Sixty percent of principals told the EdWeek Research Center they use AI for administrative tasks, and NASFAA’s April 2026 report on 1,233 financial aid professionals at 834 institutions found 45% had received no AI training of any kind. Those are two halves of one institution. The administrative side is already drafting with these tools, and almost nothing in the published catalog was built for it.
So the decision in front of a superintendent’s cabinet or a college’s vice president for administration is narrower than the vendor literature suggests. It is not whether to adopt AI, which the business office settled months ago on its own. It is what each function needs to be able to do, and what may never be typed into a chat window at all.
The catalog was built for teachers; the use is in the back office
Look where the sector’s training energy went. RAND’s American School District Panel found 48% of districts had trained teachers on AI use by fall 2024, double the 23% a year earlier, with a sharp split by district wealth: 67% of low-poverty districts against 39% of high-poverty ones.
The published catalog mirrors that priority — lesson planning, learning design, academic integrity, policy writing. None of it is wrong. It is aimed at a different building.
The NASFAA report measures what happens in an administrative office that gets none of it. Among financial aid professionals surveyed in January and February 2026, 54% had used AI for aid work in the prior six months. Of those with any training, 43% described it as self-directed and only 15% as formal institutional training; just 23% said it was specific to their function. Nine percent were aware of any policy governing AI use in their office, and 41% reported no review process for AI-assisted work at all.
That is not reluctance. It is unsupervised adoption, and the 70% who named lack of training and expertise as a challenge want a rule they have not been given.
Which is why one institution needs several curricula, a split we set out by function in the role-based enablement playbook. Issuing seats is not the same as building the skill to use them, which is the whole argument for treating enablement as work rather than procurement.
Two records regimes, not one
Settle the data question before any prompting, and settle it as two questions. Education stacks a second records regime on top of student privacy, and practitioners feel the difference even when the training material does not.
The first is the Family Educational Rights and Privacy Act, which covers education records. The second is federal tax information, which arrives with aid applications and carries its own handling requirements. NASFAA’s respondents raised the second one unprompted, and guidance on where the two lines fall appeared in 44 write-ins as a distinct and urgent need. An office can be scrupulous about student records and still mishandle a tax transcript.
Vendor positions here are checkable and narrower than most buyers assume. Microsoft’s Copilot Chat documentation states that Copilot Chat “supports FERPA compliance for prompts and responses for properly configured implementations.” The next sentence adds that “FERPA compliance does not apply to web search queries as they are not covered by the DPA.” One window, two lanes, different coverage. A staff member who does not know that distinction exists cannot respect it, and a personal login on someone’s credit card sits outside all of it.
Which makes this part of the session a sorting drill rather than a lecture. Staff bring twenty real items from their own week: a transcript request, a verification worksheet, a residency determination, an appeal letter, a board memo. Each goes into one of three piles: into the covered account as-is, in once identifiers are stripped, or never at all. Ten minutes of argument about the edge cases teaches more than an hour of policy reading, and the way wealth managers and advisory firms draw that line transfers almost unchanged.
One test shows whether it stuck: ask three staff separately what may be entered where. Three different answers mean you have a policy document, not a rule.
What each office gets trained on
Administrative work in a school district or college is language work under deadline, which is what these tools are good at. The curriculum differs by office because the review step does.
Admissions, enrollment, and the registrar
- Correspondence from templates the institution already uses. The assistant drafts from the letter library and the record of prior contact; a human reads every word before it sends. The skill being taught is reading a draft fast and catching the one invented sentence.
- Plain-language rewrites of the standing documents. Every institution has the same two dozen notices — deadlines, holds, residency, transfer credit — written at a reading level that generates phone calls. Rewriting them once, under supervision, is the highest-yield hour of the session.
- Inbox and queue triage. Fifty messages ranked by urgency with a one-line reason each, which the office overrides at will.
- Policy reading comprehension. Two catalog policies, one student’s question, a side-by-side of what each requires. The task that gets an audible reaction.
Two things stay off the list. No admission, aid, or academic-standing decision is made or recommended by the tool, and nothing asserts a dollar figure or an eligibility outcome. A statement drafted by a model and sent on institutional letterhead is the institution’s statement.
Financial aid and the business office
This is where compliance exposure concentrates and the training has to be strictest. NASFAA found the use sits in writing: internal emails and memos (68%), brainstorming (51%), proofreading (49%), summarizing regulations or meeting notes (48%), and drafting messages to students (84% among users). Train those tasks properly rather than pretending they are not happening.
- Summarizing guidance, then confirming the citation. Whoever acts on the summary opens the cited source and confirms the language. A fabricated citation reads exactly as confident as a real one.
- Drafting correspondence about a process, never about a determination. Explaining what verification requires is a drafting task. Deciding whether a student passes it is not.
- Reconciliation and worklist triage. Which forty records, grouped by cause, with the common thread named. Staff are good at this and slow at it.
- Narrative sections of reports from numbers that already exist. The assistant arranges the prose; a person ties every number back to the source.
The hard line: professional judgment stays with the professional, and nothing involving federal tax data goes near a tool that is not covered for it. Medical practice administration draws the line in the same place — anything that becomes a number on a submitted document gets checked against the source by a person.
People operations, advancement, and communications
- Job descriptions, posting copy, and interview guides from the institution’s own framework, with a human owning every evaluative statement about a candidate.
- Board and community communications, where the skill is tone control across three audiences from one set of facts.
- Advancement writing — case statements, stewardship letters, grant narratives — where a small team gets the largest return per hour trained, on the same arithmetic we work through for nonprofits with no spare capacity.
What it costs next to what the hours cost
The invoice is the small number. The staff hours are the large one, and almost no training page prices them.
Bureau of Labor Statistics data puts employer compensation cost for office and administrative support work in state and local government at $47.17 per hour worked in June 2026. Twenty staff cost roughly $940 an hour to assemble, before tuition.
| Option | Published price | Hours each | Price plus the hours it burns |
|---|---|---|---|
| Catalog subscription (Coursera for Teams, $399 per user a year, 2 to 499 licenses) | about $7,980 a year for 20 | 10, if anyone finishes | about $17,400 in year one |
| Sector certificate course (SUNY CPD’s Higher Education Professionals track, 6 weeks, 3 to 5 hours a week) | $300 to $400 a course | 18 to 30 | $1,150 to $1,815 to send one person |
| Open-enrollment live workshop (AI for Education, virtual) | $399 for one to two people, catalog to $799 | 3 to 6 | low, useful for one internal owner |
| Facilitated private session on your own documents | $2,000 to $15,000 at market rates | 3 to 4 | the fee plus about $3,800 of staff time |
Two readings. The cheap options are not cheap: a self-paced rollout nobody finishes is the most expensive line per unit of capability gained. NASFAA’s split of 43% self-directed against 15% formal training says most of this learning already happens that way. And a priced sector course is a sound buy for one purpose, getting an internal owner fluent enough to answer questions in week six. It does not change what the other nineteen do on Monday.
Where catalog courses stop
Catalog courses are well made and they stop in the same place: at your documents.
A general course teaches prompting on a business memo. The registrar’s office does not write memos. It writes against a catalog, a student information system, and a letter library, and the output quality depends on which of those three the person thought to supply.
A course cannot supply any of them, so the learning transfers at about half strength and then decays.
The second gap is the sign-off rule: who reads the draft before a student or a trustee sees it. That depends on how the office is staffed, so it cannot be pre-written, and settling it in the room routinely changes the workflow more than anything else.
Credentialing is the honest exception. Association and university continuing education carries credit staff already maintain, and no private session replaces it.
Proving it moved in ninety days
Pick the measures before the session, because afterward everyone has a view to defend. All four live in systems you run.
- Cycle time on one correspondence type. Same category of request, receipt to response out, for ten before and ten after.
- First-pass completeness on one form. Incomplete submissions are where downstream rework begins, and the fix is upstream in how the request is written.
- Response time in one student-facing queue. Median hours to a first substantive reply, the measure families feel.
- Review coverage. The share of AI-assisted outbound work a named person signed off on. NASFAA’s 41%-with-no-review finding makes this the one to start with.
Add the recall test: three staff, asked separately, stating one data rule.
Login counts and satisfaction scores are not on the list. They measure the session. These four measure the institution, and turning a plausible gain into a defensible one is the subject of our guide to proving capability moved.
One caution. Student information system vendors are shipping their own AI features, which automate defined transactions. Staff fluency covers everything between them, which is most of the day. Confusing the two purchases is how an institution ends up owning a workflow product nobody uses well.
What to buy first for a twenty-person administrative team
Four hours, split by office, on your own documents, with a written data rule as the deliverable.
Run enrollment services and the business office apart for ninety minutes each, because the review rules differ and a combined room drifts toward whichever group talks more. Bring everyone together for the last hour to settle the two-regime data rule and name the owner — a director with hours assigned, not an implied volunteer. Send that person to a priced sector course afterward, so week-six questions have somewhere to go.
Then leave it alone for ninety days and read the four measures. If cycle time has not moved and nobody can state the rule, the problem is the sign-off rule rather than the training — an hour’s work to fix.
Frequently asked questions
What should AI training for education admin staff cover?
Three things, in this order: which account and which document category, the four or five drafting tasks that office does daily, and who signs off before anything leaves the building. Everything else, including how the models work, is background reading.
Can school or university staff use AI with student records?
Only inside an institutional account whose terms cover education records, and only with a human reading the output before it is used. A personal consumer login carries no institutional coverage, so pasting an identifiable student’s situation into one is a disclosure.
How much does AI enablement for a school or university administrative team cost?
Facilitated private sessions run roughly $2,000 to $15,000 at market rates depending on length and group size. Sector certificate courses sit near $300 to $400 a course per person, catalog subscriptions about $400 a person a year. Add staff time at about $47.17 an hour of employer cost, usually the larger number.
Should administrative staff and teaching staff be trained together?
No, for the task work. Their documents, review steps, and failure modes have almost nothing in common, and a combined session shortchanges whichever group is smaller. The data rule is the exception: it has to be identical institution-wide or it will not hold.
Is Microsoft Copilot or Google Gemini acceptable for student data?
That depends on which product, tenant, and feature, and the answer comes from the written terms rather than a sales conversation. Microsoft’s documentation shows how specific this gets: Copilot Chat supports FERPA compliance for prompts and responses in properly configured implementations, and that commitment does not extend to web search queries. Verify the terms for the exact product you license.
How is this different from the AI policy work we already started?
Policy says what is allowed. Enablement builds the skill to do the allowed thing well. NASFAA found 9% of staff aware of any policy and 41% with no review process, and policy alone would have closed neither number.
Does this replace administrative staff?
Institutions are not treating it that way: 73% of directors in the NASFAA survey said their institution had not at all considered AI as a substitute for staffing. The realistic effect is less time on first drafts and queue sorting, more on cases that need judgment.
How do we know in ninety days whether it worked?
Time ten items of one correspondence type before and ten after. Watch first-pass completeness on one form and median response time in one student-facing queue. Measure the share of AI-assisted work someone signed off on, then ask three staff separately to state the data rule. If the rule comes back three ways, start there.
Key takeaways
- The published catalog was built for teachers. RAND found 48% of districts had trained teachers on AI by fall 2024; 60% of principals report using it for administrative work.
- Unsupervised adoption is the state of play in administration: NASFAA found 54% of financial aid staff using AI, 45% with no training of any kind, 9% aware of a policy, and 41% with no review process.
- Education carries two records regimes. Student records and federal tax information have different handling requirements, and practitioners raised the second one unprompted because no training material addresses it.
- Vendor commitments are narrower than they sound. Microsoft states Copilot Chat supports FERPA compliance for prompts and responses in properly configured implementations, and that the commitment does not extend to web search queries.
- Price the hours or the comparison is wrong. At $47.17 an hour of employer cost, a 20-hour self-paced course is a $19,000 payroll item across twenty staff before tuition.
- Measure in your own systems: cycle time, first-pass completeness, queue response time, review coverage on AI-assisted work, and three staff stating the same rule.
If you are deciding what to put in front of your enrollment services and business office teams, book a free AI readiness call. We will go through your document categories, both records regimes, and the four measures, and tell you which office to start with.
Arthur Wandzel
