A prospective student fills out an RFI form on your website using “Bob Smith.” Two weeks later he calls the admissions office and gives his full name, “Robert J. Smith,” to a work-study student who creates a second lead record. A month after that, he attends a virtual open house under “R. Smith” and a third record appears. Your CRM now has three partial views of one applicant, no single owner, and no shared history. This is not a hypothetical. It is the normal condition of most admissions CRMs, and it is exactly the condition an AI recruiting agent cannot work around.
Salesforce is pushing Agentforce hard into higher education recruiting, and the pitch is straightforward: an AI agent that answers prospective student questions, qualifies leads, and starts applications around the clock, freeing recruiters for the conversations that need a human. Unity Environmental University is Salesforce’s flagship reference for this. According to Salesforce’s own customer story, Unity deployed an Agentforce agent named “Una” starting with the RFI process, the highest-volume and most repetitive point of contact in admissions. Salesforce projects the deployment will save recruiters roughly 17 hours a week and reduce advisor workload by about 25 percent, as part of a broader plan to grow enrollment (Salesforce, “Unity Environmental University Selects Agentforce, Education Cloud”).
That is a real result, achieved by a real institution, and it is worth taking seriously. It is also not a plug-and-play outcome. Unity’s deployment sits on top of a specific data model, a defined RFI workflow, and months of preparation most institutions have not done. Before a VP of IT signs off on turning Agentforce loose in an admissions org, the data underneath it needs to be trustworthy enough for an autonomous agent to act on without a human checking every response.
Adoption is outrunning readiness
Higher ed has moved past the AI-curiosity phase. A joint EDUCAUSE and CUPA-HR survey of higher ed employees, fielded between September and October 2025, found that 94 percent had used AI tools for work in the prior six months, a number EDUCAUSE describes as near-universal. Admissions was named as one of the three functions most likely to benefit from further AI integration, alongside institutional research and advising.
Readiness has not kept pace with that usage. The same body of research points to governance, clarity of policy, and risk mitigation as the areas lagging furthest behind actual adoption. Ellucian’s third annual Higher Education AI Survey, drawing on responses from 779 faculty and administrators across more than 300 institutions between September and November 2025, found data security and privacy to be the leading barrier to AI adoption at both the personal level (61 percent) and the institutional level (56 percent), even as overall AI usage climbed to 90 percent (Ellucian, “3rd Annual Higher Education AI Survey”). People are using AI. Institutions are not yet governing it. An admissions agent that talks to prospective students in real time sits directly in that gap.
There is a broader signal here too. Gartner’s June 2025 research note predicts that more than 40 percent of agentic AI projects will be canceled by the end of 2027, citing escalating costs, unclear business value, and inadequate risk controls as the leading causes (Gartner, “Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027”). Agentforce for admissions is not exempt from that pattern. An agent built on duplicate leads, undocumented eligibility rules, and no defined handoff path is a strong candidate for the cancellation pile, regardless of how capable the underlying model is.
What actually breaks a recruiting agent
None of this is really an AI problem. It is an operations problem that AI makes visible faster and more publicly. A chatbot that gives one prospective student conflicting answers about financial aid deadlines because it pulled from two different lead records does not fail quietly. It fails in front of the applicant, and sometimes in front of their parent, at the exact moment your admissions team is trying to close the deal.
The fix is not a smarter model. It is a healthier org.
A pre-flight checklist before turning on any Agentforce feature
Before enabling an Agentforce recruiting or admissions feature, walk through this list. Each item is something a mid-size institution’s IT team can audit in a few weeks with the right access, not a multi-year transformation project.
Data hygiene and deduplication. Run a duplicate and matching rule audit across Lead, Contact, and Application objects. Standardize picklist values for program, major, and lead source so the agent is not reasoning over five different spellings of the same major. An agent that cannot tell whether it is talking to one prospect or three will confidently give three different answers.
Permission sets and record-level access. Define exactly what the agent can see and touch, separate from what a human recruiter can see and touch. FERPA does not distinguish between a person and a bot reading a student record without authorization, and an agent’s access scope needs the same scrutiny you would apply to a new employee, not a default “give it what it needs” setting.
Documented workflow and eligibility rules. If your program eligibility rules, waitlist logic, and deadline exceptions live in the institutional knowledge of two staff members rather than in the system, the agent has nothing reliable to draw from. Write the rules down, in the CRM, before asking a machine to apply them at scale.
Escalation paths with clear triggers. Decide in advance what the agent should never answer on its own: financial aid appeals, accommodation requests, complaints, anything emotionally charged or ambiguous. Define who receives the handoff, how fast, and what the agent tells the prospect while they wait. An escalation path that exists on paper but was never tested against a real conversation is not an escalation path.
Monitoring and a human review loop. Someone needs to read a sample of agent conversations every week, not every quarter, and have the authority to adjust the rules when the agent gets something wrong. Set up logging and an audit trail from day one, both for compliance and for catching drift before a pattern of bad answers becomes a pattern of lost applicants.
Readiness first, features second
We have seen versions of this pattern before Agentforce existed. An institution buys a capability, turns it on, and finds out six weeks later that the underlying data could not support it. Salesforce and Ellucian both build capable platforms. Neither one can fix a CRM that has never been governed.
This is the work Sanguine Tech Group does with universities running Salesforce and Ellucian Banner: the unglamorous audit of data quality, permission structure, and workflow documentation that has to happen before an AI feature goes live, not after it starts producing complaints. We are not going to tell a client to turn on Agentforce because the sales cycle calls for it. We are going to tell them what their data can support today and what it needs before it can support more.
If your admissions CRM has not had a hygiene and governance review in the last year, and you are fielding pressure to stand up an AI recruiting assistant anyway, that is the conversation worth having before the feature request, not after the first prospective student gets three different answers to the same question.
Sources
- Salesforce, “Unity Environmental University Selects Agentforce, Education Cloud” – https://www.salesforce.com/news/stories/unity-environmental-university-agentforce-education-cloud/
- EDUCAUSE, “New Survey: More than 70% of Higher Education Administrators Have a Favorable View of AI Despite Low Adoption To-Date” (EDUCAUSE/CUPA-HR “Impact of AI on Work in Higher Education” research) – https://www.educause.edu/about/corporate-participation/member-press-releases/new-survey-more-than-70-of-higher-education-administrators-have-a-favorable-view-of-ai
- Ellucian, “Ellucian’s 3rd Annual Higher Education AI Survey Signals Shift from Individual AI Use to Institutional Strategy, Data Privacy Still the Top Barrier” – https://www.ellucian.com/newsroom/ellucians-3rd-annual-higher-education-ai-survey-signals-shift-individual-ai-use
- Gartner, “Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027” – https://www.gartner.com/en/newsroom/press-releases/2025-06-25-gartner-predicts-over-40-percent-of-agentic-ai-projects-will-be-canceled-by-end-of-2027
