It will also enable more efficient allocation of personnel and reduction in costs. And, although not directly, it can also lead to improvement in insurance premium collection rate.
One of the goals of making calls to demand payment is to filter out those who have simply forgotten to pay, which directly leads to payment if addressed.
There are three major types of persons with delinquent insurance payments: (1) those who simply forgot to pay, (2) those who are unable to pay for economic reasons, and (3) those who intentionally do not want to pay.
For those who forgot (1), they will pay right away once you can get through them by phone. In the case of (2) and (3), however, detailed consultation and specific explanations are needed before you can convince them to pay. For those who are unable to pay (2), there are relief provisions, such as reduction and exemptions for insurance premiums. And for those who are not willing to pay (3), rigid measures, such as seizure of property, can be imposed. Having a Call Center makes it possible to allocate collection staff specifically to handle consultations and provide explanations. In this respect, the Call Center has a significant role in improving collection rates.
Going forward, Kawasaki City also plans to adopt AI in optimizing home-collection operations.
To improve insurance premium collection rate, the city also conducts home-collection operations by sending commissioned agents to directly visit homes of delinquent residents to demand payment. Since there are many cases of absence from home during such visits, the city hopes to also improve efficiency of home collection by using AI to predict the time zones that residents are at home.
Although Kawasaki City has one of the highest insurance premium collection rates among the designated cities in Japan, its combined uncollected income from the National Health Insurance and the Advanced Elderly Medical Service System still amounted to more than 4 billion yen in FY2017. This indicates that improving collection rates is a common issue that needs to be addressed by municipalities throughout Japan administering the National Health Insurance and the Advanced Elderly Medical Service System. Therefore, the initiatives of Kawasaki City to employ AI may serve as a useful reference for other municipalities in Japan.
(SANKEI DIGITAL SankeiBiz Editorial Desk)