Start broad, then force specificity
“AI and healthcare” is a domain, not a project. “How accurately do three consumer AI tools identify fabricated citations in a defined set of health questions?” is much closer to a study. A capstone may instead ask how an organization should respond to a bounded operational problem.
Specificity creates a method.
Check feasibility before becoming attached to the question
Do you have the data? Will you need ethics review? Can you recruit enough participants? Does the advisor know the method? Is the software available? Are key records accessible? Can the analysis be completed within the semester?
A fascinating question that depends on unavailable data is not yet a feasible project.
Match ambition to the degree milestone
A master’s thesis is training, not a requirement to solve the field’s biggest problem. A capstone is usually a bounded demonstration of integration and applied reasoning. Leaving a clear next study for later is often a sign of appropriate scope, not failure.
Run a pre-mortem
Imagine the project is behind schedule two months from now. What caused it? Recruitment? Data cleaning? Advisor feedback? Software learning? An overbroad literature review? Build the design to reduce the most likely bottleneck.
Choose a project that produces a useful artifact
A clean analysis, manuscript, policy memo, program evaluation, research protocol, dataset, poster, or technical portfolio can remain valuable beyond the course. The artifact does not need to determine the topic, but it is useful to know what concrete evidence of skill the project will leave behind.
Score the topic on feasibility before becoming attached to the wording
A topic can be intellectually strong and still be a poor thesis or capstone if the required data are inaccessible, recruitment is unrealistic, measurement takes too long, the methods are outside the available expertise, or the question cannot be answered within the program timeline. Feasibility is therefore part of research quality, not an administrative concern added afterward.
One useful test is to write the smallest credible version of the project. Define the population, main exposure or intervention, outcome, data source, and analysis at a level specific enough that another person could identify what would need to happen next. If the project only works when every component is unusually ambitious, it may need to be narrowed.
Compare candidate topics on the same dimensions
| Dimension | Question |
|---|---|
| Importance | Would the answer change understanding, practice, or a later decision? |
| Data | Can the required information actually be obtained? |
| Methods | Are the necessary methods learnable within the available support? |
| Time | Can approvals, collection, analysis, and writing fit the calendar? |
| Scope | Can a clear primary question be answered even if secondary ideas are dropped? |
