Decisions that gave something up to make the data true

Spending the disaster budget on tenants instead of a second data centre

In 2022 our continuity plan protected the wrong thing. Our infrastructure team had proposed a second data centre in western Japan so that lease, billing and accounting systems could be restored within four hours, at about ¥1.2 billion over five years. The proposal followed the group's continuity standard, which set four hours for every critical system. But when I asked what a tenant would need from us an hour after a strong earthquake, none of our systems supplied it. Our lease records held contracts managers, not safety leads, and each building's management system reported only to its own screen and our monitoring centre.

I proposed moving lease and accounting systems to cloud infrastructure with recovery in another region and a forty-eight-hour objective, for about ¥310 million, and spending ¥560 million on strong-motion sensors in our 46 largest buildings, a floor-by-floor status service for tenants and a restart order for 388 lifts agreed with their maintainers. Our general manager of finance objected that an earthquake near month end could delay rent invoicing and the close, and he was right. The group's internal audit function objected that we would breach its standard. We gave up the four-hour ledger, I wrote the exception request myself, and the group's risk management committee approved it with a condition that we test the forty-eight hours each year.

Our legal department raised a harder question: whether a provisional status could be read as an assurance. We agreed five states, none of them called safe, each stamped with the time and the person who set it. The September 2023 drill with 120 tenants produced a first status in 38 minutes. In February 2025 a weekday-morning earthquake of intensity 4 in our area stopped 61 lifts. Every building had a published status within 26 minutes, and 58 lifts were back within three hours. One storeroom with a fallen ceiling panel had been shown as inspected, no damage found, because its door was locked; we corrected it 40 minutes later.

The group's continuity standard now names tenant status as the first critical service. I would do two things differently. I would bring tenants' own safety leads into the design in the first month rather than the twelfth, and I would have treated locked rooms as unchecked from the start.

Closing 131 vendor routes into our building networks

The 2022 inventory began as a routine audit and became the most contested project of my first three years. Across 71 buildings we found 143 remote-maintenance connections into building networks, excluding the lift makers' monitoring links: mobile routers in plant rooms, telephone-line modems and vendor-managed VPN boxes. Thirty-one vendors were involved, and for 37 connections nobody in our group or among the vendors could tell us who used them or why. Any one of them could have been a path into the controls for air conditioning, access or car parks.

I proposed closing every connection that could be moved and routing all remote work through one gateway owned by Masaragi Properties and operated by the monitoring centre at Masaragi Facility Services, which would approve each session, record it and end it. The cost was ¥540 million to build and about ¥70 million a year to run. Two groups disagreed. Mizuchi Elevator, which maintains most of our lifts, insisted that its monitoring links stay outside any gateway, arguing that a delay in its alarm path could leave passengers trapped longer. Our own facility services company feared that night-time faults would wait while someone approved a session.

I accepted the lift maker's argument and declined only its request to connect those links to our building management systems for energy reporting; instead we verified that each link is physically separate. For the night-time concern, we set a fifteen-minute approval standard and pre-approved sessions for named emergency engineers. The trade-offs cost money: two vendors raised maintenance fees by a combined ¥18 million a year, and a car-park system supplier ended its contract rather than change its practice, which cost us ¥26 million to replace.

By March 2025, 131 connections had closed. Twelve remain, each with a named owner and a dated plan. Median approval time at night is six minutes. In 2024 the gateway refused a session from a vendor laptop that had been flagged as compromised by the vendor's own security team, and we learnt of the compromise from our records before the vendor told us.

I would change one thing. I let the inventory run for nine months before talking to vendors, and several learnt of the plan from our facility managers. I would brief the largest five on the first day.

Paying for handover data in a station tower's construction contract

In December 2021 we signed a ¥64 billion design-and-build contract with Tsubakida Construction for a 27-storey office and retail tower above our busiest suburban interchange station. The contract required a building information model at completion, as most do. Experience with our previous new building, an office block opened in 2016, told me what would follow: a model file nobody in maintenance could open, a stack of as-built drawings and an equipment list typed up by our own staff over eighteen months while the warranty period ran out.

I asked the head of our development department to add an asset-information requirement: a register of every maintainable item with sixteen attributes, including maker, model, serial number, maintenance cycle, warranty end and a room or zone code shared with the building management system and the tenant portal. Tsubakida priced it at ¥240 million for about 3,400 items and warned of schedule risk. The head of development disagreed with the whole idea. The cost fell in his budget, the opening date was fixed with the railway's timetable change, and in his view maintenance data was an operating problem to solve after handover. He was not wrong about the timetable.

We compromised. Finishes and fixtures with no maintenance cycle were dropped, bringing the register to 2,150 items and the price to ¥160 million, which my department's budget funded rather than the project's. In return, I accepted the register in two tranches and agreed that a shortfall would not delay opening. Acceptance was by sampling: my engineers checked ten per cent of items against installed nameplates and live control points. The first sample found 9.2 per cent of records wrong; Tsubakida corrected its process, and the second found 1.4 per cent.

The tower opened in March 2025 with its register loaded into our maintenance system six weeks earlier. In its first year, reactive work orders per thousand square metres were 27 per cent lower than in the 2016 building's first year, and the thirty-year repair plan was costed from the register rather than estimated.

What I would do differently: I did not place any duty on tenants' fit-out contractors to update the register, and by mid-2026 about six per cent of records on tenanted floors were out of date. That clause belongs in every fit-out agreement, and I have now added it.

Restating a J-REIT's energy figures before a benchmark submission

In 2019, my third year seconded to Masaragi REIT Management, I was preparing the energy data for the J-REIT's annual sustainability report and its submission to an investor benchmark. The report showed energy intensity covering every property in the portfolio of 38. Tracing each figure to its source, I found that 41 per cent of the floor area had no meter data at all. In single-tenant warehouses and two retail buildings the tenants bought their own electricity, and the figures had been estimated from floor area using published averages, in the same table as metered data and without a label.

Nothing had been concealed deliberately; the estimates were in a working file. But investors could not tell them apart from measurements, and some of the warehouses were almost certainly using far more energy than the averages suggested. I recommended restating the previous three years, reporting measured coverage separately and negotiating data clauses with the tenants concerned. The head of investor relations and the asset manager's president preferred to wait two years until meters were installed and disclose the change then. Their concern was reasonable: the benchmark grade would almost certainly fall, and the REIT was planning an equity offering.

I presented both options to the investment corporation's board, its executive director and two supervisory directors. They chose to restate. The 2019 report disclosed measured coverage of 59 per cent and labelled every estimate. In 2020 the benchmark grade fell by one level, and two investors asked hard questions during the offering roadshow.

We negotiated green-lease data clauses with 14 of the 15 tenants who bought their own supply, spending ¥86 million on meters and loggers. One tenant declined, and its building is still shown as estimated. Measured coverage reached 81 per cent by 2022, when the benchmark grade recovered, and 88 per cent in the year to March 2026. Two of the REIT's larger investors later told the asset manager that the restatement was why they trusted its later data.

I would do one thing earlier. I discovered the problem in May and took it to the board in August, after three months of internal debate. The supervisory directors were entitled to hear it in May, with the debate still open.