Understanding Artificial Intelligence in Your Role
Artificial intelligence is changing how work gets done across the University. This guide explains what AI is, the language used to describe it, and where it can responsibly support the work of business and financial management — written for administrators, not engineers.
Introduction to Artificial Intelligence
What is AI?
Artificial intelligence refers to computer systems that perform tasks we normally associate with human reasoning — recognizing patterns, understanding language, making predictions, and generating text or images. Most of the tools generating attention today are a specific kind of AI called machine learning, in which a system learns patterns from large amounts of data rather than following rules a programmer wrote by hand.
The best-known examples are tools built on large language models, such as ChatGPT, Microsoft Copilot, and Google Gemini. These are a form of generative AI: given a written instruction, or prompt, they produce fluent, original responses.
Each field sits inside the one around it — the tools you hear about most live at the center. Large language models (LLMs) sit within deep learning and power the generative AI at the core. Click any ring for a definition.
AI
A useful distinction. Today's tools are narrow AI — very capable at specific tasks like drafting or summarizing, but without genuine understanding or judgment. The "general" AI of science fiction, able to reason across any problem, does not exist. Treat these systems as fast, tireless assistants that still require your oversight.
One important limitation: because these systems generate the most statistically likely response, they can state incorrect information with complete confidence. This is known as a hallucination, and it is the single most important reason to verify anything you rely on.
Why it matters now
AI is no longer a research curiosity. It is embedded in the software your teams already use — email, spreadsheets, procurement platforms, and financial systems — and staff are experimenting with public tools whether or not the institution has issued guidance. For a business manager, that creates three responsibilities you cannot delegate:
- →Judgment. Knowing where AI genuinely helps versus where it introduces risk to accuracy, compliance, or trust.
- →Data stewardship. Ensuring confidential and student data never flows into tools that are not approved for it.
- →Fairness. Understanding that models learn from historical data and can reproduce bias, so consequential decisions keep a person accountable.
You do not need technical depth to meet these responsibilities. You need a working vocabulary and a clear sense of appropriate use — which is what the rest of this guide provides.
Core Concepts
This section explains the big ideas behind AI, in the order that makes them easiest to follow. Each one has its own page. If you just need a quick definition of a single word, use the Glossary instead.
Start here: seven ideas, in order
Types of large language model
See the dedicated page on what an LLM is and how it works →
Not all LLMs are the same, and the differences matter for procurement, data handling, and cost. Two distinctions come up most often. Vendors will use these terms; here is what they mean for you.
Owned by a company and accessed as a paid service over the internet. You send a prompt, their servers run the model, and you get a result. The inner workings are not public.
Examples: OpenAI GPT, Anthropic Claude, Google Gemini. Trade-off: most capable and easiest to adopt, but your data leaves campus and you depend on the vendor's terms and pricing.
The trained model itself (its "weights") is published for anyone to download and run — including on hardware the University controls, so data need never leave campus.
Examples: Meta Llama, Mistral, Google Gemma. Trade-off: strongest data control and no per-use fee, but you (or IT) must host, secure, and maintain them.
A common point of confusion: "open-weight" means you can download and run the model — it does not necessarily mean the training data or full source is public, and it is not the same as free-to-use consumer apps. "Open-source" is a stricter term implying the code and often the data are shared too; many "open" models are open-weight only.
Runs on the vendor's servers. Nothing to install, always the latest version — but prompts travel off-campus, so it must be an approved service for any sensitive data.
Runs on University-controlled hardware. Data stays in-house and usage is private — the option often preferred for confidential or regulated information, at the cost of setup and upkeep.
You will rarely choose a model yourself, but knowing these categories lets you ask the right question of any vendor: "Is this open-weight or proprietary, and does our data stay on campus or leave it?" The answer drives both the compliance review and the cost.
AI in Finance & Budgeting
Where it helps
Financial work involves large volumes of structured data, repetitive processing, and routine written reporting — exactly the areas where AI is most useful. The highest-value applications for a campus business office include:
Project revenue, enrollment, or spending trends and test budget scenarios far faster than manual modeling.
Flag unusual transactions, duplicate payments, or expense-policy exceptions for human review.
Read invoices, extract line items, and match them to purchase orders, reducing manual data entry.
Turn figures into a first-draft variance narrative or board summary for you to review and correct.
Summarize spending across vendors or surface key terms and dates from long contract documents.
Assistants trained on your policies can answer common staff questions about travel, reimbursement, or coding.
Cautions
Finance is a high-stakes, highly regulated domain. The same qualities that make AI fast also make unexamined use risky.
- Verify every number. AI can misread a figure or invent one. Treat all output as a draft requiring reconciliation to source systems.
- Keep a human accountable. AI can inform decisions on budgets, awards, or personnel — it should never make them. Ownership stays with a person.
- Mind the data. Do not paste ledgers, salary data, or vendor contracts into public tools. Use only platforms the University has approved and contracted for institutional data.
- Retain your records. AI-assisted work is still subject to audit, public-records, and retention requirements. Document how a result was produced.
Providing an AI service: the financial view
The sections above concern using AI. But a business manager is often on the other side of the transaction — the person who has to fund, budget for, and account for an AI capability offered to a department or the wider campus. That is a different discipline. AI costs behave unlike traditional software, and understanding the dimensions below is what separates a sustainable service from a budget surprise.
1. Know which cost model you are buying into
Each delivery model prices differently, and the model dictates how your budget behaves — fixed and predictable, or variable and usage-driven.
A fixed fee per licensed user. Predictable to budget, but you pay for every assigned seat whether or not it is used — so the risk is shelfware. Match licenses to actual demand and reclaim idle ones.
You pay for what you process, usually per unit of text (“tokens”) and by model. Efficient at low volume, but costs scale with use and can be hard to predict. Requires usage monitoring and, often, chargeback. See prompts, tokens & cost for how this is metered.
Up-front investment in GPU hardware plus ongoing power, hosting, and staff. No per-use fee, but a large fixed cost and a refresh cycle. Economical only at sustained, predictable volume.
2. Budget for total cost of ownership, not the license
The sticker price is rarely the real cost. A defensible budget accounts for the whole lifecycle:
Integration, configuration, data preparation, and any custom bot or knowledge-base build.
Admin, support, and training time — often the largest and most overlooked line.
Security review, risk assessment, and ongoing monitoring — a real cost even when unbudgeted.
Price increases, model deprecations, and the cost of migrating off a tool later (lock-in).
3. Fund it fairly and control the spend
- Decide the funding model. Central subsidy, per-unit chargeback, or a hybrid — each shapes adoption and equity. Consumption pricing usually needs chargeback to stay sustainable.
- Set guardrails on variable cost. Usage caps, budget alerts, and steering routine work to cheaper models keep a token-based bill from surprising you.
- Right-size continuously. Review seat and usage data on a schedule; reclaim idle licenses and retire pilots that did not land.
- Pilot before you scale. A small, measured pilot reveals real per-user cost and value before a campus-wide commitment.
4. Tie cost to value
Ultimately the question is not “what does it cost?” but “what do we get for it?” Define the outcome you expect — hours saved, faster close, fewer errors, better service — and measure against it. An inexpensive tool no one uses is pure waste; a costlier one that removes a bottleneck can pay for itself. Build that comparison into the budget request, and revisit it at renewal.
Practical Use Cases for Managers
Everyday tasks AI does well
You do not need a special project to benefit. These are low-risk, high-value tasks where a general assistant saves time today — provided no sensitive data is involved.
Guardrails
A simple set of rules keeps AI use safe and defensible. Share these with your team.
- Use University-approved tools for any work data.
- Treat output as a draft and verify facts and figures.
- Keep a person accountable for every decision.
- Disclose AI assistance where transparency is expected.
- Enter student records, SSNs, or PII into public tools.
- Paste confidential financial or HR data into a chatbot.
- Let AI make final hiring, budget, or award decisions.
- Assume output is accurate, current, or unbiased.
When in doubt, ask a simple question: would I be comfortable if this input or output appeared in a public-records request or an audit? If not, stop and consult your IT or information-security office before proceeding.
Frequently Asked Questions
AI is a capable assistant, not an authority. It drafts, summarizes, and suggests at speed — but a person keeps judgment, accountability, and stewardship of University data. Use it to do your work faster, never to decide your work for you.
AI Services
The main kinds of AI tools institutions provide, with common examples from several vendors. Your institution's choices will differ. Each has its own page — select one to learn what it is, when to use it, and how to get access.
Start with your approved tools listOpen page →ChatGPT, Claude, Gemini, Copilot. The enterprise edition matters more than the brand.
Open page →AI inside Google Workspace or Microsoft 365, where the work already happens.
Open page →Q&A over your own documents, with citations.
Open page →Summaries and action items in Zoom, Teams, or Meet.
Open page →New AI features inside finance, HR, procurement, and service platforms.
Open page →A secure, university-run, multi-model home for generative AI.
Open page →Open-weight models run on institutional hardware, so data stays in-house.
Open page →References
Frameworks and resources for AI guidance, policy, and support. Check your own institution's resources for current rules and for questions specific to your unit.
External links open in a new tab. Where this guide and your institution's official guidance differ, follow your institution's guidance.