Subscribe
There is a quiet crisis unfolding in business school classrooms across India and the world. It is not the kind that makes headlines. But it is reshaping the very foundation of what we claim to measure when we evaluate our students.
A significant portion of the assessments we currently use can now be completed with reasonable competence by a large language model (LLM) in under sixty seconds. This includes case analyses, essay questions, term papers, and reflective journals. ChatGPT, Claude, Gemini, and their successors are not novelties any longer. They are productivity tools that our students carry in their pockets. The question is no longer whether students are using them. The question is whether we are designing assessments worthy of a world in which they exist. For most business schools, the answer is an uncomfortable no.
Today’s LLMs can summarise large volumes of text, structure arguments in polished prose, generate frameworks and reports, answer conceptual questions across business domains, and even role-play personas. That last capability matters more than it sounds. An LLM can convincingly simulate the student who studied the chapter.
What AI does considerably less well is exercise genuine contextual judgment with real stakes attached. It cannot draw on lived experience in a specific organisation. It cannot navigate ambiguity under social pressure or demonstrate values under conditions where values cost something. This gap is exactly where the future of business school assessment must be anchored. AI can produce. It cannot authentically embody.
Traditional assessment rests on three assumptions that AI has quietly demolished. The first is the reproduction assumption: that reproducing knowledge in a structured format proves learning. AI can reproduce virtually all business school knowledge on demand. The second is the isolation assumption: that exams in isolation ensure the work is the student’s own. But business is never conducted in isolation. Leaders use advisors, databases, and AI tools. The isolation assumption trains students for a world that does not exist. The third is the product assumption: that we should assess the final essay rather than the thinking that produced it. AI has decoupled product quality from thinking quality entirely.
Continuing to grade the product while AI writes the product is not merely pedagogically unsound. It is institutional self-deception.
Reimagining assessment does not mean abandoning rigour. It means relocating it. From the output to the process. From the text to the person. From recall to reasoning.
Process-visible assessment asks students to show their thinking, not just their conclusions. A reflective memo submitted alongside each assignment asks: what did you struggle with, what did you change after your first draft, and why. Version-controlled submissions require students to show multiple drafts with annotations. These methods do not prevent AI use. They make AI use visible and accountable. A student who used an LLM well and built on it critically has demonstrated something valuable. A student who submitted the output verbatim has demonstrated nothing.
The oral examination is one of the oldest assessment formats in academia and one of the most AI-resistant. When a faculty member asks follow-up questions in real time, the conversation quickly separates genuine understanding from borrowed prose. Questions such as “why did you choose this framework?” or “how would your recommendation change if the company were family-owned?” cannot be answered by AI on a student’s behalf in a live exchange. Mandatory oral components, even brief ones, and random oral audits of submitted assignments would raise the integrity of assessment significantly. Producing polished text now costs nothing. The ability to speak to one’s ideas with precision is a genuine signal of learning.
Organisation-specific, embedded projects are inherently more AI-resistant because LLMs are trained on public data and perform poorly on local, proprietary contexts. A student asked to analyse a well-known company’s competitive position can use AI freely. A student asked to help a mid-sized Hyderabad-based textile exporter decide whether to invest in a direct-to-consumer digital channel, after two hours of conversation with the owner, is engaged in something an LLM cannot replicate.
AI-augmented assessments go further: rather than pretending students do not use AI, design tasks where AI use is declared and itself evaluated. Prompt engineering tasks grade the quality of a student’s prompts and their critical evaluation of the outputs. Adversarial AI critique asks students to identify the logical gaps and ethical blind spots in an AI-generated strategy report. These assessments build the capabilities that will make graduates valuable in AI-augmented workplaces.
Indian B-schools must navigate specific regulatory constraints. Frameworks set by UGC, AICTE, NAAC, and NBA still privilege standardised end-semester examinations. A pragmatic path forward is a hybrid architecture: retain centralised exams for foundational knowledge validation while shifting the internal assessment component toward richer formats. This can begin immediately, without waiting for regulatory reform.
But the hybrid architecture must go further than tinkering with internal marks. The weightage of the end-term examination itself must be reconsidered. In a typical in-person MBA programme, the end-term carries forty percent of the final grade. In online programmes, that figure rises to seventy per cent. Both numbers are difficult to defend in the age of AI.
The forty per cent figure still means a student’s grade is heavily determined by a single sitting under artificial conditions. A student who has done outstanding project work and shown genuine intellectual growth over fifteen weeks can be undone by one bad morning. That is not a measurement of learning. It is a measurement of test-day performance.
The seventy per cent figure for online programmes is harder still to defend. It exists for logistical convenience, not pedagogical reasons. Assigning seventy per cent of the grade to the format most susceptible to AI assistance is a structural problem the sector can no longer ignore.
A compelling case can be made for capping end-term examinations at less than twenty per cent of the final grade in both in-person and online programmes. The exam can still verify baseline knowledge and provide a standardised check across cohorts. But the remaining eighty per cent should reflect the full arc of learning through continuous assessment, oral evaluations, live projects, and portfolios. For autonomous institutions and deemed universities, the regulatory flexibility to make this shift already exists. The barrier is not regulatory. It is inertia.
The goal of management education has never changed: develop leaders who can think clearly, act ethically, and create value under uncertainty. What has changed is the environment in which those leaders will operate.
In that environment, producing prose without understanding it is worth nothing. Generating a framework without the judgment to know when it applies is worth nothing. What is worth something is the capacity to ask better questions than the machine, to bring genuine context to generic tools, and to exercise judgment that is accountable to real people in real situations.
If our assessments measure those things, AI is not a threat. It is a mirror showing us, with uncomfortable clarity, what we were never actually measuring in the first place. It is time to look in that mirror and redesign what we see.
Sanjay Fuloria is a Professor and Director of the Centre for Distance and Online Education (CDOE) at ICFAI Foundation for Higher Education (IFHE), a deemed-to-be University in Hyderabad. He teaches Managing Digital Transformation and Quantitative Methods in the MBA programme and researches at the intersection of AI, consumer engagement, and digital transformation.
Views expressed are the author’s own and don’t necessarily reflect those of Down To Earth