Technology became abundant. Judgement did not

Somewhere in your organisation, probably this week, someone will make a technology decision in under an hour that the company will live with for five years. It will feel small. Most of the expensive ones do. I want to explain why that moment, multiplied across every team you have, has quietly become the most consequential thing a modern organisation does, and why almost nobody manages it.

Earlier this year my team assessed the cloud environment of an established business in a regulated industry. Two weeks, read-only access, structured review. The findings report contained four high-risk security items, three reliability gaps, and the conclusion that 31% of the monthly cloud bill was recoverable.

Here is the part worth sitting with. The engineering team was good. Genuinely good. The architecture had been built by capable people using excellent technology from one of the most sophisticated platforms ever made available to businesses. Nothing in that environment was broken because someone lacked skill or because the tools failed.

So what produced four high-risk findings and a third of a bill that didn't need to exist?

Most people guess skills. Some guess tooling. A few guess budget. In my experience the honest answer is none of these, and it points at something much larger than one company's cloud estate.

Every one of those findings was a decision. A decision made quickly, under pressure, without structured evidence from the environment it affected. Each one was reasonable at the time. Each one was never revisited. And each one quietly charged interest for years.

What the team had was a judgement problem, and so, I'd argue, does almost every organisation operating today.

The constraint moved

For most of computing history, technology was the scarce thing. Compute was rationed. Software took years and capital to acquire. Infrastructure decisions were forced through procurement committees precisely because the thing being procured was expensive and rare. Scarcity imposed discipline from the outside. You couldn't make many technology decisions, so the ones you made got scrutiny by default.

That world is gone. A credit card now provisions in minutes what once took a data centre programme. Any engineer can stand up infrastructure before lunch. Any department can subscribe to software without asking. AI capability that didn't exist three years ago is available to anyone with a browser.

Bring a CIO from 1995 into a modern engineering stand-up and the thing that would astonish them would arrive before the technology did: a 26-year-old provisioning, in the gap between two agenda items, more computing power than their entire former data centre held. Then show them the decision process wrapped around that act. It hasn't changed since their day. That asymmetry is the whole story of this paper.

Organisations can feel this. In Flexera's State of the Cloud 2026, 84% of organisations say managing cloud spend is a top priority, and expected cloud spending is set to rise 28% in 2026. Spend keeps growing and control keeps slipping, at the same time, in the same companies.

When technology was scarce, the constraint on outcomes was access. Now the constraint is the quality of the decisions made about it. What to adopt. When. In what order. What to leave alone.

Access was the old bottleneck. Judgement is the new one. And here is the uncomfortable asymmetry: our capacity to acquire technology has grown exponentially, while our capacity to decide well about it has barely grown at all. Human working memory, attention, and reasoning under uncertainty are the same as they were. The decisions multiplied. The deciders didn't.

Decision Debt

I want to give this problem a name, because unnamed problems don't get managed.

Decision Debt (noun)
The accumulated cost of technology decisions made without evidence. It compounds silently, and it is invisible until an event makes it visible.

The mechanism works like a loan. Every significant technology decision made on assumption rather than evidence borrows against the future. The borrowing is rational in the moment: teams are shipping, the pressure is real, and the decision has to be made today. But the interest accrues whether or not anyone is watching. An architecture chosen for speed in year one determines unit economics in year three. A permissions model improvised during a launch becomes the audit finding two years later. A workload sized by guesswork bills every month, forever, until someone measures it.

The data describes this pattern with unusual consistency.

The interest payments on Decision Debt:

• Average enterprise cloud waste (Flexera 2026): 28–35%
• Waste attributed to lack of visibility (Anodot 2025): 54%
• Organisations with a misconfiguration incident each year (DataStackHub): 60%
• Cloud security failures on the customer's side (Gartner): 99%
• Average time to detect a misconfiguration: 180 days

Read those as one story: the debt feels free right up until it doesn't.

The waste is caused by not being able to see. The security failures are caused by configuration choices, which is to say decisions, on the customer's side of the line. And the average gap between a bad decision and its discovery is half a year. The platforms underneath all of this work as designed. These figures are the interest payments on Decision Debt, and the 180-day detection window is why the debt feels free right up until it doesn't.

Picture the room where the debt comes due. A due diligence call, months from now. Someone across the table asks a precise question about cost per transaction, or recovery time, or who can access what, and the honest answer in the room is a feeling. Every organisation carrying Decision Debt has that room somewhere in its future. Only the date and the audience vary: an acquirer's diligence team, a regulator, an incident bridge at two in the morning, a finance committee watching a bill in a foreign currency climb. What makes the moment expensive is that years of undischarged decisions get repriced all at once, on someone else's schedule.

Governance debt, cost opacity, architecture drift: organisations experience these as separate problems and assign them to separate teams. They are one problem wearing different clothes. They are what accumulates when decision velocity outruns evidence velocity.

The confident decision trap

Here is where most organisations respond incorrectly, and the incorrect response is instructive because it feels so much like the right one.

Faced with mounting technology uncertainty, the instinctive move is to project more confidence. Executives are trained to sound certain. Vendors are paid to sound certain. Internal business cases arrive with single-point estimates and no stated ranges, because ranges look weak in a steering committee.

The research on this is one of the most useful counterintuitive findings I know. Larcker and Zakolyukina at Stanford analysed roughly 30,000 earnings-call transcripts and found that executives whose companies later had to restate their results used measurably more extreme positive language and fewer hesitation markers than executives whose statements held up. Unhedged confidence, at scale, is a statistical signature of the claims that fail. Meanwhile van der Bles and colleagues showed experimentally in 2019 that stating a numeric range around an estimate does not reduce an audience's trust in the source, while vague verbal hedging does.

The register most organisations reward — confident and unqualified — is the register most associated with being wrong.

Sit with what that pair of findings implies. The register that feels weak, an honest range with a stated basis, is the one that actually preserves trust. Sophisticated audiences, boards, investors, regulators, are already calibrated this way, whether or not they could cite the studies.

I call the underlying failure the confident decision trap: mistaking the feeling of certainty for the presence of evidence. The trap is comfortable because confidence is cheap to manufacture and evidence is not. A slide can assert. A dashboard can reassure. Only a structured look at the actual environment can tell you whether the assertion is true.

And this is the moment to talk about AI, because AI is about to make the trap much deeper before it makes anything better.

The AI options paradox

AI is the greatest abundance event yet. It generates plans, architectures, analyses, and recommendations in seconds, fluently and persuasively, at essentially zero cost. Which means the number of plausible options in front of every technology leader is about to grow by an order of magnitude while the evidence for choosing between them grows not at all. More possibility, same judgement. I think of this as the AI options paradox: the tool that multiplies your choices does not multiply your ability to choose.

Two findings from the human-AI research should shape how every leader handles this. First, Dell'Acqua and colleagues, in a field experiment with 758 BCG consultants, found that AI assistance sharply improved performance on tasks inside the model's capability boundary and degraded it on tasks just outside, and the boundary is invisible in advance. They called it the jagged frontier. Capability falls off a cliff at the edges, and the cliff is invisible from where you stand.

Second, a field experiment with roughly 1,000 students found that unrestricted access to AI answers improved practice performance by 48%, and then produced a 17% decline on the exam once the AI was withdrawn. The gain was real and the dependence was realer. The effect nearly disappeared when the design forced people to attempt an answer before the AI supplied one.

Both findings say the same thing about organisations. AI amplifies whatever decision capability already exists. Teams with strong judgement use it to test more options against evidence, faster. Teams without it use AI to generate confident-sounding justifications for decisions they were already going to make, which is Decision Debt with better production values. The fluency of the output tells you nothing about which is happening, and that is precisely the danger.

I am genuinely optimistic about AI; we use it deeply in our own work, and I believe human and AI collaboration is one of the defining opportunities of this decade. Optimism about the tool is exactly why I'm insistent about the operator. An organisation that adopts AI before building decision capability is handing a faster car to someone who hasn't learned where the brakes are.

Judgement is a capability, and capabilities can be built

The good news is the part I find most people haven't considered. We talk about judgement as though it were a personal talent, something a great CTO simply has. Some of it is. But organisational judgement, the ability of a company to reliably make good technology decisions, is a capability. It has components. Components can be built. Compute can be rented by the second; judgement is still built the old way, one evidenced decision at a time. In my experience, four components do most of the work.

1. Evidence before action. Significant decisions get made against structured evidence from the actual environment they affect, gathered before the commitment, not after. There is a reason the AWS Partner programme reports that a Well-Architected Review typically identifies 20 to 40% in cost reduction: an environment examined systematically almost always disagrees with the picture of it that lives in people's heads. The environment is telling you something; the question is whether anyone is structurally required to listen.
2. Sort decisions by reversibility. A reversible decision deserves speed and a light process, because the cost of being wrong is a correction. An irreversible one — a core platform commitment, a data model, a multi-year contract — deserves full decision hygiene, because the cost of being wrong is years. Most organisations apply one uniform process to both, which guarantees the reversible ones are over-governed and the irreversible ones are under-evidenced. Sorting them is a one-meeting change with a decade of payoff.
3. Speak in ranges, and separate two questions. For every material estimate, state the range and the basis. Then keep two questions distinct: how good is our evidence, and how likely is the outcome? "Well-supported by three comparable cases, with a 15 to 25% return depending on adoption" is a different claim from "likely significant returns," and the first is both more honest and, per the research, more trusted. Boards can work with a range. What they can't work with is confidence that turns out to have been decoration.
4. Make decisions compound. A decision, once made, should leave a residue the organisation can learn from: what was decided, what evidence supported it, what would prove it wrong, and when it gets revisited. Written down, briefly, at the time. This is how a company gets smarter with every decision instead of merely older. A decision with a written basis can be audited against reality; a decision that lives in someone's memory cannot.

Imagine the version of your organisation five years from now where every significant decision left that trace: the evidence, the reasoning, the tripwire that would prove it wrong, the date it gets revisited. New leaders inherit judgement instead of folklore. The company remembers why. Almost nothing on a balance sheet compounds like that.

Everything above can be built with the technology an organisation already owns, which is rather the point. What it asks for is the choice to treat decision quality as an asset you invest in deliberately, the way the last generation invested in technology access.

The company I opened with understood this better than most, and I want to be precise about what changed for them, because the most durable part sits past the numbers. The recoverable spend mattered. The security findings mattered more, since they were resolved on the company's schedule rather than an incident's. But the durable change was that their next significant decision, and the one after that, was made against a documented, evidenced picture of their own environment instead of against memory. They stopped borrowing. That is the entire game.

The stewardship question

I'll finish with the belief underneath all of this, because I think it's the real dividing line for the next decade.

The organisations that will compound through the coming abundance are the ones that treat judgement as their scarce, buildable asset, and every technology choice as a chance to strengthen it. The ones that will struggle are the ones still solving a scarcity problem that no longer exists, accumulating technology and Decision Debt in equal measure, confidently.

There is a test I'd offer any advisor, internal or external, in this new environment: are they trying to become more necessary to you, or progressively less? An advisor whose incentive is your dependence will keep the judgement and sell you its outputs. A steward builds the capability inside your organisation and expects to be needed less over time. In an abundant world, the second kind is the only kind worth keeping, and the willingness to say "don't buy that yet" is how you recognise them.

That conviction is why CloudFruition exists, and it shapes how we work in ways a services catalogue can't show. We assess before we recommend, because evidence before action applies to us first. We teach the method as we apply it, since a capability that lives only in your advisor is a dependency wearing a friendly face. And the measure we hold ourselves to is a strange one for a consultancy: being needed less each year by the organisations we serve. Judgement, once built, belongs to you.

So here is the question to take into your next leadership meeting, and it takes one minute to answer honestly. Pick your organisation's last significant technology decision. What structured evidence from your own environment informed it? If the answer is a document, you're building capability. If the answer is memory, experience, and vendor material, then you now know what your Decision Debt looks like at the moment it's being taken out.

Technology became abundant. Judgement didn't. The organisations that close that gap deliberately will make the next decade look easy from the outside, and everyone else will keep wondering why owning more technology than ever feels like controlling less of it.

John is the founder of CloudFruition, a technology decision capability partner. CloudFruition helps organisations build the capability to make better technology decisions through evidence-based assessment, advisory, engineering, and human-AI collaboration.

Sources cited: Flexera State of the Cloud 2026 · Anodot Cloud Cost Survey 2025 · DataStackHub Cloud Misconfiguration Statistics 2025–2026 · Gartner (confirmed through 2026) · AWS Partner Programme data, AllOps 2026 · Larcker & Zakolyukina, Stanford Rock Center (~30,000 earnings-call transcripts) · van der Bles et al., 2019, Royal Society Open Science · Dell'Acqua et al., 2023, Harvard Business School (N=758) · IZA Discussion Paper 18338 (N≈1,000) · Client finding drawn from an anonymised CloudFruition assessment engagement, 2026.