Who makes the better call: a manager with a wide-ranging career or a narrow one; a doctor who trusts their instincts or one who sticks to clinical guidelines; an algorithm that assesses job candidates or human recruiters?

In this month's round-up of INSEAD research, we look at how people and machines make high-stakes decisions, including those above. Other insights featured cover how generative AI can produce ads that are attractive and on-brand, and the finer points expatriates need to ponder when it comes to buying or renting a new home. 

Does a varied career make managers better at spotting winners?

Is a manager who has worked in numerous firms better at predicting which new products will succeed? It depends on where that experience comes from.

By tracking managers' share purchases ahead of product launches, Philipp Meyer-Doyle and his fellow researchers found that those who had worked across several firms and functions read their company's prospects better. But managers who had moved between industries underperformed peers with less varied work experience. The gaps widened further when markets were especially volatile.

Read the paper

Transforming digital advertising with generative AI

Generative AI can churn out endless ad images, but many are either not attractive or not on-brand. Remi Daviet and his co-author developed a cost-effective active learning approach that trains AI to meet both objectives. They paired an image generator with two prediction models and tested the system in a live advertising campaign for an outdoor activities company. The generated ads drew more clicks and were more consistent than those from a professional human designer or an aesthetics-focused rival AI model.

Daviet’s research provides a validated methodology for addressing two challenges in creative design: finding effective ideas and keeping them on brand.

Read the paper

When doctors stray from clinical guidelines

Guidelines tell doctors when to order follow-up ultrasound scans, but doctors don't always follow them. At Singapore's largest obstetrics and gynaecology hospital, Xinyu Liang, Srishti Arora and Michael Freeman found that over-scanning occurred in 29.1% of decisions, nearly six times the rate of under-scanning. Each deviation led to an average of 0.54 additional scans over the next 12 months. This effect was smaller when doctors noted their reasoning for ordering the scans.

The study shows that, by retrospectively analysing unstructured medical records, hospitals can assess the impact of guideline departures and implement measures to reduce low-value deviations while allowing doctors to use their discretion where needed. 

Read the working paper

To use or not to use an AI hiring tool

In this case study, Julien Clément and Victoria Sevcenko outline how OmniCorp, a fast‑growing Singapore software firm, is tempted to adopt HireMe, an HR analytics tool that uses AI to scrape candidates’ public profiles and score them on five measures: skill‑match, sociability, conscientiousness, leadership potential and cultural fit. The tool could cut recruiting costs and replace early interviews. But do its scores really predict who gets hired, or just reflect past biases? 

Using data on 450 past candidates, students assess the tool’s predictive power, identify selection bias and debate the fairness and legal risks of algorithmic hiring.

Access the case

Rent, buy or belong?

Should you buy a home or keep renting? In this case study featuring an expatriate couple in Paris by Pierre Mabille, the answer involves more than the cost of a mortgage. Students work out what the couple can afford versus what they should buy, and why low rental yields in Paris can make buying a poor financial bet. In the process, they learn to evaluate affordability, apply the principles of mortgage amortisation, and put a price on the harder-to-measure benefits of owning: stability, freedom and a sense of belonging.

Access the case