What the research found

The interviews explored how people understood fairness, how they tried to put it into practice and what happened when AI became involved. I developed seven themes from their accounts, presented here in plain language.

These themes describe patterns in a qualitative study, not how frequently something happens across all businesses. The evidence comes from what participants reported, rather than direct observation of their organisations.

Fairness meant different things across people and processes

People used the same word while referring to different concerns: treating people consistently, taking account of individual circumstances, following a sound process or meeting legal obligations. Even within recruitment, what counted as fair depended on which part of the process someone was considering.

Agreement that fairness mattered could therefore conceal significant differences. Following the same policy did not necessarily mean people were making the same judgement about the same concept.

The motivation for fairness varied from idealistic to cynical

Participants described ethical commitments, legal requirements, reputation and expectations of business benefit. These reasons could coexist, but they did not carry equal weight in everyday decisions.

Legal compliance was a particularly strong operational driver in the accounts. Business benefits were often asserted with less clarity about the evidence behind them, while personal commitments did not necessarily translate into organisational resources or priorities.

Confidence in Fairness was easier to express than demonstrate

Participants could be confident that decisions were fair without being equally clear about how to establish that. Organisations collected data, but the categories and measures available did not answer every question participants considered important.

This was not simply a shortage of information. It also concerned the relationship between fairness and the things used to represent it: what was counted, what was left out and what conclusions the measures could support.

Fairness often lost out to other business & personal priorities

Fairness operated alongside pressures for speed, cost control and commercial performance, as well as preferences and established ways of working. Participants described situations in which these competing considerations took precedence.

Those conditions shaped the room people had to act. The findings do not require an assumption that every compromise was cynical or that commercial priorities were illegitimate; they direct attention to which trade-offs were being made and what they meant in practice.

Businesses addressed fairness with widely varying piecemeal solutions

Participants described policies, training, process controls, technology and efforts to influence individual behaviour. These measures addressed different parts of the problem, but did not necessarily form a connected response.

A control designed for one stage could leave another untouched. Where evidence of fairness was uncertain, it was also difficult to establish whether an intervention was doing what people hoped.

Making a decision and judging its fairness were different tasks

The accounts raised questions about the difference between producing a decision and deciding whether it was fair in its circumstances. A process could be consistent and computationally manageable without resolving every contextual judgement involved.

Participants differed in how they understood that distinction and the implications for human involvement. These findings concern the technologies and experiences discussed in the interviews; they are not a definitive assessment of everything AI can or cannot do today.

Responsibility did not align neatly with control & influence

HR participants’ accounts suggested that retaining responsibility for people decisions did not necessarily give them control over all the ways AI shaped those decisions. The analysis also identified a separation between HR involvement and wider organisational efforts to govern AI.

That directs attention to the relationship between the people expected to oversee a decision and those who can change the systems, priorities and processes behind it.

Bringing the findings together

The findings suggest that fairness cannot be understood solely by examining an algorithm or asking whether a person approves the final result. Its treatment also depends on organisational priorities, different meanings, available evidence and the distribution of influence.

The thesis uses management theory to help explain these patterns and to develop arguments about upstream governance and differences in understanding. Those are interpretations and contributions built from the evidence, rather than additional events directly observed in the interviews.

Read how the study was carried out or explore the contributions and continuing research. For the broader context, return to the research overview.