AI in higher education
AI should improve institutional outcomes, not demonstrate that you are using AI.
Most universities now have more AI opportunities than they can pursue. The scarce thing is not ideas. It is the judgement to choose among them and the institutional capability to convert a chosen one into a measurable result.
The problem as it actually presents
An institution accumulates AI use cases from every direction. Faculty experiment. Administrative functions propose. Vendors arrive with demonstrations. The board asks what the AI strategy is.
What follows is usually a portfolio of disconnected pilots, each defensible on its own, none clearly tied to a number the institution is measured on. Two years later there is considerable activity, a reasonable amount of spend, and no outcome anyone can point to.
The failure is rarely technological. The model works. The tool is used. What does not happen is the institutional change around it: the process redesign, the clear ownership, the data foundation, and the shift in how people actually work. Without that, a working technology produces no better result.
How I work on it
Start from the outcome
Not "where can we use AI" but "which outcomes do we most need to improve, and can AI materially move them?" Student outcomes, institutional productivity and growth are the three places the answer usually lives, and naming the specific measure comes before naming the technology.
Make the strategic choices explicit
A strategy is a set of decisions about what to pursue now, what to experiment with, what to defer, and what not to pursue at all. It also names the capabilities that must exist first. These are often unglamorous, like an integrated student data foundation that three separate use cases all silently depend on.
Design for the institution, not the pilot
A pilot can work around fragmented data and exceptional effort from a small team. Institution-wide implementation cannot. Ownership, process, governance and incentives decide whether a capability becomes a result, and scaling is what exposes the problems the pilot was able to avoid.
Measure the result, not the activity
Adoption counts and query volumes tell you the technology is being used. They do not tell you it was worth using. The measures that matter are the ones that justified the initiative: conversion, progression, placement quality, turnaround time, capacity released, cost per enrolled student.
What I bring to it
Two things, and the combination is the point.
I spent thirteen years building software as an engineer, a founder and a CTO in a derivatives risk business, so I can tell you what a system will genuinely do, where implementation becomes hard, and when a vendor's claim is doing more work than the product.
And I spent close to fourteen years running business schools, accountable for the outcomes AI is supposed to improve. I have made the institutional decisions on CRM, on Coursera for Campus, on ChatGPT Edu. Buying technology, deploying technology, adopting technology and producing an outcome are four different things, and the distance between the third and the fourth is where most institutional AI efforts quietly stop.
The purpose is not to help an institution use more AI. It is to help it decide where AI deserves attention, and to make those choices into implemented, measurable change.
Articles on this
Where should a university actually use AI?
The challenge is not identifying what AI can do. It is deciding where AI can materially improve an institutional outcome that matters.
AI use cases are not an AI strategy
Strategy requires choices about what to pursue, what capabilities to build, and what to deliberately defer or not pursue.
Why AI transformation is primarily an institutional problem
Data, processes, systems, roles, incentives and governance usually determine whether AI produces institutional value.
From AI experiment to institutional outcome
A successful pilot shows that something can work. The harder test is whether the outcome that justified it actually improves.