The Carrying Cost of More: Why Good Decisions Accumulate Into Burden

The Problem With Things That Work

Failure is comparatively easy to recognize. A product that doesn’t sell, a system that doesn’t work, or a report that nobody reads eventually produces enough signal that someone acts on it.

The harder problem is recognizing when something is still working, still producing value, and perhaps still improving, while the next increment is no longer worth what comes with it.

Businesses face this judgment constantly, usually without naming it. Another feature may genuinely improve a product. Another metric may reveal something useful. Another layer of automation may save real time. Another AI capability may solve a problem that could not be solved a year ago. Considered individually, each decision can be entirely rational, and in most organizations each will be approved on exactly those terms.

But organizations don’t experience those decisions individually. They experience them cumulatively.

I’ve started using the term carrying cost for what accumulates: the ongoing attention, coordination, maintenance, and risk an addition imposes for as long as the organization keeps it. The term is borrowed loosely from inventory management, where carrying cost describes the expense of holding stock rather than acquiring it. The distinction matters. The cost of acquiring something is paid once and usually appears in a budget. The cost of carrying it is paid continuously, often in time and attention rather than money, and rarely appears anywhere.

The concept becomes useful the moment it is applied to a decision. Before approving anything significant, four questions bring the carrying cost into view.

An addition that can’t answer them hasn’t been fully evaluated, however strong its promised value.

Why Good Decisions Accumulate

If every addition is individually defensible, accumulation can’t be explained by poor judgment in any single case. It is explained by how organizations are built.

New initiatives have sponsors. Someone proposes them, champions them, and is credited when they launch. Most performance systems reward building, implementing, and shipping, because those things are visible and attributable. Almost no one is rewarded for retiring a report, consolidating two tools into one, or declining a feature whose burden would outweigh its value.

Carrying costs, by contrast, have no owner. They are spread across support staff, analysts, engineers, compliance teams, and end users, each absorbing a small increment that never appears as a line item. The person who approved the addition rarely feels its weight.

And ending something creates visible risk. Whoever proposes retiring a system or a report must accept responsibility if something goes wrong, while the benefit of removal, a lighter organization, is diffuse and hard to claim.

Research helps explain why the pattern persists. It consistently finds that people overlook subtractive solutions when trying to improve something, and that past investment makes established practices hard to abandon.^1,2

Together, these forces work like a ratchet. Additions are favored when options are generated, credited when results are attributed, and protected when commitments are reviewed. No single decision is the problem.

Where the Carrying Cost Shows Up

Carrying cost shows up in operations, in people’s attention, and in the coordination work that multiplies around every new system. It looks slightly different depending on what is being added.

A product feature has to be tested against every future change, supported when customers struggle with it, and accommodated in each subsequent design decision. Software systems in active use grow more complex over time unless someone deliberately works to simplify them.⁶ The burden reaches customers too. Consumers often favor feature-rich products when choosing, then report lower satisfaction after using them.¹⁰ The feature that looked like pure upside at the point of decision carries a cost that surfaces later, in a different part of the business.

Metrics carry a quieter cost. Each new measure competes for the same limited supply of attention, and every dashboard someone must review is time not spent acting on what it shows. Metrics also change behavior. Once a number becomes a target, people can begin managing the number rather than the work it was meant to reflect.

Automation may be the clearest case of value and burden arriving together. In a paper that remains remarkably current, Lisanne Bainbridge described the ironies of automation.³ Automating a process removes routine work but leaves people responsible for the exceptions the system cannot handle, which are by definition the hardest cases. Meanwhile, their skills erode because they no longer do the routine work that kept them in practice, and they are asked to monitor for rare failures, something humans do poorly. Raja Parasuraman and Victor Riley later showed that people both overrely on automation and underuse it, in ways the original business case rarely anticipates.⁸

Artificial intelligence inherits all of these burdens and adds its own. Engineers at Google, in a widely cited paper led by D. Sculley, argued that machine learning systems are especially prone to hidden technical debt: the model is often a small fraction of the production system, surrounded by far larger infrastructure for data, configuration, serving, and monitoring, and changes anywhere can alter behavior everywhere.⁹ Current risk-management guidance treats AI oversight as continuous across the life of a system rather than a one-time approval,⁷ and economic research shows that technologies like AI require substantial complementary investment in processes, skills, and structure before their benefits appear.⁴ The capability can be purchased quickly. The organizational capacity to carry it is built slowly.

When Value and Burden Rise Together

What these examples share is easy to miss. In none of them does the addition stop producing value. The feature still works. The metric still reveals something. The automation still saves time. The AI system still extends capability.

What changes is how that value compares with everything that comes with it.

The value of successive additions of the same kind typically rises quickly and then flattens, because the best opportunities tend to be pursued first. Carrying cost tends to rise steadily and then more steeply, because each new element must be coordinated with everything already in place. A team of five people has ten possible working relationships. A team of ten has forty-five. The same arithmetic applies to features that must work with other features and systems that must hand off to other systems.

The result is that net value can peak and begin to decline while total value is still rising. That’s where the Enough Point sits.

This is why the question leaders ask has to change as additions accumulate. Early on, asking whether something creates value is a reasonable proxy for whether it is worth doing, because carrying costs are still small. Later, almost everything under consideration will create some value. The question that matters becomes whether it creates enough additional value to justify everything that comes with it.

Why the Burden Stays Hidden

If carrying costs are real, why are they so rarely caught? Part of the answer lies in how organizations account for their decisions.

Budgets capture the platform purchase or the development team. They rarely capture the hours spent reconciling a new metric with an old one or explaining a confusing feature to customers. Those costs are absorbed inside existing roles, so they trigger no review. They simply make everyone somewhat busier.

Timing works against visibility as well. Benefits are evaluated when the decision is approved. Carrying costs arrive gradually and continue indefinitely. By the time they are material, the decision is old and its sponsors may have moved on.

Most important, complexity doesn’t arrive looking like complexity. It arrives as a reasonable request from a customer, a sensible improvement from a capable team, or a promising capability a competitor has already adopted. No one decides to make an organization complex. It happens one reasonable decision at a time, and because the accumulated whole is never put to a vote, it is never evaluated.

What AI Changes

Everything described so far predates artificial intelligence. What AI changes is the economics of addition.

Tasks that once required a development project, a specialized hire, or a vendor engagement can now be prototyped in days, sometimes by people with no technical background. That’s a genuine expansion of capability. It also means the cost of adding capability is falling much faster than the cost of carrying it.

Early signals suggest many organizations are underestimating the difference. Gartner predicted in 2025 that more than 40 percent of agentic AI projects will be canceled by the end of 2027, citing escalating costs, unclear business value, and inadequate risk controls.⁵ Through the carrying-cost lens, the forecast suggests a familiar pattern: projects approved for their visible promise before their supporting burden was fully understood.

AI also weakens a quiet constraint. The friction of building things used to slow the rate at which organizations took on new obligations. Friction was never a good governance mechanism, but it did limit accumulation. As it disappears, deliberate practices have to replace it, or the ratchet will turn faster.

The objective, as I argued in my last article, should not be maximum AI. It should be sufficient AI, applied where it creates meaningful net value.

Making the Burden Visible

Because these forces are structural, exhortations to simplify tend to fade. What helps is changing a few routines.

The four carrying-cost questions belong in every significant approval. Next to them sits a question most planning processes never ask: could we solve this by removing something instead? Because people rarely consider subtraction unless prompted, the prompt has to become a habit.¹

Other routines belong after approval. Ownership should extend across the life of an addition rather than ending at launch, and a review date set at approval replaces the default of keeping something with a simpler test: if we were deciding today, would we adopt this again? Reporting what was removed alongside what was added, and giving credit for both, changes what the organization learns to value.

Finally, the accumulated whole needs its own review. Individual proposals will usually pass individual review. Only a periodic look at everything the organization is carrying shows whether it is still proportionate to what it returns. A quarterly portfolio review can ask which reports, tools, approvals, and workflows still change a decision—and which are being carried mainly because no one has been authorized to stop them.

Carrying What We Choose

None of this is an argument against growth, measurement, automation, or AI. Sometimes the right answer will be substantially more of all of them.

It’s an argument for asking a second question alongside the familiar one. We’re very good at asking whether something creates value. We’re much less practiced at asking what we’ll have to carry because of it. As AI keeps expanding what organizations can do, that second question will matter more.

The difficult decision may not be recognizing when more is possible.

It may be recognizing when more is still possible, and no longer worth it.