Last Christmas, the ghost was a credit card bill
Last December I wrote a Christmas story about a ghost that turned out to be a credit card bill. It was about how the enthusiasm of holiday spending can hide a rather ordinary financial reality. In January, the bill arrives, whether we are ready or not.
Now Halloween is approaching, and apparently the ghosts have moved from our homes into our offices. This one is not interested in rattling chains or making the lights flicker. It arrives as a beautifully prepared presentation, complete with impressive numbers, a reassuring executive summary and a recommendation that sounds entirely sensible.
There is just one small problem. Nobody seems to know who actually thought it through.
A perfectly sensible decision that nobody really examined
Imagine a fictional manufacturing company reviewing a proposal to reduce inventory by 20%. The business case looks attractive: lower storage costs, less capital tied up in stock and a welcome improvement in cash flow. AI has helped prepare the analysis, compare scenarios and assemble the presentation.
AI can usefully support work like this.
The finance director likes the impact on working capital. Operations sees an opportunity to run a leaner business. Procurement expects it can manage supplier deliveries. Sales hopes that the released cash will support other investments.
But finance assumes operations has checked the production implications. Operations assumes procurement has verified supplier lead times. Procurement assumes sales has considered what happens if a major customer places an unexpected order. Sales assumes somebody in the room has already challenged the proposal.
The slides are polished, and there are six more items on the agenda. The proposal is approved.
A few months later, a supplier delay coincides with a surge in customer demand. Stock runs short. The company has released cash, but it cannot fulfill every order. What looked like a sound financial improvement has created a commercial problem.
Reducing inventory might still have been the right decision. Perhaps it simply required a different safety stock policy. Our fictional company did not necessarily make the wrong choice, but it failed to test some of the assumptions that mattered most.
And now comes the uncomfortable question: who was responsible for that?
Meet ghost judgment, the colleague nobody remembers hiring
I recently came across the expression ghost judgment in an article from TNG, the Swedish recruitment company. It describes a situation where a decision looks carefully reasoned, but nobody has properly examined the underlying evidence or genuinely taken responsibility for the conclusion.
A recommendation can contain accurate figures and sensible calculations while still missing an important part of business reality. The ghost is the missing judgment between a polished recommendation and the decision someone makes with it.
A 2025 global study by the University of Melbourne and KPMG, based on a survey of more than 48,000 people across 47 countries, found that 66% of employees reported relying on AI output without evaluating its accuracy. That figure includes people who said they did so rarely, not just habitual users. Even with that qualification, it is a useful reminder that human review cannot be assumed.
One colleague prompts an AI tool, another tidies the output, a third forwards the recommendation. Each assumes somebody else has checked it. By the time the presentation reaches the boardroom, it has acquired an authority nobody consciously gave it.
The ghost has found a comfortable seat.
AI gives an old organizational weakness a much faster route
Poorly challenged recommendations existed long before generative AI. Business history is full of convincing spreadsheets and confident presentations that overlooked an important assumption.
What changes with AI is not merely how many recommendations appear. It is how quickly a plausible recommendation can move from an initial idea to a management decision, and how easily it can be repeated across teams.
A single set of assumptions can become the foundation for multiple forecasts, project plans or departmental recommendations before anybody has questioned it. Fast execution can then turn a local error into a wider problem while the organization is still discovering what happened.
AI can also be used to expose weaknesses in an argument and test alternative scenarios. The technology is not the villain of our Halloween story. The danger is allowing the speed of producing answers to outpace the organization's capacity to evaluate them.
Business acumen helps people spot what the numbers leave out
To challenge a recommendation intelligently, people need enough business understanding to know what deserves questioning. Telling an employee to "check the AI" will not help much if they do not understand how the decision affects the organization.
Consider the inventory proposal. Financial literacy helps explain why lower inventory releases working capital. Business acumen takes the analysis further: What happens to production reliability, supplier relationships, customer commitments and the ability to respond when demand changes?
A business can improve one metric while damaging something more valuable elsewhere. Reducing costs might weaken quality. Extending customer payment terms might win sales but put cash under pressure. Delaying investment might improve this year's profit at the expense of next year's capacity.
These are not arguments against the decisions themselves. They are reasons to consider the complete business system before acting. This is also where second-order thinking becomes useful: what happens after the immediate improvement?
Business acumen is the ability to understand how an organization creates value and how choices affect other functions, risks and results over time. It includes recognizing when one department's sensible decision becomes another department's problem.
AI gives us greater access to sophisticated analysis. That is a real advantage. But access to an impressive analysis should not be confused with the ability to decide whether it makes sense for a particular business.
A quiet meeting is not necessarily an aligned meeting
Even people with strong business acumen can make poor collective decisions when they consider only their own part of the business. Finance, operations, sales and procurement may all be acting sensibly within their responsibilities while pulling the organization in different directions.
This is where organizational alignment becomes essential. Alignment means understanding the common objective and the trade-offs required to achieve it, not simply agreeing with the recommendation that happens to be on the screen.
In our fictional meeting, each function had a reasonable reason to support lower inventory. But nobody was willing to ask whether the proposed reduction remained sensible when those perspectives were brought together.
There is a particularly dangerous form of agreement that I suspect many of us recognize. A proposal is presented, the most senior person nods, a few others agree, and the room goes quiet. The meeting appears remarkably efficient. Yet silence may reflect uncertainty, reluctance to disagree, or an assumption that somebody else knows more.
Speed of agreement is not proof of shared understanding.
Productive disagreement is often a sign that the organization is doing the work of alignment. People compare assumptions and consider consequences beyond their own department. The objective is not endless discussion, but a decision that can withstand different perspectives.
Our ghost rather enjoys functional silos. They give it plenty of places to hide.
Accountability belongs to people, while governance sets the conditions
Accountability starts with a person who can explain and own a decision. Someone should understand the assumptions, the alternatives considered and the risks accepted. If the outcome changes, that person must also know when to revisit the recommendation.
That responsibility cannot be assigned to a presentation, however polished, or to an AI tool that helped create it.
But accountability also depends on how the organization works. Leaders and governance structures together establish which decisions need formal review, whose expertise must be consulted and where approval authority sits. Leaders must also make it acceptable to question an apparently convincing recommendation without being seen as obstructive.
These are two different responsibilities. A decision owner is accountable for the choice. The organization is responsible for creating a process and culture in which that ownership is meaningful.
Clear ownership should not require endless committees. It can make decisions faster by removing ambiguity about who evaluates the evidence, who provides input and who decides. What it should prevent is a recommendation circulating through several hands without anybody truly examining it.
Better judgment must be practiced before it is needed
This creates an important challenge for Learning and Development. Organizations can publish AI guidelines and teach employees to ask better questions. Both are useful. But knowing a checklist is different from applying judgment when the information is incomplete, the meeting is running late and different colleagues want different outcomes.
People need opportunities to practice decisions, challenge assumptions and examine consequences. They benefit from discussing a problem with colleagues who see it differently, discovering what they overlooked and reflecting on why a particular choice worked or failed. Realistic business scenarios and experiential learning can provide that practice without making customers or shareholders pay for every mistake.
Learning alone will not solve weak accountability or poor governance. Equally, governance is less effective if the people expected to exercise judgment have not developed the business understanding or confidence to do so.
Three questions to ask before the next approval
Fortunately, you do not need a Halloween ritual to deal with ghost judgment. Before approving a consequential recommendation, especially one prepared with AI, try three ordinary questions.
What must be true for this recommendation to work? Identify the assumptions and test the ones that could change the outcome. In our inventory case, supplier reliability and demand volatility deserve attention alongside financial savings.
Who sees this differently, and have they been heard? Bring in the perspectives missing from the discussion. A decision that works for finance may create a difficult problem for operations or customers. Alignment means testing those trade-offs before approval, not discovering them afterward.
Who owns the reasoning as well as the decision? The owner should be able to explain why the recommendation was accepted, what risks remain and when the outcome will be reviewed. An approval is more meaningful when someone can defend the thinking behind it.
Keep the ghosts for Halloween, not the boardroom
At Celemi, we help organizations build business acumen and shared understanding through experiential learning and team-based business simulations. People practice business decisions, see the consequences and learn to challenge assumptions together. Those capabilities are valuable far beyond AI, but AI gives us another reason to develop them.
The technology can be an excellent thinking partner. We should welcome its contribution while keeping responsibility for decisions where it belongs.
One small confession: AI helped me draft and edit this article too. I reviewed the arguments, challenged the assumptions, and take responsibility for the final text. Hopefully, no ghosts slipped through the editorial process.
So if an especially convincing presentation arrives in your inbox this Halloween, admire the efficiency. Appreciate the formatting. Then ask the question the ghost would prefer you forgot:
Who has really thought this through?