Every decision a company postpones is a loan it takes out against itself, and nobody sends the statement. The principal is what the fix would have cost on the day it was first raised. The interest is everything that attaches to the unfixed thing while it waits. Organisations price the principal correctly and ignore the rate entirely, which is why a two-month repair turns into a one-year programme and nobody can say exactly when that happened.
Postponement is a liability that accrues, not a pause
A postponed decision does not sit still. That is the part boards get wrong, and they get it wrong consistently, because the language of postponement is the language of stillness. We are holding off. We are parking this. We will revisit in the spring. Every one of those phrases describes an object at rest.
The object is not at rest. The system around it keeps moving, and every movement makes the eventual correction bigger. New work gets built on top of the thing that should have been replaced. New integrations attach to the interface that should have been retired. New people join who learn the workaround as if it were the design. None of that is visible on the day of the decision, and all of it is billed on the day of the reckoning.
What makes this so hard to see from the top of an organisation is that the postponed item behaves exactly like a solved item for a long time. It does not produce incidents. It does not generate escalations. It does not appear in a monthly report, because monthly reports track activity and there is no activity. A decision that nobody is working on produces no evidence that anybody is not working on it.
So the executive committee moves on. And it is right to move on, in the narrow sense that on that particular Tuesday there really was something more urgent. The error is not in the individual arbitration. It is in the assumption that the arbitration can be repeated indefinitely at the same price.
The debt metaphor was always about the interest, not about the mess
Ward Cunningham introduced the debt metaphor in a 1992 experience report about a financial software product, and the point he was making has been quietly inverted ever since. He was not describing bad code. He was describing the gap between a shipped understanding of a problem and the correct one, and arguing that shipping early against an imperfect understanding is a legitimate financing decision, as long as you go back and refactor once you understand better.
The metaphor works because of the second half. Debt is a reasonable instrument. It becomes destructive when you service it without ever repaying principal, or when you take it out without knowing the rate. Cunningham’s warning was about the organisation that borrows continuously and never returns, until every hour of new work goes into carrying the interest and nothing is left for the product.
Martin Fowler later split the concept along two axes, reckless against prudent and deliberate against inadvertent, and that grid remains the most useful thing a non-technical leader can hold in their head. The interesting quadrant for a CEO is prudent and deliberate. That is the quadrant where the organisation knowingly takes on a short-term cost because the market timing justifies it, writes it down, and comes back. It is the only quadrant where postponement is a strategy rather than a habit.
The quadrant most companies actually live in is prudent and inadvertent. Every individual decision was defensible. Nobody wrote anything down. There is no register of what was deferred, no date attached to any of it, and no owner. The organisation is carrying real debt with no loan documentation, which means it cannot tell you the balance and cannot tell you the rate.
Physical infrastructure publishes the bill that organisations never see
Public asset managers have spent decades learning this arithmetic in the open, and their numbers are worth borrowing because they are audited and they are large.
The United States National Park Service assesses the condition of its assets every year and publishes what it would cost to catch up. At the end of fiscal year 2025, that deferred maintenance and repair figure stood at an estimated twenty-four billion dollars across roads, buildings, utility systems and other structures. That number is not a budget request. It is an accounting of work that was correctly identified, correctly costed, and not done, year after year, by an organisation with competent engineers and a clear inventory.
The American Society of Civil Engineers does the same exercise at national scale in its Infrastructure Report Card, and its 2025 edition puts the investment gap at three point six trillion dollars over ten years, across roads, bridges, drinking water, energy, levees and the rest. Eight categories improved in that edition. The gap is still measured in trillions.
Two things in those numbers matter for a company that has never owned a bridge.
The first is that the deferral was never a decision to accept a permanent loss. It was always a decision to accept a delay. The permanent loss arrived later, as a consequence, and it was always larger than the delay appeared to justify at the moment it was accepted.
The second is that these organisations can state the figure at all. They have an inventory, a condition assessment and an annual recalculation. Most companies have none of the three for their own technical estate. They cannot produce the equivalent number because nothing in their management system is designed to produce it. The absence of the number is not evidence that the liability is small.
The rate is set by what attaches, not by the calendar
Here is the part that boards consistently mis-model. They assume the cost of a deferred fix grows with time, in a straight line, roughly in step with inflation and salary drift. Eighteen months of delay, eighteen months of extra cost, annoying but proportionate.
That is not how it grows. It grows with attachment.
A decision left unmade becomes more expensive in proportion to the number of things that come to depend on the unmade state. If nothing attaches, the cost barely moves. A reporting tool that nobody else integrates with can sit unfixed for three years and cost roughly the same to replace at the end as at the start. That is a healthy deferral, and executives should make more of them without guilt.
If the item is central, the curve is entirely different. A billing platform that should have been replaced two years ago has since acquired four new consumers. Each of those consumers was built against the current behaviour, including the behaviour that was wrong. Replacing the platform now means replacing the platform and reworking four things that did not exist when the decision was first declined. The organisation did not defer one project. It manufactured four more.
This is compounding in the literal sense. The base grows, and the next period’s growth is calculated on the new base. And it is why the honest answer to a CEO who asks what a delay will cost is a question in return: what is going to plug into this in the next twelve months?
That question is answerable. It requires knowing the roadmap, the acquisition pipeline and the integration backlog, which are all things the executive committee already has. It does not require any technical knowledge at all. It requires somebody to put those three lists next to the list of deferred decisions, which almost nobody does, because the second list does not exist in writing.
Organisations underprice postponement because nothing ever invoices them
Every other commitment a company makes announces itself. A lease has a schedule. A loan has a statement. A supplier has an invoice with a due date and a collection process behind it. A hire has a start date and a payroll line. These commitments are visible because a counterparty exists whose job is to make them visible.
Deferred decisions have no counterparty. Nobody is on the other side of the transaction with an incentive to remind you. The system does not send a letter. It simply keeps working, less well each quarter, in a way that stays below the threshold of executive attention until it crosses it all at once.
This creates a structural bias that has nothing to do with the intelligence of the people involved. In any prioritisation exercise, items with a date outrank items without one. That is correct behaviour for a queue. The deferred technical decision has no date, no claimant and no escalation path, so it loses every week, forever, until it stops being a queue item and becomes a wall.
And when it finally arrives, it does not arrive as a request for arbitration. It arrives as a constraint. The commercial project that the board approved in January cannot ship, because it runs through the thing nobody wanted to touch in the year before last. At that point the decision has already been made, by default, by the organisation’s own past behaviour. The executive committee is not choosing. It is paying.
There is a second-order effect that is worse. The correction, when it finally happens, is undertaken in the worst possible conditions. It is now urgent, which means it is now expensive. It is now blocking revenue, which means it will be scoped by panic rather than by design. And the people who understood the original context have frequently left, which means part of the budget goes to rediscovering what was already known three years ago.
A company that had been postponing for three years
I was called in by a mid-sized services group. Not a technology company. A business with physical operations, field teams, a real estate footprint and a chief executive who had grown the company competently for eleven years without ever needing to have an opinion about a database.
The presenting problem was a delivery one. A commercial programme that mattered to the year was six months late, and each monthly review produced a new explanation that the executive committee could not evaluate. They did not want an audit. They wanted a translation.
I asked for something that had nothing to do with the programme. I asked for the minutes of every management meeting going back four years, and permission to read them all.
It took two days. What came out of it was a list of eleven separate occasions on which the same underlying decision had been raised, discussed seriously, and deferred. Not ignored. Discussed. There were paragraphs about it. There were arguments, some of them good ones.
Each deferral had a defensible reason attached. A difficult trading year. A senior departure that needed backfilling before anything structural could be attempted. An acquisition that had to be absorbed first. A year in which the chief executive had explicitly decided that stability mattered more than improvement, which, given the context of that year, was the right call.
The thing that stopped me was not the number eleven. It was that the eleven entries were textually almost identical. Same framing, same estimated effort, same named individuals, four years apart. The item had been photocopied forward across sixteen quarters without anybody ever recalculating it.
Because the rest of the business had not stood still. In those four years the group had added two subsidiaries, replaced its invoicing tool, and roughly doubled its field headcount. Every one of those three changes had been connected to the component that nobody wanted to touch. Each connection had been built quickly, against the current behaviour, by people who were told the underlying thing would be dealt with later and who had no reason to doubt it.
So the original estimate, one quarter of work, was now fiction. Not because a quarter had been optimistic in the first place. It had been about right in the first place. It was fiction because the object being estimated had changed shape underneath the estimate. The same work, done in the current environment, was closer to a full year, and it now required coordinating three business units that had not been involved when the question was first asked.
I put that in front of the chief executive as two numbers and one sentence. One quarter, four years ago. One year, now. The difference was not inflation, and it was not anybody’s incompetence. It was the cost of four years of things attaching to a decision that was never made.
His reaction is the reason I keep telling this story. He was not upset about the year. He was upset that he had never seen the number. He had approved every one of those eleven deferrals, and at no point had anyone told him that the price of the thing he was deferring was moving while he deferred it. He had believed, reasonably, that he was choosing between doing it now and doing it later at the same price. He had actually been choosing between doing it now and doing something four times larger later, and no one in the room, including the people who knew the system best, had ever expressed it that way.
The second-order damage was worse than the direct cost. The six-month delay on the commercial programme was caused by the compounded item. So was a hiring problem, because the two people who understood the component had become impossible to replace and equally impossible to promote out of it. So was a quality issue in field reporting that the operations director had been treating as a training problem for eighteen months.
Three problems, three owners, three separate action plans, one cause. None of the three owners could have seen the cause from where they sat, because each of them only saw their own symptom.
We did not fix the component in that engagement. That was not the mandate and it would not have been the right sequence. What we produced was a register: every deferred technical decision in the group, what it would cost to correct today, what it would cost in twelve months given what was scheduled to attach to it, and what it was currently blocking. Fourteen items. Nine of them were genuinely fine to keep deferring, and saying so out loud mattered as much as flagging the other five. Two of the five were repriced upward by a factor of three between the original estimate and the current one.
That register is now reviewed twice a year. It is a page and a half. The chief executive’s own description of what changed was that he had stopped making these decisions blind. He is still deferring things. He is deferring them knowing the rate.
Pricing a later takes four questions
None of these require technical knowledge. All of them require somebody with no stake in the answer to ask them.
The first question is when this was first raised. Not when it became a problem. When it first appeared in a meeting. If the answer is more than a year ago, the original estimate is no longer valid and should be treated as unknown rather than as a baseline.
The second question is what has been connected to it since. Every integration, every new team, every process built around the current behaviour. That list is the interest that has already accrued.
The third question is what is scheduled to connect to it in the next twelve months. Acquisitions, product launches, new markets, replatforming of adjacent systems. That list is the rate going forward, and it is the only forward-looking number in the exercise.
The fourth question is what this is currently blocking that nobody has attributed to it. This is the one that changes the conversation in the room, because it usually turns two or three separate problems, each with its own owner and its own plan, into one problem with one cause.
Answer those four and you have converted an invisible liability into a priced one. You may well decide to keep deferring. That is a legitimate outcome and often the correct one. The difference is that you will be deferring a known amount at a known rate, which is the definition of a financing decision rather than an omission.
Later is a position, and it has a price
The organisations that get destroyed by deferred decisions are almost never the ones that made bad calls. They are the ones that made a long series of individually reasonable calls without ever writing down what those calls were accumulating.
Nobody in those companies decided to run a leveraged position on their own technical estate. They decided, eleven times, that this quarter was not the quarter. The leverage was the by-product.
A chief executive can read a balance sheet, a covenant, a maturity schedule. The one liability that never appears in any of those documents is the set of decisions the company has chosen not to make. It is real, it compounds, and it will present itself in full, on a date it chooses rather than one you choose.
The work is not to stop postponing. It is to know the rate before you sign.
Sources
- Ward Cunningham, The WyCash Portfolio Management System, OOPSLA ‘92 Experience Report
- Martin Fowler, Technical Debt
- Martin Fowler, Technical Debt Quadrant
- National Park Service, Infrastructure by the Numbers: deferred maintenance and repairs
- American Society of Civil Engineers, 2025 Infrastructure Report Card
FAQ
Why does a postponed technical decision usually cost far more than the original estimate by the time it's finally addressed?
Because the cost doesn’t grow linearly with time — it grows with attachment. If nothing else depends on the deferred item, delaying it barely changes its cost. But if it’s central to the business, every new integration, team, or process built on top of the current (wrong) behavior has to be reworked when the fix finally happens. A quarter of work deferred for four years while three business changes attach to it can become a full year of work involving units that weren’t even part of the original conversation.
What are the four questions that let a non-technical leader price a deferred decision without any technical knowledge?
When was this first raised (if over a year ago, treat the original estimate as invalid); what has connected to it since (the interest already accrued); what is scheduled to connect to it in the next twelve months (the forward-looking rate); and what is this currently blocking that nobody has attributed to it (often the question that reveals several separate ‘problems’ actually share one uncounted cause). None require technical expertise — just someone with no stake in the answer willing to ask them.
