This is Lesson 5 of The Student Accounts AI Field Guide Series, a companion learning series alongside our 2026 State of AI in Student Accounts report.
Through the first four lessons in our The Student Accounts AI Field Guide Series, you received a framework for writing prompts, suggestions for how to think through FERPA issues, and a playbook for bringing AI work into the open at your office. But there's one moment that still stops a lot of people, including experienced users: the blank chat box.
You have a task in front of you that you are pretty sure AI could help with. You open your favorite AI tool. Then you freeze, unsure of where to start. What do you type? How much background does it need to help you with your work? Does it have prior context to help with the task? Do you describe the report or other materials you want reviewed, or do you paste the entire report or documents? This hesitation stalls many good intentions. It is also where the myth of the "perfect prompt" does the most damage. If you believe you have to get the instructions exactly right before you begin, the easiest thing to do is not begin at all.
Here's a simpler way in: start with the outcome, not the prompt.
Flip the conversation: say what you need, then ask what it needs
Tell the tool what you're trying to accomplish, then ask it what it needs from you.
For example:
The AI can ask for the missing context. You can then answer one question at a time, rather than trying to write a perfect set of instructions up front. This isn't a replacement for the prompting framework in Lesson 1. However, it is a way to build that same prompt together, in conversation with AI. You are the expert on the problem: you know about the content you are asking for help with, where the data comes from, and what "matching" should look like. AI is good at recognizing what information it needs to structure the work. By working this way, you use your skills while letting the AI do what it does best.
The questions AI asks back are often useful on their own for achieving your goals. Examples are “For which reporting period? Which fields should match? Should timing differences between posting dates count as a discrepancy?” Answering these questions often helps you reach the result you are seeking. As you think through the answers, you can get more precise about what you know, what you want to know, and what data you need to provide to get to those answers.
The three-minute moment
This approach came up at our recent session, AI in the Bursar's Office: What Is Actually Working, which Meadow presented at the annual CoHEsion conference in partnership with John Noah of Harding University.
John shared an example from his office in which his colleague spent about three hours tracking down a bank-reconciliation discrepancy. To show his colleague the power of AI, John gave the same two files to an AI and explained in his prompt what they were looking for in the bank records. The AI identified the discrepancy in about three minutes. This experience sparked his colleague's curiosity about where else AI might help in their work.
This example demonstrates several important points. You don’t need a perfect prompt for AI to help you in your work. You simply need a clear goal, solid data inputs, and enough background knowledge to judge whether AI’s responses are correct.
Using pilot tests like this can help spark curiosity among your colleagues, as we discussed in Lesson 4. Sometimes the most effective way to get buy-in for using AI among colleagues is to show them simple examples like the one we just discussed. This lets them experience the time savings and potential automations for other tasks.
Choose a problem you can check
Not every task will produce a result that is easy to verify. But John's example illustrates a useful first step: choose a problem you understand well enough to check the answer, even if doing the work manually takes a long time.
This combination is the sweet spot. If a task is quick to do by hand, AI won't save you much time. If it is too complicated and you can't tell whether the answer is right, you can't trust it. You should consider these examples of use cases as a strong starting point:
- Comparing reports. Spotting discrepancies between two versions of a report or lining up figures from one term to the next. AI flags what changed; you decide what matters.
- Drafting a sensitive student email. Getting a strong first draft of a hard message about a canceled payment plan, a registration hold, or a balance sent to collections. You bring the policy and the tone; AI gets you there faster.
- Documenting a process. Turning rough notes on a month-end task into step-by-step instructions a cross-trained colleague could follow. You know the steps; AI organizes them.
- Summarizing guidance. Pulling out the parts of a long federal or institutional policy document that actually affect your office. AI narrows it down; you confirm what applies.
In each case, you already have the expertise to judge the output. AI is accelerating work you know how to do, not doing work you can't evaluate.
Try it with one task.
Pick one recurring task from the list above, or one of your own. Then:
- Describe the outcome. "I want to identify what doesn't match between these reports so I can have an accurate reconciliation and investigate any discrepancies before the end of the term." Explain what done looks like, not how to get there.
- Ask what's missing. "What do you need from me to do that?" Answer the questions one at a time. If a question doesn't make sense for your situation, explain why to help get you to the answer you are seeking.
- Provide only appropriate information. Use an approved tool and follow your institution's rules for student data. Bank files, ledgers, and student account exports can contain exactly the kind of information Lesson 3 asked you to think carefully about. When in doubt, remove or mask identifiers first, or describe the structure of the data instead of sharing the data itself.
- Check the result. Treat the output as a starting point, but not a final answer. If AI says the discrepancy is a duplicated deposit on the 14th, open the source file and confirm the output.
What it sounds like in practice
The same approach works for writing tasks, not just analysis. If you need to contact a student whose payment plan was canceled after missed payments, you might open with:
"I need to write an email to a student whose payment plan was canceled for missed payments. I want it to be accurate and clear about what happens next but not discouraging. What do you need from me to draft it?"
A good response will ask for things like the current balance and due date, whether the plan can be reinstated and how, what holds are now in place, who the student should contact, and what they've already been told. Then you answer each of these questions. The result is a draft built on your actual policy rather than generic language, and you didn't have to anticipate every detail before you started.
Notice that none of those questions require the student's name or ID. You can draft the template with placeholders and fill in the specifics in your own system if you choose.
Write down the number.
If it helps, note how long the task usually takes and how long it took with AI, including the time you spent checking the output. That gives you something concrete to share with colleagues, and a better basis for deciding whether the approach is worth repeating.
It also connects directly to Lesson 4. When you're ready to bring AI into a conversation with your manager or compliance partner, "three hours down to three minutes, and here's how I verified it" is a far stronger opening than "I've been trying AI." One specific, checked result is worth more than general enthusiasm.
It may help to keep a simple log if you can: the task, the tool, the time before and after, and whether the result held up. A few weeks of entries become the kind of evidence that moves a decision.
One honest note on checking the work
AI can be wrong, even when it sounds confident (that’s one of the reasons we often say it is like an eager intern). It may point to a discrepancy that isn't there, miss one that is, or write an email that sounds right but misstates a policy detail. It is essential to check every AI output.
Not every attempt will work. Sometimes the files are too messy, the question is too vague, or the tool simply isn't suited to the task. That's not a reason to stop. As the ASU example in Lesson 4 put it, most experiments fail; the move is to change the variable and try again. Rephrase the outcome, give it a cleaner file, or try a different task.
The takeaway
Your first breakthrough may not come from mastering prompts. It may come from bringing one familiar problem to AI and letting it help you work backward. A perfect prompt is not the goal, a good answer is. And sometimes you need to work through that together alongside the tool you are using.
This is Lesson 5 of The Student Accounts AI Field Guide Series, a companion learning series alongside Meadow's State of AI in Student Accounts report. Each lesson covers a practical topic grounded in data from the field.
Meadow's research for this series draws on The State of AI in Student Accounts, based on a March 2026 survey of 147 professionals across student accounts, billing, finance, collections, and One Stop/shared services.
