Are We Making Decisions, or Producing Hallucinations?

“Every hallucination is an artistic creation of the soul; the brain that creates it shapes and adorns it with its own aims and purposes.” – Alfred Adler

AI never leaves the agenda.

There is one concept we hear constantly, especially in connection with large language models (LLMs): AI hallucination.

LLMs can build incomplete or faulty patterns and then produce wrong outputs in extremely confident language. This is usually flagged as “the soft underbelly of AI.”

To me, the real question is different:

Is hallucination a problem only for algorithms?

1. Decision bias: conclusion first, data second

In business, a great many decisions are still made through instinct and inclination rather than after data analysis. The supporting arguments are produced afterwards.

A kind of reverse engineering.

In this approach, the whole construct rests on the consistency of the human mind — because the mind does not want to revise a decision once it has been made. But consistency and truth are not the same thing.

As in every other part of a company, this kind of approach is extremely dangerous in logistics operations. Because topics such as:

• capacity planning
• inventory levels
• new warehouse investment
• route optimization

and SLA targets all demand decisions grounded entirely in data.

If the decision comes first and the data is then collected to fit the justification, the process is not analytical. In Adler's words, it becomes a mental construct adorned with aims and purposes.

At that point it is no longer a technical process but a psychological one.

2. How operational hallucination forms

Hallucination in operations is like fixing the answer and building the equation afterwards. Data is used not to make the decision but to support it.

Imagine a logistics manager at a manufacturing company who decides, instead of buying 3PL services, that “we can do this 20% cheaper with our own team.”

Having made that decision, he selectively brings forward the following data:

• high unit-price quotes from 3PL providers, inflated by heavy specifications (because it is the specification that determines the cost)
• only “direct labor” in the staffing cost
• the “loss of control” risk associated with outsourcing.

But the following invisible costs are ignored:

• Human resources burden: severance and notice pay, accrued annual leave, reinstatement lawsuits and similar costs.
• Overhead allocation: meals, staff transport, electricity, water and other shared costs.
• Operational risks: inventory discrepancies, damages, missing items, penalties.
• Economies of scale and flexibility: the 3PL provider's technology infrastructure and the resources it shares across its other clients.

And what happens in the end? The purpose of collecting data shifts from finding the best solution for the company to serving the manager's own objective.

This brings a concept into view: job insurance.

“Job insurance” is the shield some employees build to make themselves indispensable to the organization. Sometimes it means monopolizing information, sometimes making it more complicated than it needs to be, sometimes manipulating the data outright.

This too is a form of corporate hallucination.

3. The parallel between AI and the human mind

When AI builds faulty patterns and draws confident but wrong conclusions from data, we call it AI hallucination.

When a human being does exactly the same thing, we usually do not notice.

The mechanism is strikingly similar in both cases:

• Faulty premises → “This warehouse is inadequate anyway.”
• Selective use of data → considering only the peak period.
• Ignoring alternatives → never examining a different layout.
• Post-decision rationalization → “This outcome was inevitable.”

When algorithms do it, we call it a “model problem.” When people do it, we call it a “mistake” — or “experience.”

In truth both share the same problem: forming the conviction first and arranging the data around it afterwards.

Demanding perfection from technology while treating human bias as normal is a serious contradiction.

The real issue is this:

We audit the algorithms. How much do we audit our own decision processes?

Mini tool: a decision audit

Before and after making a decision, ask these five questions:

• What measurable data is this decision based on?
• Could data arriving after the decision change my mind?
• Have I looked at alternative scenarios?
• Is there data that would support the opposite decision?
• Did we make a decision, or did we go looking for data to fit one?

If the answers are not clear, you are in risk territory.

Conclusion

With algorithms:

• routes are optimized.
• inventory accuracy improves.
• cost visibility is achieved.

But no system on its own can eliminate mental bias.

Without a transformation of mindset, no transformation succeeds — digital transformation included.