Admin & Operations
Using AI for Administrative and Operational Work
Guidance for employees on four common places where AI shows up: drafting communications, working with sensitive data, automating routine processes, and procuring new tools.
Getting Oriented
The use of AI in administrative or operational settings is shaped by several sources of information and guidance, including Virginia Tech’s Responsible and Ethical AI Principles, the university data Risk Classification Standard, the use of approved tools to protect privacy, and IT Governance.
Responsible Use
In day-to-day work, the role of AI should be considered in the places where it most commonly shows up: when drafting communications, working with sensitive data, automating routine processes, and procuring new tools. Each recommendation below is followed by the related principle for Responsible and Ethical AI in parentheses.
Drafting communications and content
AI is often used to assist in writing emails, memos, web copy, event materials, and talking points along with creating images and video assets.
- As the author, you remain responsible for the accuracy, voice, and any commitments made in the text. (Human Judgment)
- AI often gets names, titles, and dates wrong. Read and verify AI-generated content before sending.
- Edit AI-generated content to be in your own voice.
- Consider when it is important for your communications to have warmth and authenticity. (Human-Centered Benefit)
- Sensitive messages, sympathy, conflict, tough news, should not be AI-generated.
- Form letters and routine confirmations are exactly where AI shines.
- Remember any work-related content you produce is considered official Virginia Tech communication. (Fairness & Transparency)
- Don’t have AI generate quotes attributed to people, statistics about programs, or policy interpretations without verification.
- Confirm all communications follow university style.
Working with sensitive or regulated data
Protected data, e.g., student records, employment files, financial information, and health data, falls under federal regulations that consumer AI tools are not built to handle.
- Match the tool to the data class. (Data Security & Privacy)
- Use only tools approved for the relevant data classification on the Virginia Tech approved tool list.
- Be sure to sign in to approved tools with your Virginia Tech credentials.
- Do not use free-tier consumer apps or personal paid accounts as they don’t carry institutional contracts and related data protections.
- Always anonymize or aggregate information, if possible.
- Consider whether using AI to assess or analyze sensitive data is appropriate. (Responsible & Ethical Use and Human Judgment)
- Decisions affecting individuals, e.g., admissions, personnel, financial aid, accommodations, and conduct cases, require named human judgment. AI can inform but should not be used to make decisions.
- Always consult existing university policies if you are unsure which decisions require human review.
- Document the human reviewer for any AI-assisted decision about an individual.
- Any AI-assisted decisions affecting individuals should be appropriately documented and disclosed. (Fairness & Transparency)
- Keep prompts and outputs alongside the human reviewer's notes.
- Coordinate with HR or General Counsel before using AI in any consequential decision.
Automating routine processes
AI can handle routine tasks like sorting tickets, drafting form responses, summarizing reports, and formatting data. When AI moves from a drafting tool to an autonomous workflow, additional governance applies.
- Automation is a real win when it gives you back time to spend on judgment-heavy work. Identify the repetitive tasks that would benefit most. (Innovation for Good)
- Identify tasks that are high volume and low judgment.
- Pilot with a small batch and a real human in the loop before scaling.
- An automated process still produces decisions you own. Build in checkpoints, audit trails, and a way to override. (Human Judgment)
- Sample and verify AI-generated output regularly.
- Provide a human’s contact information to people receiving AI-generated output.
- If an individual receives an AI-generated reply, they should be able to reach a human to ask questions. (Fairness & Transparency)
- Disclose when AI is used.
- Provide a human’s contact information to people receiving AI-generated output.
- Automation should reduce friction and provide efficiency for everyone involved. (Human-Centered Benefit)
- Test your automated process with individuals who will use it, including people with disabilities and limited English proficiency.
- Be prepared to respond to complaints and make adjustments.
Procuring or piloting AI tools
Departmental AI tools often start as informal pilots on free tiers. Consult with IT procurement to approve a tool before launching workflows that depend on new tools.
- Free tools usually mean your data trains the model. (Data Security & Privacy)
- Read and understand the data-handling terms before beginning a pilot.
- Request Low-Risk Low-Cost Software approval.
- If your pilot workflow will require access to Moderate- or High-Risk data, you must consult with IT procurement for approval.
- An informal pilot creates expectations and dependencies. (Fairness & Transparency)
- Define and disclose the pilot’s scope, duration, and exit criteria up front.
- Ensure stakeholders understand when the pilot workflow decision will be made.
- Ensure piloted tools are accessible to people with disabilities. (Responsible & Ethical Use)
- Confirm WCAG-AA compliance and screen-reader compatibility before pilot adoption.
- Keep a non-AI workflow path available.
- Before deciding on a tool, test it against your real edge cases. (Human Judgment)
- Identify three of your hardest tickets, letters, or data sets to test using a new tool.
- Only use Low-Risk data for testing until the tool receives a higher level of approval.
When to stop and get help
Some situations require you to consult your supervisor, IT procurement, or the IT Security Office before you proceed.
- You’re about to paste high-risk data into an AI tool, e.g., names paired with grades, social security numbers, health information, financial-aid records, or human resources files.
- Visit the Data Risk and Privacy web page.
- AI-generated content will affect an individual, e.g., admissions, personnel, financial aid, accommodations, and conduct cases.
- AI can inform but should not be used to make decisions.
- For decisions about people, you may need to consult your supervisor, Human Resources, or University Legal Counsel.
- A vendor pitches a “free” AI tool that would access protected data.
- You should work with VT Procurement and potentially IT Governance before you proceed.
- An automation you built starts producing incorrect or unverifiable outputs.
- Pause the automation and investigate.
- Alert stakeholders immediately about the scope and timeline of the failure.
- An individual objects to an AI-generated communication.
- Follow up in person and gather feedback. Carefully consider if the individual’s concerns should change your workflow.
- A vendor changes terms or pricing mid-pilot.
- Pause the pilot until data privacy terms can be reviewed.
- Consult with VT Procurement or University Legal Counsel.
Virginia Tech Resources
- 4Help — receive help using AI tools.
- AI Working Committee (a standing committee within IT Governance) — contact for policy questions and advice.
- TLOS AI Professional Development — self-paced courses and live workshops for instructors.
- University Libraries AI Resources, Tools, and More — guides for citing AI, evaluating AI-generated sources, and using AI in research.