Process Inflation and the Diffusion of Accountability

Each intervention can be rationalised as improving the quality of the hiring decision. Collectively, however, they do not necessarily produce a better understanding of the candidate.

Many systems continue to operate on relatively thin representations of a person: CVs, application forms, keyword matches, interview narratives and standardised assessments. Technology has reduced the marginal cost of processing those signals. It has therefore become possible to screen more candidates, involve more stakeholders and collect more documentation.

Processing weak information more extensively does not make the underlying information stronger.

Some additional process is valuable. Structured interviews, relevant work samples and properly validated assessments can produce materially better evidence than informal judgment alone.³ The concern is not assessment itself. It is the accumulation of stages without a clear account of what each stage measures, what new information it contributes or whether it improves the decision.

This helps explain why conventional credentials remain disproportionately influential. Pedigree, previous employers, job titles, sector experience and prior private-equity exposure are highly legible. They are easy to communicate internally and easy to defend after the event.

Judgment, operating instinct, adaptability, decision-making under uncertainty, influence and the capacity to recognise what matters in an unfamiliar situation are considerably harder to observe. Yet these less-legible characteristics may be precisely what differentiates exceptional performers.

The problem is therefore not simply insufficient information. It is a measurement architecture that can privilege what is easy to observe and defend over what may actually predict performance.

Much of the visible application of AI in recruitment remains downstream: sourcing candidates, parsing CVs, generating outreach, screening applications, ranking profiles, summarising interviews and administering workflows. Large recruiting datasets show how application volume, evaluation complexity and interview stages interact across modern hiring funnels.¹

That can create substantial productivity improvements. But if the candidate representation and evaluation model remain unchanged, AI primarily allows organisations to perform the existing process faster, cheaper and at greater scale.

Selection researchers make the relevant distinction explicit: an AI tool still requires evidence of reliability, validity and fairness for its intended use. Technical sophistication is not itself proof of predictive value.²

Meanwhile candidates are using the same technology to optimise CVs, applications, cover letters and interview responses against those systems. Both sides become increasingly effective at optimising against each other's filters.