If you are a CRM, IT leader, architect working in Higher Education exploring opportunities with Gen AI, my interview with Carrie Marcinkevage and Akhil Kumar authors of a research article Generative AI in Higher Education Constituent Relationship Management (CRM): Opportunities, Challenges, and Implementation Strategies is a definite read for you. By reading this interview and downloading the article, you will learn key metrics to evaluate Gen AI usecases, measuring your university readiness and adoption strategies and tactics to implement Gen AI across the university. In the below interview, Akhil Kumar goes by AK and Carrie goes by CHM where I collaborated with them to create this blog interview.

Research Approach

What sparked your interest in researching GenAI for higher ed CRM specifically? Was there a particular problem or moment that catalyzed this article?

AK. I invited Carrie to give a talk on CRM in my Enterprise Information Systems class about two years ago. We met for lunch later where the idea of collaborating on some CRM-related research germinated. Last summer we had several zoom calls to brainstorm our ideas and we hit upon the topic of GenAI in CRM.

CHM. What Akhil said!! I’d add that I’ve been hearing so much about Agentforce that I wanted to dig deeper into agentic AI’s potential as well as explore other kinds of Generative AI types and use cases.

You chose a grounded theory approach with interviews and open-ended responses. Why was that the right fit for this research, and did anything in the data surprise you?

AK. This topic is new and the work is exploratory. There are no real showcases of actual use on campus of GenAI technology that we know of. Salesforce has only recently announced its Agentforce product. Hence, we felt a reasonable starting point would be to go out and talk to folks to see what they feel and are doing in this space.

CHM. Agreed – we’re still at the cusp of this technology’s potential. We’d have loved to study real results, but no one has enough data yet. This research approach allowed us to craft a consistent set of questions and collect data from a good cross-section of the industry: technology suppliers, implementation consulting partners, and universities. We then synthesized themes and crafted frameworks and recommendations grounded in the data. I’m pretty sure I’d have bet on some of the outcomes, such as the importance of a good data foundation. One outcome that surprised me was how many kinds of GenAI there are and where they might be individually applied across the student lifecycle. Here are some examples found in Table 2 of our research:

How did your own backgrounds—academic, tech, or consulting—shape your lens as you analyzed the findings? Were there any biases you had to check along the way?

AK. I am a faculty member with interests in business process management and information systems more broadly. Carrie was the CRM director at the business school and now she is managing CRM strategy for the university.  Her PhD degree was closely related to CRM. In fact, she was the prime driver of this project. With her numerous connections in industry, consulting and academia she was able to find the right folks to interview.  I am sure there were biases, e.g. in selecting the interviewees, in kinds of questions we asked, interpretations of the responses, etc. We did use a tool to analyze the transcripts of the responses, yet biases can still creep in.  We ourselves and most of our subjects are technology enthusiasts. This could have led us to present an unreasonably rosy picture of the prospects of this nascent technology, while neglecting some drawbacks.

CHM. Akhil captured that very well. We did utilize a trusted qualitative research methodology to conduct the research, which helped us identify and mitigate bias as much as possible. The methodology helps ensure reliability and trustworthiness in the process and interpretation of the data. As noted, we’re technology enthusiasts seeking to understand its potential – that’s an automatically positive position. And our sample was pioneers and forward-thinkers in the industry, also a skew toward the positive.

Was there a particular interview quote or participant insight that stuck with you or shifted your thinking while writing the paper?

AK. There were many. But one that sticks with me was an interview with a senior professor at a  university. He brought up issues of scope creep, politics, too many committees, and other thematic issues that helped shape my thinking. 

CHM. One of the participant quotes that stuck with me was about the fine line we must walk – aiding the innovators while bringing along the less enthused:

“You’ve got the innovators. You’ve got the people who want to be the first to change. And then you’ve got the people that are more comfortable and set in their ways, who have been doing this forever and ever and have been hugely successful. So why would they want to change? And you have to somehow figure out how to bring these two groups together, because you don’t want to be too fast and put the university and students at risk. But you don’t want to be too slow and be left behind, because clearly this is moving forward, and this is having an impact on universities, on work, and on society as a whole.”

I appreciated the insight on a wider scale, talking about how difficult it is to do this. I also appreciated the reference to risk – that going too fast could put our students and data at risk, but going too slow means we’re left behind and not preparing our students for the world they are sure to face upon graduation.

If you had to summarize your biggest “aha” moment from the entire research process in one sentence, what would it be?

AK. For me it was realizing that CRM should be viewed through a process lens. After all, many of our interactions with prospective students, current students and alumni are part of a lifecycle and should be seen as such rather than as individual transactions. 

CHM. For me the “aha” started before the research process. It started with my own exploration into GenAI by reading Ethan Mollick’s great GenAI primer Co-Intelligence: Living and Working with AI. His introduction suggests that we each need three sleepless nights to wrestle with the implications of GenAI in our future. He goes on to very rationally outline various risks, benefits, and potential scenarios for GenAI’s societal consequences. In the end, he suggests embracing GenAI as a co-intelligence, a virtual collaborator, colleague, and partner. That got me thinking about all the ways GenAI could benefit higher ed CRM if we give it a chance.

So I’ll use a quote from the book that summarizes my exploration: “Always invite AI to the table.”
Mollick, E. Co-Intelligence: Living and Working with AI; Portfolio: New York, NY, USA, 2024; ISBN 0-593-71671-X.

GenAI Strategy

You’ve highlighted six core types of GenAI in higher ed CRM. Which of these will become the “killer app” over the next two years, and why?

AK. Agents/chatbots. Say, for instance, for interacting with prospects. These agents are dealing with outsiders, and there is low risk of revealing any sensitive information. A next step would be an agent for course advising to help students find the right major, select courses, perform a degree audit, etc. Here there is a higher risk of student information being leaked, so more care is needed. Yet these are piecemeal applications that can standalone and are relatively easy to develop and deploy. 

CHM. The biggest externally facing killer app will surely be agentic AI – virtual assistants with much more capability than we currently have. These agents go far beyond early chatbots with limited responses to recognized user “utterances.” They can complete an entire service activity for someone, from answering a question to completing forms or scheduling. From a service and wayfinding perspective, virtual agents have tremendous potential for high value. They are, however, also high effort. I’d put them in the high impact, high effort “Strategic Projects” box of strategic initiatives in the chart below. Another high potential that’s more internally facing is data summarization – providing an organized, succinct natural language summary of many data points, such as summarizing a prospective or current student’s profile and key needs. I’d put this in the high value, low effort “Quick Value” box. Summarizing a candidate or student profile for admissions committees or student advisers could offer significant time savings as well as better constituent service.

Source: Adapted from standard Impact/Effort Matrix

The article proposes an eight-factor readiness framework. In your experience, which criterion most commonly derails GenAI projects before they start—strategy, data readiness, or something else?
AK. This is uncharted territory and few people understand this technology at any level of depth. Hence, I would recommend starting with small applications where it is easy to illustrate the tangible benefits from the technology in terms of higher productivity, lower costs, feasibility, and ease of implementation leading to a decent ROI.  Large scale applications that require access to enterprise wide data integration or excessive focus on security should be left for later. 

CHM. I’d love to say strategy, as that’s where it all begins. But the single factor we heard again and again, and the one I heard almost every Education Summit session on AI? That was data, data, data. So many institutions have data spread around the university in various pockets and disparate systems. I call it the “It’s a Wonderful Life” syndrome, like the beloved 1946 Jimmy Stewart film (substituting “data” for “money”): “The data’s not here. Your data’s in Joe’s house, right next to yours. And in the Kennedy house, and Mrs. Macklin’s house, and a hundred others.” Institutions should start with projects where they can get the data they need, prove the value, and use it as a case for data collaboration. Otherwise, we could create poor or even harmful outcomes, as one of our participants said, “at a blazing pace in creative new ways.”

CRM GenAI Readiness Evaluation Framework

We’ve been told that the biggest practical contribution from the research is the 8 factor framework for evaluating potential use cases in Table 4, with a scoring example in Figure 4. https://www.mdpi.com/2073-431X/14/3/101

(This can give readers a practical heads-up on where most projects stumble.)

You mention higher ed’s dual nature—progressive thought leaders working alongside institutional inertia. What are your top tips for getting ‘slow adopters’ on board with GenAI initiatives?

AK. The leaders should focus on educating slow adopters on the value of this technology in making their jobs richer and more productive. Incentives should also be employed such as bonuses and prospects of promotion for successful efforts. 

CHM. Slow adopters may be good change management “buffers” – people who bring up potential risks and pitfalls that early adopters should listen to and account for. They can make the outcome better. But true “resistors” are harder to convince, as their resistance could come from job threat fear, technology aversion, comfort zone inertia, or organizational politics. For those cases, I’m a big proponent of success breeding success. Start with those who are ready and willing (that’s one of the eight factors). Utilize those wins to demonstrate value. Before long, even slow adopters may see opportunities for their own benefit. If they don’t, best to first ask if it matters…can we work around and outside them, or is their participation truly necessary? If it is, we can activate champions who’ve achieved success to help motivate those resistors. Hopefully we’ll have enough positive ROI to let the results speak for themselves. Please check the link on leveraging the appreciative inquiry process by Carrie, which will provide you with a framework to increase adoption.

Implementation & Operational Focus

What’s an example of a higher ed GenAI use case where implementation failed—or nearly failed—and what did it teach you?

CHM. I honestly don’t think we have the data for this yet. I’m all ears for schools starting out and experimenting, but there just aren’t enough stories yet. Everyone is still in the consideration or experimentation phase except a few pioneers. Or they’re in the Hamlet phase – too many options and mired in indecision. If anything, the failure is in not taking the leap to try something.

Can you please provide any experiences on a Gen AI project early stages of implementation which can help the audience prepare for a project?

Several universities talked about their perspectives on and entry into GenAI at the Salesforce 2025 Education Summit. One particular session that talked about experience with GenAI and Agentforce was Planning an AI Roadmap: Preparing for Agentforce. There are some very interesting examples across different universities and use cases.

Penn State is developing the system and data foundation for a large initial agentic AI use case for Spring 26, so we should have the capacity for good data collection based on that project.

In the report, you emphasize ‘do data first.’ Could you unpack what a typical “data readiness sprint” might look like for a university exploring GenAI for student services?

CHM. This could turn into a chicken and egg moment: pick a use case first, or pick the best case of data readiness first? Leaning towards use cases first, our matrix would suggest evaluating the use cases that have all the other elements – is it valuable enough, will people readily adopt it, etc. If you have several of those, which one has the best available data? From there, a data readiness sprint would involve starting with the end in mind. Part one is the data assessment. What kinds of activity will this GenAI be doing: decisions, actions, answers, recommendations, summaries? What data elements, individual and aggregate, will it need in order to perform those activities? Where is that data, how much is there, and what’s its quality? You could use the V’s of big data (Volume, Velocity, Variety, Veracity, and Value) to evaluate the data. Part two is data readiness, preparing the data by doing whatever cleaning and collecting is needed. Of course, I’d offer the advice that anyone who says integration is easy is misled or lying, so be wary of starting a GenAI journey with a use case that requires many new integrations. 

Ethics, Trust & Privacy

You caution against algorithmic opacity and hallucinations. How do you suggest institutions validate GenAI outputs—especially when lives or careers (e.g., mental health, advising) are affected?

AK. This is a million dollar question! AI does make mistakes. But then humans make mistakes too, sometimes serious ones. You should test AI thoroughly until its error rate drops below that of humans.

CHM. I do think it’s important to identify that both technology and humans make mistakes. When we assume humans are de facto better than AI, I think that demonstrates a wishful hubris on our part. If humans didn’t make mistakes, we wouldn’t need safeguards and performance reviews and coaching. We’ll need protections and guardrails for AI just like we do for humans. Again, let’s think of it as a co-intelligence, and let’s offer it training, support, reinforcement, and performance management like we would a human counterpart. We’ll need expert design, curated training data, RLHF with high performing humans, extensive testing, and consistent monitoring and reinforcement. There is no case in which GenAI should be left to perform entirely on its own, just like we’d never leave a new employee to perform entirely on their own. As we suggest in the article, don’t just keep the human in the loop; keep the human at the helm.

Given the pressure to innovate fast, how should HEIs balance experimentation with ensuring ethical, FERPA-compliant, and bias-mitigated GenAI models?

AK. Start small. Emphasize continuous monitoring. Build controls into AI. Develop specialized models on top of foundation models tailored to HEIs specific needs. Train AI to evaluate itself. 

CHM. Arizona State University is a good example here. They started with a parking service virtual agent. It was a small use case where they had a limited enough data set with sufficient data quality. And importantly, it was a low-risk use case. It’s much less harmful for an AI agent to make a mistake on a parking fee than tuition.

Penn State is clearly at the forefront of exploring GenAI in higher education. How is this research influencing Penn State’s own CRM strategies or student experience today?

CHM. Like so many, we’re in the early stages of AI exploration in CRM. We are working on an enterprise CRM strategy with multiple implementations and about a million integrations. We say in the paper to “do data first.” So we are focusing effort on the infrastructure to support it. We have a potential major agentic AI project that should start in January 2026 if we can ready ourselves with the data and culture we’ll need to succeed.

For students interested in AI, data science, or education innovation, what makes Penn State’s undergrad or graduate programs stand out? Are any hands-on projects, labs, or courses directly connected to this research?

AK. We are taking many initiatives to prepare students for an AI world. We had an AI Innovation Day at Smeal in March. There was another campus-wide event on AI in May with sessions on how to incorporate AI into our teaching and learning. We are adding a unit on AI in our machine learning course. Also, trying to see if we can dedicate an entire course to AI and at what level. Should it be at the freshmen level for all students? As part of the MIS major? What course will it replace?

CHM. In addition to our work at the business school, Penn State’s College of Information Science and Technology launched a B.S. in Artificial Intelligence Methods and Applications, which is a fantastic jump start into the field. Our World Campus offers stackable credentials including Graduate Certificates in Foundations of Artificial Intelligence, Natural Language Processing, and Computer Vision, plus a Master of Artificial Intelligence program comprising the certificates and additional courses. I’m tempted by this one, but I probably have enough degrees! 

Carrie and Akhil thank everyone involved in their research and invite others to contribute to continued learning and growth in the field. They’d love to hear your experiences.

Please feel free to post your comments below on any questions, and I would be happy to connect you with Carrie and Akhil. You can also email me at buyan@eigenx.com for questions.