Buy design authority back from the integrator

In 2018 Japan's Ministry of Economy, Trade and Industry warned that companies unable to renew their ageing core systems faced a "2025 digital cliff": rising maintenance costs, retiring engineers and systems nobody could safely change. The year has passed. Many companies did renew. Many others are still running systems built in the 1990s, now on hardware whose makers are ending support. I have worked on both sides of that problem, first at an integrator and now as a client, and my view is that the cliff was never mainly about technology. It was about who held design authority over a company's systems, and most Japanese companies had given it away.

The usual way a Japanese company buys a system is a fixed-price contract for work with a large integrator. The company writes requirements, sometimes with the integrator's help; the integrator designs, builds and delivers; the company accepts the result. It is an understandable arrangement. It gives the finance department a ceiling, the board a single accountable vendor and the IT department a small headcount.

What the fixed price really buys

I spent ten years at a systems integrator, and from the inside a fixed-price contract looks different. The price is fixed, so every ambiguity in the requirements becomes a negotiation about change requests. The integrator protects its margin by settling design questions early and in its own favour, often choosing what is easiest to build and support rather than what the client's business would choose with full information. The client rarely notices, because the client has few people able to read a design document critically.

When the system goes live, the design decisions stay with the vendor. Nobody at the client can explain why a rate is calculated in a particular way, because nobody at the client decided it. Fifteen years later, when the hardware is ending support, the client asks the same integrator to replace the system, and the integrator proposes another fixed-price project. The client has no alternative, because the knowledge needed to consider one lives elsewhere.

I watched this cycle from both sides with the settlement system at Seriyama. I helped to write it as a junior engineer at Kotoshiro Systems between 1999 and 2001. When I joined Seriyama in 2016, four people understood its rate engine, and all four worked for a third-tier subcontractor. Kotoshiro's renewal proposal in 2020 was a fixed price of ¥9.8 billion with a single cutover weekend. It was a competent proposal. It would also have repeated the arrangement that had created the problem.

Counting the tiers

The second thing the fixed price hides is the subcontracting pyramid. On my largest Kotoshiro project, fourteen of sixty engineers were Kotoshiro employees. The rest came from second-tier and third-tier companies, each adding a margin and each a step further from the client. The engineers who understood the details had the shortest contracts and the least authority. When the project ended, they moved to other projects, and their knowledge went with them.

At Seriyama we now ask every vendor to disclose the tier structure beneath a statement of work: how many companies, how many people at each level, and who among them holds knowledge we will need later. Some vendors were reluctant. Two declined to bid. The disclosures were sobering. On one maintenance contract, the people who actually changed our code were three companies removed from the one we paid.

Disclosure alone does not fix this, but it changes the questions a client can ask. If the critical knowledge sits at the third tier, the client can ask for those engineers by name, contract with their employer directly or bring the knowledge in-house. A client that does not know the tiers exist can do none of these.

Insourcing the judgement, not the keyboard

The alternative I proposed, and the board approved in February 2021, was not to build everything ourselves. We have 220 technology staff, and we will always buy engineering capacity. The change was in what we keep. Requirements, architecture and data design are now done by Seriyama people. Engineering capacity is bought under quasi-mandate contracts, in which the vendor supplies named people's time and effort rather than delivering a finished product at a fixed price. Because directing another company's engineers day to day under the wrong kind of contract can amount to disguised labour dispatch, which Japanese law does not allow, our contract types and working arrangements are reviewed together by our legal team.

The phased renewal has cost ¥5.9 billion to March 2026 against an estimate of ¥7.4 billion, with the largest wave still ahead. There is no ceiling. Our CFO was right that this exposes the company to cost risk a fixed price would have moved to the vendor, and we hold a contingency reserve because of it. What we bought with that exposure was the ability to change course. Three waves have gone live, each teaching the next, and the people who designed them work for us.

Insourcing judgement has costs beyond money. Experienced architects are scarce in Nagoya, and we compete for them with automakers and their suppliers. We pay above our usual scales for a small number of roles, which created tension with other parts of the company. Some of our first in-house designs were worse than a vendor's would have been, because our people were learning. I accepted those costs as the price of rebuilding capability that had been outsourced for thirty years.

Integrators are not the villains of this account. Kotoshiro's engineers were careful and loyal, and the best of them now work with us under contracts that name them. The fault lay in an arrangement both sides found comfortable for decades: the client avoided headcount, the vendor secured a long annuity, and the knowledge drifted to wherever the margins were thinnest.

My suggestion for boards is to ask three questions before approving any major renewal. Who in the company can explain the current system's most important calculation without calling a vendor? How many subcontracting tiers sit beneath the proposed contract, and where does the critical knowledge sit among them? And when this system needs replacing in fifteen years, who will hold the design? If the answer to all three is the integrator, the company is buying the same problem again.