Industries
AI for education
Student-services assistants, course assistants and transcript processing for universities and colleges, grounded in institutional content and connected to the SIS and LMS. Admissions, grading and interventions stay with staff.
The sector
What's different here
Enquiry volume at universities is seasonal and predictable. Admissions, financial aid, registration and IT help peak at the same points every year, most answers sit in the institution's own pages and policies, and the student-specific part, such as a hold or a missing document, sits in the SIS.
The EU AI Act puts admission decisions, evaluation of learning outcomes and exam proctoring in its high-risk category, and it bans emotion recognition in education outright. An assistant that informs students is a very different project from one that scores them, and we keep the two apart.
Funding and trust are the practical limits. Few institutions have new money for AI, and faculty are right to be wary of tools that hand students answers. Course assistants work when they are grounded in the instructor's materials and instructors can see how they are used.
22%
of respondents report an institution-wide AI strategy, although 57% say their institution treats AI as a strategic priority. EDUCAUSE survey of about 800 respondents, Nov 2024.[1]
2%
of respondents in the same study say new funding sources cover their institution's AI costs.[1]
0.73–1.3 SD
learning gain for Harvard physics students using a purpose-built AI tutor compared with an active-learning class, in a randomised trial with 194 students (0.63 SD by OLS). One course, two lessons and a custom-designed tutor, so it says nothing about general chatbots.[2]
Workflows
Where AI does useful work
Each of these runs inside systems the sector already uses. Steps that carry real risk wait for a person.
Student services assistant
Admissions, financial-aid, registration and IT questions answered from institutional content, with student-specific status from the SIS.
- 1Answer general questions from the institution's knowledge base
- 2For signed-in students, read holds and missing documents from the SIS
- 3Send deadline reminders tied to each student's status
- 4Escalate to the right office with the question and context
- 5Log unanswered questions for the content owners
Staff decide anything involving eligibility, exceptions or money. The assistant only informs.
SIS (Ellucian Banner, Workday Student) · CRM (Slate, Salesforce Education Cloud) · SSO/SAML · Website CMS
Course assistant in the LMS
A tutor grounded in the course's own materials that prompts students to reason instead of handing out answers.
- 1The instructor selects the materials the assistant may use
- 2Students open it inside the LMS through LTI
- 3It asks guiding questions instead of giving final answers
- 4Instructors see usage and the most common questions
- 5Assessments are excluded from its sources
The instructor approves the source materials and tutoring settings for each course.
LMS (Canvas, Moodle, Brightspace) · LTI 1.3 · Course materials repository · SSO
Feedback drafting for instructors
Draft comments aligned to the rubric, which the instructor edits before students see them.
- 1The instructor provides the rubric and the submission
- 2A draft comment is written against each rubric criterion
- 3The instructor edits, rewrites or discards it
- 4Grades are entered by the instructor, not the model
The instructor approves every comment and sets every grade.
LMS gradebook · Rubric library · Document storage
Transcript and credential evaluation
Incoming transcripts are read and mapped to existing equivalencies, so evaluators start from a draft.
- 1Ingest transcripts and course descriptions
- 2Extract courses, grades and credits
- 3Match against the institution's equivalency tables
- 4Flag courses with no existing equivalency
- 5An evaluator confirms before credit is posted to the SIS
A credit evaluator approves every transfer credit before it is posted.
SIS (Oracle PeopleSoft Campus Solutions) · Document capture · Equivalency database · Admissions CRM
Accessibility and course content
Alt-text, transcripts, quiz banks and translations drafted for staff to review.
- 1Pull course media and documents from the LMS
- 2Draft alt-text, captions and transcripts
- 3Draft quiz questions and translations from the course materials
- 4Staff review and publish
Course staff review all generated content before students see it.
LMS · Media platform (Kaltura, Panopto) · Accessibility checker
Systems and rules
What the design has to respect
The systems we expect to connect to, and the regulation that decides what an AI system may do with the data in them.
- SIS (Ellucian Banner/Colleague, Workday Student, Oracle PeopleSoft Campus Solutions, Anthology)
- LMS (Canvas, Moodle, Blackboard/Anthology Learn, Brightspace/D2L)
- CRM (Slate, Salesforce Education Cloud)
- Identity (SSO/SAML)
- Website CMS
- Document management
- IT service desk (ServiceNow)
- Proctoring tools
- EU AI Act, Annex III point 3
- AI used for admission or access, evaluating learning outcomes, assessing the appropriate level of education, or proctoring is high-risk. Under the Digital Omnibus these obligations are scheduled to apply from 2 Dec 2027.
- EU AI Act, Art. 5
- Emotion recognition in educational institutions has been prohibited since Feb 2025, except for medical or safety reasons.
- FERPA
- Vendors receiving education records must meet the school-official conditions: a legitimate educational interest, direct control by the institution, and limits on use and re-disclosure. That means a contract or DPA, not consumer AI accounts.
- COPPA and state student-privacy laws
- COPPA covers children under 13, and the amended rule's compliance date was 22 Apr 2026. State laws add their own limits, and GDPR applies to students in the EU.
- Accessibility
- Student-facing assistants must meet WCAG and, for US public institutions, ADA Title II.
Risks
What goes wrong, and how we design for it
- Tutors that hand out answers
- Socratic prompting, assessments excluded from the sources, and usage visible to the instructor.
- Biased admissions or at-risk scoring
- We don't build admission or scoring models as a first project. Early-alert flags go to advisers, who decide on any intervention.
- Student data leaving institutional control
- Models run under an institutional contract or in your own cloud tenant, and SIS access is limited to the fields each workflow needs.
- Wrong financial-aid or deadline information
- Answers cite the current policy page and its date, and questions about an individual's aid go to the financial-aid office.
- Low faculty adoption
- Instructors choose the sources and settings for their course, and nothing is switched on in a course without them.
Where to start
A student-services assistant with SIS lookups
It informs rather than decides, so it stays outside the Annex III admission and assessment categories. It takes on a seasonal load you can measure, and the same retrieval setup is reused later for course assistants. Automated grading and proctoring are poor first projects.
- Index admissions, financial-aid and registration content with an owner for each section
- Sign a DPA that meets the FERPA school-official conditions before any student record is read
- Connect read-only SIS lookups for holds and missing documents after SSO sign-in
- Escalate to the right office with the question and the student's context attached
- Measure enquiries handled, response time at peak and, for admitted students, summer melt
In the sector
Publicly reported examples
What others in the industry have announced. These are not our clients.
Georgia State University, Pounce[3]
A randomised trial covering more than 7,000 admitted students. The vendor reports summer melt fell 21% and enrolment rose 3.3%. Pounce was a rules-based chatbot from 2016, built before large language models.
California State University, ChatGPT Edu[4]
Deployed to more than 460,000 students and over 63,000 faculty and staff across 23 campuses, as announced by OpenAI and the university system.
Use cases
Related use cases
- Customer support automationAn agent answers routine questions from your own policies and account data, takes a small set of allowed actions, and hands everything else to a person with the case already summarised.
- Company knowledge searchStaff ask a question in plain language and get an answer drawn from your own documents, with citations, limited to what each person is already allowed to see.
- Document processingInvoices, forms and contracts are read, checked against your own records and posted to the right system, with low-confidence fields and failed checks sent to a person.
Sources
- 1.EDUCAUSE, 2025 AI Landscape Study (Feb 2025)
- 2.Scientific Reports, AI tutoring outperforms in-class active learning (Jun 2025)
- 3.Campus Technology, Using AI chatbots to freeze summer melt in higher ed (Mar 2017)
- 4.Axios, California State University brings ChatGPT to students and staff (Feb 2025)
Figures are as reported by each source on its publication date. Examples are public industry cases, not AI Sewa clients.
Next step