Not All Hiring Is Equal: When AI Should Automate, Assist, or Step Aside

By Danilo Dukanović.

We’ve spent the last few years pouring billions into AI for HR and recruitment. Vendor decks promised higher productivity, less bias, and sharper hiring decisions (Chowdhury et al., 2023). 

Reality has been slower. A recent MIT-linked analysis shows that 95% of enterprise generative-AI pilots fail to deliver a measurable impact on profit and loss (Potluri & Serikbay, 2025; Estrada, 2025), and recruitment is no exception. 

We are now standing at the edge of the AI hype cliff. Organisations are starting to ask themselves: 

Not “Will AI replace HR?” but “In which context does AI actually make sense?”  

Two Lenses for When AI Belongs in Hiring 

To answer the important question, two lenses help bring clarity. 

The HIRE framework (Will et al., 2023) compares AI with humans in recruitment across four dimensions: 

At the same time, Phillips’ long-standing framework for investing in people also works when deciding how much to invest in AI in HR. His five investment strategies force us to weigh benefits, risks, and payoff instead of following the crowd (Phillips & Phillips, 2002): 

1. let others do it; 
2. invest the minimum;  
3. invest with the rest;  
4. invest until it hurts;  
5. invest as long as there is a payoff  

Both frameworks point to a simple truth: 

Not all hiring is equal – and AI investment should reflect that. 

Below is a practical way to make sense of that complexity across three broad role types: 

A) “Swap-Out” Roles 

Role type: High volume, high churn, low retention focus. 
Examples: Manufacturing, hospitality, service work. 
Typical question: “Can they show up, do the basics safely, and start tomorrow?” 

Assessment here focuses on simple, observable conditions: 

For these roles, the evidence is clear: 

On Phillips’ scale, this hiring type warrants: 

There is no competitive advantage in having humans manually process hundreds of similar applications when a reliable model can do it faster and more consistently. 

Automate almost everything early. Reserve human time for safety issues, regulation, or onboarding – not checking if someone can work weekends. 

B) “Scale-Up” Roles 

Role type: High volume, but talent is an investment. 
Examples: Banking, FMCG, consulting – analyst-to-partner paths, merchandiser-to-director, junior-to-senior management. 

These organisations don’t need “capacity”; they need pipelines. They should be hiring for: 

Here, emerging research makes the picture more complex. 

AI-inferred personality models (Fan et al., 2025; Dukanović & Krpan, 2025) show good reliability and convergent validity, but more variable predictive validity compared with traditional psychometrics. The HIRE review shows that candidates perceive AI processes as less fair and less trustworthy, even when AI performs competently (Will et al., 2023). 

AI is excellent at handling volume and structuring information, but still clumsy at deeply understanding people in the way these roles demand. 

Phillips’ strategies offer clarity: 

This balance is especially important in markets where pressure pushes companies toward over-automation.  

The UK graduate market is one example: ~17,000 graduate vacancies versus ~1.2 million recent graduates in 2025. Under this strain, many organisations treat graduate hiring like a Swap-out exercise, relying on opaque cut-offs or AI-only filters. Short-term efficiency increases, but long-term damage includes weaker pipelines, reduced diversity, eroded trust, and a brand that signals “commodity employer.” 

The future here is small, elite, tech-savvy People teams supported by multiple specialised tools across sourcing, onboarding, development, and succession. AI should assist, not replace – enabling consistency and structure while humans make integrated, high-judgment decisions. 

C) “Don’t-Mess-This-Up” Roles 

Role type: High-stakes, high-demand specialist and leadership roles. 
Examples: CTOs, senior ML scientists, partners in professional services, senior clinicians. 

These candidates are rare and highly literate in technology. They have options. Imagine asking a prospective CTO or Big 4 partner to complete an AI-only interview. Many will either game the system using their own AI tools (because they can), or walk away entirely. 

Research on applicant reactions shows that senior candidates react negatively to overly automated, hard-to-challenge evaluation processes (Nørskov et al., 2024; ICO, 2025). As stakes rise, the expectation for human contact increases, not decreases. 

Phillips’ framework makes the investment logic obvious: 

Here, it is crucial to keep the front-stage human. AI should work behind the scenes as a decision-support analyst, not as the interviewer or the “face” of your employer brand. 

Using the Three Types in Practice 

The goal of this model is to provide a framework for evaluating the use of AI in recruitment processes depending on context: 

Conclusion 

Seen through HIRE and Phillips together, the debate shifts from “Will AI replace HR?” to a more useful question: 

For each role type, how much should we invest in AI — and where must humans stay firmly in the lead? 

That is the conversation worth having as we step off the edge of the hype cliff and into the decade of real adoption. 

About the Author 

Danilo Dukanović is a psychologist and behavioural-science entrepreneur. He earned a BSc in Psychology, then an MSc in Cognitive Psychology at the University of Novi Sad. After two years in boutique management consulting at CS&C in Podgorica, he completed a second master’s degree at the London School of Economics in Behavioural Science, specialising in technology and behavioural science. He then founded Recrewty and serves as CEO, helping organisations hire, promote, and enable internal mobility using behavioural science and AI. Recrewty works with leading banking, retail, and telecom enterprises across Central and Eastern Europe.  

References 

Chowdhury, S., Lukac, M., Ikeda, S., & Sato, H. (2023). Unlocking the value of artificial intelligence in human resource management: Efficient tools, quality concerns and the role of HR managers. Human Resource Management Review, 33(4), 100995. https://doi.org/10.1016/j.hrmr.2023.100995 

Dukanović, D., & Krpan, D. (2025). How well can an AI chatbot infer personality? Psychometric comparison of machine-inferred and questionnaire-based traits in selection (Unpublished manuscript).