Turn an operations problem into an AI work sample you can defend
You have an MBA and an operations job at a business that does not sell AI. You want to explore work involving AI, but the first portfolio question stops you: what would you build that draws on your experience?
Start with a decision you already understand. Perhaps you coordinate store issues, reconcile competing requests or prepare a weekly operating review. Choose one small part of that work and write down what a useful AI-assisted output would have to get right.
The exercise below produces a project brief you can use yourself or bring to someone you are considering paying for career support. Its purpose is to make your next learning need specific: defining a problem, building a demonstration, testing it or explaining the evidence.
Choose a task you can judge
The New York State Department of Labor describes transferable skills as abilities that carry between careers, including planning, communication and problem-solving. Use those as a starting inventory, then attach each to work you can explain. “Planning” becomes “I decide which unresolved store issues need an owner before the weekly review.” Transferable skills and career exploration.
Pick a task where you can distinguish a useful answer from a plausible but wrong one. For this exercise, avoid a project whose quality you can judge only by whether the demonstration looks polished.
Write three lines:
- User: Who receives the output?
- Decision: What will that person decide or do with it?
- Boundary: What must the system leave for a person to resolve?
If those lines remain vague, narrow the task before choosing software.
A fictional example: the weekly issues brief
Priya is an MBA graduate working in operations at a regional retailer. She wants to explore AI-enabled operations work. Her practice project turns invented store issue notes into a brief for a regional manager.
Her first idea is “an AI operations assistant.” That is too broad to test. She narrows it:
Given ten fictional issue notes, produce a table containing the store, issue, stated owner and stated deadline. Flag missing information and contradictions. A manager reviews the table before assigning work.
She creates the notes from scratch. She does not copy employer records, customer details or internal documents. The sample is labeled “independent practice project using fictional data.”
Her role is now clear: she defines the workflow, writes the expected answers, inspects the output and explains what a manager must review. The technical build can be scoped around that brief.
Write the test before polishing the demonstration
NIST’s AI Risk Management Framework connects an AI system’s intended use and business context with evaluation. Its Map function addresses purpose and context; its Measure function addresses testing and documentation. These are useful prompts for scoping a practice project, rather than a credential the project earns. NIST AI RMF Core.
For Priya’s project, three test cases might look like this:
| Fictional input | Expected behavior | What to record after a test |
|---|---|---|
| Store 12 reports a broken scanner; no owner or date appears | Preserve the issue; mark owner and deadline missing | Whether either field was invented |
| Two notes name different owners for the same issue | Flag the conflict for review | Whether both names and the conflict survived |
| A note says an issue was resolved on Tuesday | Keep the resolved status and stated date | Whether it was incorrectly presented as open |
Add ordinary cases as well as difficult ones. Save the inputs and expected answers before running the tool. Keep the original outputs, including mistakes, so another person can follow what happened.
For a first pass, Priya could count whether each required field matches her answer sheet and record every unsupported addition. She could also time the manual version and the assisted version, including review and corrections. A smaller drafting time is only one observation; the review work belongs in the comparison too.
Do not fill the results column with an estimate. Run the exercise, then describe the actual observations and the conditions under which you obtained them.
Make the human decision visible
A table that sends an issue to the wrong owner is a different problem from a table with an awkward heading. Decide which errors require the output to be corrected before anyone uses it.
Priya’s brief includes two rules: missing owners stay missing, and conflicting instructions go to the manager. She also names the person who checks the table against the original notes. Those choices belong in the demonstration alongside the prompt or interface.
NIST separately addresses human oversight in Map 3.5 and documentation of test sets, metrics and tools in Measure 2.1. Use those questions to strengthen your explanation of who checks what and how you tested it. NIST AI RMF Core.
Finish with a one-page evidence brief
Set aside one working session to draft the following. Leave a field blank if you have not done that work yet.
- Problem: One sentence describing the user’s decision.
- Relevant experience: What you personally understand about this workflow, with an example from your own responsibilities.
- Practice materials: What inputs you created and how they are labeled.
- Expected behavior: The output format, essential facts and cases requiring human review.
- Test record: The tool and version used, date, inputs, outputs, corrections and observed results.
- Next gap: The single part you cannot yet do or explain confidently.
Keep your existing experience and your new practice separate in the wording. “Managed regional issue reviews; independently prototyped and tested a fictional-data briefing workflow” makes the two activities legible. Use it only if both accurately describe what you did.
Use the brief to choose your next support
If you cannot define a useful output, seek feedback on the workflow and scope. If you can define it but cannot build it, identify the particular tool or technical step you need to learn. If the demonstration runs but you cannot explain its errors, focus on testing and review.
Bring that specific gap to any career-program conversation. Ask the provider to explain whether its current support covers it, who would review your work and what you would need to do yourself. Our questions to ask before enrolling cover the broader purchase decision; the brief gives you something concrete to discuss.
Review the current MakeTheLeap program with that question in hand. You are choosing help for a defined next step, with a work sample whose scope and evidence you can explain.