What Not to Delegate to Artificial Intelligence in Management

A company starts using an algorithm to score candidates for a vacancy, saving the recruitment team hours of work. Months later, it discovers the algorithm has been systematically penalising one group of candidates, without anyone consciously deciding that. Nobody wrote that rule. But somebody had to answer for it, and the algorithm could not be that person.

Which decisions still require a human?

Three types of decision resist delegation, even to the most advanced AI. Decisions about people, such as hiring, promoting, disciplining, or dismissing, where human context weighs as much as the data. Decisions that define values, such as accepting or turning away a client for ethical reasons, where no purely numerical criterion gets it right. And decisions that carry public accountability, where somebody has to be willing to explain, under their own name, why the company chose that path.

Can an AI take responsibility for a business decision?

It cannot, for a simple reason: responsibility implies the capacity to answer for something, and an algorithm faces no personal consequences when it gets things wrong. It does not lose its job, its reputation, or a night's sleep. Whoever signs off on an AI-supported decision remains a real person, with a name and a title, and it is that person who carries the weight of having accepted it, even when the suggestion came from a machine.

How do you tell AI support apart from a substitute for judgement?

The test is simple to state, though hard to meet in practice: can the decision-maker explain, in their own words, why they accepted the AI's recommendation, or are they simply trusting it blindly because 'the system said so'? In the first case, real support exists. In the second, judgement has been replaced and dressed up as efficiency, and it is exactly in that second case that the costliest mistakes tend to happen, because nobody feels they have the authority, or the obligation, to question the machine.

Why is it risky to delegate people decisions to an algorithm?

Because an algorithm learns from past data, and the past of any organisation carries its own biases, including the ones nobody has ever said out loud. A company that uses AI to screen candidates without first examining its own hiring biases ends up automating the discrimination that already existed, just faster, and with less chance of a human catching it midway through the process.

What is the Digital Ouroboros, and what warning does it carry?

The Last Asset names this risk the Digital Ouroboros, the serpent biting its own tail: a system that, instead of correcting the errors it inherits, amplifies them at a scale and speed no human reflection can keep pace with. An organisation that rolls out AI without first examining its own internal patterns, as the book puts it, ends up 'automating its complexes' rather than automating efficiency.

What real example shows this risk playing out inside a company?

The Last Asset cites the case of one of the world's largest consultancies, which went as far as tying internal promotions to how much its consultants used AI tools. At first glance, it looks like an incentive for modernisation. Looked at more closely, the scheme rewards adoption, not judgement. It measures who uses the tool most often, not whether that use is improving or worsening the quality of the decisions that person makes. One of the firms that advises other companies most on digital transformation ended up teaching thousands of client organisations that speed of adoption counts for more than direction.

What mistake do companies make when they ban AI from these decisions outright?

The opposite extreme is also a mistake. Banning any AI support from sensitive decisions altogether denies the team a faster, and often more complete, analysis than it could produce alone. The real problem sits with whoever uses the tool without examining their own assumptions, and with whoever treats a machine's suggestion as an order rather than an opinion to be weighed like any other.

How do you design a decision process that uses AI without losing human control?

Start by naming, for each type of decision, the person who signs off on it, before any tool enters the process. Then require that person to justify the decision without reaching for the line 'it's what the system recommended'. Finally, submit the decision process itself, not just the outcome, to periodic review. A well-designed AI decision process protects exactly this space, and a leadership diagnostic helps identify whether your team has already lost that human control, without realising it, over decisions that should never have been fully delegated.

What responsibility does senior leadership carry in this design?

It falls to senior leadership to decide, before any tool gets purchased, where the line sits between support and a substitute for judgement inside their own organisation. It is a decision about values and about risk, not a technical question to leave solely to the technology department, which is precisely why AI governance has stopped being a matter of legal compliance alone and become a core responsibility of leadership itself. A programme such as Deep Leadership 3D helps an executive team build this discernment before it is needed, rather than discovering it too late, after an expensive mistake has already happened.