University of Massachusetts Amherst

Prof. Carolina Rossini, JD, LL.M., MBA, serves as serves professor of practice at the School of Public Policy at DACSS Faculty Associate and Director for Programs at the Public Interest Technology Initiative at the University of Massachusetts Amherst. A recognized leader in technology policy, she advances research, education and cross-sector collaboration to ensure technology serves the public good.

        Most conversations about AI in higher education center on faculty and students: how professors should handle AI-generated essays, whether students are learning or outsourcing their thinking, and what new research methods become possible. These are legitimate concerns. But they skip over the population that actually makes the university function: staff.

Administrative professionals, IT teams, HR coordinators, financial aid officers, registrar employees, facilities managers, and communications offices. These people process thousands of transactions daily, navigate legacy systems, and absorb the operational shock every time a new technology arrives on campus. When institutions adopt AI, staff are often the last to receive training and the first expected to operationalize the tools. That asymmetry is where much of the risk and the unrealized opportunity sit.

What Other Sectors Already Learned

The cross-sector evidence is substantial enough to be blunt about. A 2026 MIT study reported that 95% of generative AI pilots fail to produce a measurable financial impact. Stanford's 2026 analysis of 51 successful enterprise deployments identified the determining factor: not model quality, but workflow integration. Organizations that redesigned workflows before selecting tools succeeded. Those that bolted AI onto existing processes did not. AI deployment is organizational change management with a software component. That distinction matters for staff because staff workflows are exactly what gets reorganized or ignored.

The use cases that have scaled share a pattern: bounded inputs, bounded outputs, and a human reviewer in the loop. Customer service triage, clinical documentation, and document processing. Each succeeds because AI handles high-volume structured tasks while a person retains judgment. That maps directly onto University staff work. Projects that have embedded students in state agencies and non-profits aim to build AI tools, including HR chatbots, call-center onboarding assistants, and multilingual training platforms. Each addressed a staff workflow problem.

The reports have shown that what is not scaled is autonomous AI decision-making, especially in high-stakes contexts. Wherever error costs are high and audit requirements are stringent, AI remains a decision-support tool, not a decision-maker. Universities should internalize this before deploying AI in admissions, financial aid, or advising.

What Public Interest Technology Teaches Us

My work with the Public Interest Technology Initiative at UMass Amherst, and before that, co-founding think tanks and working with organizations that address digital rights more broadly, has shaped a clear conviction: technology adoption without a governance architecture is just risk transfer. You move the burden from the institution to the individual, to the staff member deciding on the fly whether it is safe to paste student data into a chatbot.

Public interest technology programs, now present at universities across the USA, exist to close that gap. They embed governance, ethics, and community accountability into technology work from the start. For staff, this means literacy that goes beyond "here is how the tool works" to "here is what the tool does not tell you." The equity dimension is concrete: when commercial AI subscriptions cost $20 or more per month, a departmental administrator is just as priced out as a graduate student. Governed institutional platforms eliminate that barrier, but only if staff are included in the design from the beginning.

What Leadership Owes Its Staff

Institutional leadership tends to treat AI adoption as a technology procurement question. It is not. It is a workforce development, governance, and equity question. Getting those right is precisely what makes procurement decisions defensible.

“AI deployment is organizational change management with a software component.”

The cross-sector evidence on adoption is unambiguous: workers who experience AI as a productivity tool adopt it willingly; workers who experience it as surveillance or a job threat resist. One enterprise survey found 29% of employees actively sabotaging their company's AI strategy. The pattern that works is voluntary adoption of tools that solve a real problem the worker actually has.

Leaders need to act on three fronts. (i) First, provide governed tools with clear data policies and institutional oversight, so staff never make compliance decisions that belong at the institutional level. (ii) Second, invest in sustained training. Running AI literacy workshops for municipal employees taught us that effective adoption requires repeated engagement, real-world examples, and a culture where questions are normal. (iii) Third, create feedback channels. Staff sees failure modes that leadership does not. These observations are governance data, and institutions that fail to collect them are flying blind.

Recommendations for Institutions

Stop building AI strategies around faculty and students alone. Map staff workflows with staff. Co-redesign before you deploy. Build or adopt governed platforms rather than leaving staff to navigate the commercial AI landscape on their own. Treat AI literacy as ongoing professional development, not a checkbox.

And take the public interest technology framework seriously. The answer always involves governance, community participation, and accountability, as well as asking who benefits and who bears the cost. For university staff, the people who keep the institution running while everyone else debates the future of knowledge, that question has been deferred long enough.