# From topic to testable project

By Ackley Will | https://ackleywill.com

## Start with the decision

A broad topic is a useful beginning. A feasible project needs a more precise question, observable variables, accessible evidence, and a way to evaluate the answer.

## Six prompts

1. **Question:** What do you want to know, in one sentence?
2. **Evidence:** What data or literature can you actually access?
3. **Unit:** What does one observation represent?
4. **Method:** What analysis can that evidence support?
5. **Comparison:** What baseline or alternative makes the result informative?
6. **Boundary:** What conclusion would go beyond the evidence?

## Check the fit

A dataset containing employee characteristics and turnover labels, for example, may support a predictive modeling question. By itself, it does not show that adopting AI caused a change in employee retention.

The same distinction applies widely: building a prediction tool and estimating the effect of using that tool are different research tasks.

Download the canvas and write a first version before choosing a more complicated model.

---
Resource text: CC BY 4.0. Attribution: Ackley Will, ackleywill.com.