The arithmetic of legacy

Everyone agrees the incumbents should modernise. The mathematics explains why they mostly cannot — and why the ground-up build stopped being a luxury.

Ask why a century-old insurer is slow and the answers come back as adjectives: bureaucratic, risk-averse, cultural. This paper offers a different answer, and it is a number. A large insurer does not run one system; it runs a hundred or more — policy administration by era, claims by acquisition, finance by geography, each added for good reasons, each still load-bearing. The problem this creates isn’t age — it’s arithmetic.

Count the interfaces, not the systems

Connections between systems grow quadratically: n systems can require up to n(n−1)/2 interfaces. Six systems make fifteen. Thirty make four hundred and thirty-five. One hundred make four thousand nine hundred and fifty. The estate does not get linearly harder to change as it grows; it gets quadratically harder, because every new system must be reconciled with the ones already there.

An interface is more than a cost: it is a place where two copies of the same fact can disagree — a premium in the quoting tool that differs from the premium in the ledger, a risk code that means one thing upstream and another downstream. Reconciliation, the industry’s great hidden workload, is the tax paid on those disagreements. A single-record system does not reduce the tax — it removes the base on which the tax is levied.

Where the cost actually lands

The third paper in this series priced the monthly close; the tenth showed that five of an agency’s six speeds are set at bind. Both are downstream symptoms of the same structure. Queueing theory has a compact way of putting it: the time work spends in a process is set by the queues it waits in, queues form at hand-offs, and hand-offs form at system boundaries. The number of systems is, near enough, the cycle time. That is why the same bound risk can take minutes to underwrite and weeks to reach a clean bordereau row — the judgement was fast; the journey between systems was not.

Why the incumbent cannot simply rebuild

None of this is news inside a large insurer, and the obvious response — replace the estate — has been tried across the industry for two decades. The record is documented. A joint McKinsey and University of Oxford study of more than five thousand large projects found that big IT programmes run, on average, forty-five per cent over budget while delivering fifty-six per cent less value than promised, and that one in six goes so badly it threatens the existence of the company itself. These are not insurance figures alone, but insurance transformations carry two extra burdens of their own.

— The estate mirrors the organisation. Systems accrete along the lines of departments, eras and acquisitions — an observation as old as Conway’s law. Untangling the systems means untangling the history that produced them, which is why so many programmes become political before they become technical.

— The rebuild happens in flight. The old systems cannot be switched off while the new one is built. Live books must renew, claims must pay and regulators must be answered throughout, so the firm pays for both estates at once — and every migration step risks the book it is migrating.

Faced with that arithmetic, bolting intelligence onto the existing estate is a rational move. It is also a bounded one. Michael Hammer said it in the Harvard Business Review in 1990: don’t automate, obliterate — automating a fragmented process preserves the fragmentation and makes it faster. Thirty-five years later the tools have finally caught up with the advice, but only for those free to take it.

The honest objection

There is a serious argument against ground-up builds, and it deserves stating at full strength. Gall’s law holds that a complex system designed from scratch never works: working complex systems are invariably found to have evolved from simple systems that worked. Anyone who has watched a grand replatforming fail has seen Gall’s law collect its debt.

But read carefully, the law does not indict the greenfield build — it indicts the big-bang transformation. A firm that starts with one product on one lifecycle, proves it end to end, and grows the system by configuration is evolving a complex system from a simple working one, which is precisely what the law prescribes. It is the incumbent’s programme — replacing a hundred systems with a new estate in a single campaign — that attempts what Gall says cannot be done. And a second honesty matters here: each of those hundred systems exists because it worked. The constraint on the incumbent is structural, not personal. The people are able; the arithmetic is against them.

What the ground-up build buys

Freedom from that arithmetic is the biggest prize on offer. A system designed as one record can be designed backwards: the outputs — the bordereau, the tax return, the claims file, the audit ledger — define a data contract, and the contract is enforced at the moment of quote. Every field the end of the month needs is captured at source, once, because the process was shaped around the data rather than the data extracted from the process. Nothing downstream waits for a field that was never collected, and nothing needs reconciling with a copy of itself. These are design claims about how delegated-authority infrastructure should be built — which is rather the point, since almost no existing estate is free to build this way.

The conclusion

The industry’s technology debate is usually framed as new versus old, or AI versus tradition. The truer frame is quadratic versus flat. An estate of a hundred systems pays an integration cost that grows with the square of its own history, and its bravest transformations are bets against documented odds. A firm built on one record largely avoids that cost. This series has argued that what made incumbents big is now rented by the policy; this paper adds the corollary for the machinery itself. The advantage is no longer owning the most systems — it is needing only one.

Cuttleflow Systems · Perspective 14 · 33°53′S · 151°16′E · Sydney