Application volume grew every year. Maintaining rigorous selection of candidates became challenging without increasing headcount.

Introduction

Every year, the number of applications to ISB's ivi programme grew. Every year, the pressure to evaluate each one with the same rigour as a faculty-led interview grew with it. At low volumes, the evaluator could personally ensure every selection decision was rigorous. At thousands of applications, that became impossible — but lowering the bar wasn't an option for a programme built on academic excellence.

We built an evaluation engine.

The Problem

As application volumes scaled year on year, the admissions team faced a structural problem: manual evaluation couldn't keep pace without compromising consistency. Reviewing each candidate meant reading a resume, assessing a portfolio, forming a view on fit — then doing it again, thousands of times. Fatigue would introduce bias. And, a less rigorous screening process would have missed the very qualities in a candidate, the ivi programme was designed for.

The Solution

• ✓Worked directly with the Dean to map the evaluation framework — translating their judgment into a structured, AI-powered rubric that could be applied consistently at scale
• ✓Each interview was dynamically generated from the candidate's own resume, LinkedIn profile, portfolio, and submitted documents — every session uniquely relevant to that individual
• ✓NLP-based rubric scoring aligned to ISB's own criteria, generating quantitative scores, qualitative feedback, and psychometric insights per candidate
• ✓Faculty review AI-evaluated shortlists only — human time reserved exclusively for final-round interviews

Key Outcomes

• ✓Thousands of interviews conducted at scale with no compromise on evaluation quality
• ✓Every interview was personalised to the candidate's background — assessments were more revealing and harder to game
• ✓Shortlists achieved 97% agreement with final faculty assessments, validating the rubric co-developed with the Dean
• ✓Faculty time saved — all upstream evaluations handled end-to-end by Pyzo