Chief Information Officer / Japan
Four decisions measured by who was left waiting
- Net premiums written, year to March 2026
- ¥286 bn
- Motor 44 per cent; fire and allied lines 31 per cent
- To pay 80 per cent of claims after the August 2024 typhoon
- 23 days
- Against 74 days after the September 2018 typhoon
- Policyholders with a registered family contact
- 118,000
- March 2026; each registration confirmed with the policyholder
Decisions & consequences
01Rebuilding claims capacity for the fortnight after a typhoon
When I joined Kotonami in October 2020 I read its review of the September 2018 typhoon in Kansai. Most of its pages concerned a reporting website that had failed twice on the first evening. A table near the back told the more important story. Kotonami received about 61,000 claims in three weeks against a normal month of 9,000, and took 74 days to pay 80 per cent of them. Claims waited for adjusters assigned from a spreadsheet, small roof claims queued behind collapsed warehouses, and staff lent by other departments could not log in for nine days.
The board's instinct, and that of the head of claims, was to buy capacity: a larger mainframe configuration and an outsourced overflow call centre, about ¥1.9 billion together. I proposed something cheaper and less comfortable. We would protect the core system by taking first reports through a separate short form, let claims settle small losses from photographs, and keep about 900 trained people ready with pre-built access. Our head of information security objected strongly to the accounts. Dormant accounts are an attacker's favourite route, and 900 of them looked to him like 900 open doors. The head of claims worried that photograph settlement would be exploited by roofing contractors who encourage inflated claims after storms.
We compromised on both. The accounts exist but cannot be used until two named managers activate them after a disaster headquarters is declared, each holder carries a hardware token, and access lapses automatically after 45 days. Photograph settlement applies only below a limit the head of claims sets, and we sample settled claims after every event. We deferred the mainframe upgrade and gave up the overflow contract. The programme cost ¥640 million over three years, rehearsals included.
In August 2024 a slow typhoon crossed Kyushu and Shikoku. We received about 38,400 claims in three weeks, the accounts were active within six hours, and 80 per cent of claims were paid in 23 days. The claims department's own changes contributed as much as ours, and the storm was less severe for us than 2018. Our sampling reopened about 2 per cent of photograph-settled claims.
I would include agencies in the rehearsals from the start. Agencies in southern Kyushu had not seen the short form before the storm, and for two days they sent reports by fax.
Switching off policyholder exports to agencies
In 2023 our agency portal still offered a button that exported an agency's customer list, with renewal dates, addresses, telephone numbers and vehicle details, to a spreadsheet. Agencies used it about 41,000 times a month, because our own renewal screens showed one policy at a time. The files ended up on dealers' shared drives, in desk drawers and on staff's personal phones. When the sector's handling of customer information at shared agencies came under the regulator's scrutiny in 2024, we already knew our exposure was large.
I recommended removing the export entirely. Our executive officer for agency sales disagreed, and so did two large dealer groups, which ran their own customer systems and said they would move business to insurers that still allowed downloads. The sales team also feared that small agencies, which hold our oldest customers, would struggle with any replacement. I accepted that a ban without a better tool would fail within weeks, so the decision became a programme.
We rebuilt the renewal worklist inside the portal so that agencies could sort, assign and annotate renewals without taking data away, and changed it three times in the first year because the first version was slower than paper. We issued about 6,000 managed tablets to larger agencies at our own cost, with no local storage and remote wiping, and gave small agencies a smartphone app that keeps documents in a protected container. Our agency-support staff visited about 2,800 small agencies in person. The programme cost about ¥2.3 billion over two years.
The export was switched off in March 2025. One dealer group moved about ¥1.1 billion of annual premium to another insurer. Agency-related data incidents in our register fell from 63 in 2023 to 11 in 2025, though some of the fall reflects changes in how we classify them. Renewal rates moved by less than half a percentage point, and the average renewal visit is now slightly shorter than it was with the spreadsheet.
I would do one thing differently. We now let large agencies connect their own systems to ours under a contract that defines what they may hold and how we audit it. Had we offered that on the first day rather than a year later, we would probably have kept the dealer group's business.
Stopping a policy-system rewrite after ¥6.8 billion
In 2018 Kotonami's board approved a ¥19 billion programme to rewrite its motor and household policy systems from scratch and replace the mainframe in a single cut-over in 2023. A large system integrator delivered it on a contract priced by the person-month. When I arrived in October 2020, ¥6.8 billion had been spent and the programme reported amber. I asked for evidence rather than status. Only 31 per cent of the requirements had been specified to a testable standard, two of six design phases had slipped by more than a year, and the integrator's own estimate to complete had risen to ¥26 billion.
I recommended stopping. The integrator argued, reasonably, that most large programmes look worst at this stage. Our executive officer for products needed the rewrite for a new household product, and the president was uneasy about reversing a programme the board had approved. The trade-off was plain: a ¥4.1 billion write-off of work we could not reuse, a two-year delay to the new product, and five more years of mainframe running costs at about ¥1.4 billion a year.
I proposed instead to migrate one product line at a time onto a packaged policy-administration platform, configured rather than rewritten, starting with small-business property, the simplest book, and ending with motor. About ¥2.7 billion of the earlier work, chiefly data cleansing and product definitions, carried over. The Audit and Supervisory Committee asked for an independent review by a second firm before the board decided, which confirmed the integrator's revised estimate. The board approved the halt in March 2021 and recognised the write-off in that year's accounts.
Small-business property moved in 2023 and household fire in 2025, with about 1.6 million policies converted and 38 per cent of the mainframe's batch programs switched off. The replacement programme is budgeted at ¥14.2 billion, of which ¥9.6 billion had been spent by March 2026. Motor, the largest book, is due in 2027. Every phase now reports specified requirements, test results and programs retired, never a single colour.
I would change two things. I ended the integrator's involvement too completely; keeping its best designers on a separate contract would have saved us months of rediscovery. And I should have told our largest agencies about the delay in person, before they read about the write-off.
Letting older policyholders choose a relative who can act
In 2021 we reviewed six months of contact-centre calls from or about household policyholders over seventy-five. About a third were made not by the policyholder but by a son, a daughter, a niece or a neighbour, and many of those callers could not be helped because they had no authority to discuss the policy. Our staff were right to refuse them, yet a daughter in Tokyo reporting storm damage to her father's house in Ehime was asked to have him call while he lay in hospital. Complaints of that kind ran at about 1,900 a year.
The proposal from our digital team was to simplify the app for older customers, and adoption of the app was one of the company's published measures. I argued that many of these policyholders would never use an app, and that what they needed was a way to nominate someone who could. We designed a family registration service: a policyholder may nominate up to two relatives, who get their own login to view the chosen policies, report a claim and follow it. They cannot change cover, cancel a policy or change the account into which claims are paid.
The argument was about consent. Our head of digital wanted registration completed online in minutes, because every extra step loses sign-ups. Compliance and claims staff, who had seen financial abuse of older people, feared relatives registering against a policyholder's wishes. I decided that every registration would take effect only after a call to the policyholder or a confirmation by their agency in person. Registration takes about two weeks rather than minutes, and we gave up the adoption figures an instant process would have produced.
By March 2026 about 118,000 policyholders had registered a family contact. After the August 2024 typhoon, about 22 per cent of claims from policyholders over seventy-five were reported by a registered relative, and they were paid four days faster on average because photographs arrived with the first report. Complaints from refused relatives fell to about 420 a year. In two cases the call-back found that the policyholder had not agreed to the registration, and one of those relatives had tried to change a payment account.
I would launch with read-only access first and let policyholders widen it later, so that older customers could try the service without deciding everything at once.