Good growth work explains what changed for the user and how you know, including the limits of your evidence.
Original Role Atlas portfolio exercise
At a glance
- Time to build
- 8-12 focused hours (estimate)
- What you need
- A spreadsheet or SQL, basic funnel analysis, and a simple prototype or slide deck. Use synthetic events if you do not own a product.
What to ship
- An event dictionary and a reproducible funnel analysis
- A before-and-after onboarding prototype
- An experiment plan, guardrail metrics, and an honest result memo
Define value before measuring clicks
Use a fictional AI meeting assistant. Signup is not the same as value: your working activation event could be a user reviewing a useful summary from their first recording. Explain why this event would matter, then list what remains unvalidated.
This is an original Role Atlas portfolio brief. You can complete it without recruiting users, buying analytics tools, or uploading real meeting recordings.
Find a specific bottleneck
Create or obtain permitted event data for signup, sample meeting opened, summary generated, summary reviewed, and return visit. Define the unit of analysis and a time window, then deduplicate users before computing conversion rates.
For a synthetic dataset, document exactly how you generated it. Do not interpret an invented drop-off as a discovered fact about real customers.
- Segment by entry path and first-session behavior; avoid tiny groups that invite overinterpretation.
- Choose one intervention, such as a sample meeting before an upload request.
- Prototype the changed screen and explain the hypothesis it tests.
- Specify a primary metric, such as seven-day activation, and guardrails such as errors, cost, or abandoned uploads.
Write the decision before the victory lap
If you control a real product and have permission, randomize eligible users and plan duration and sample size before reading results. If you only have a prototype, run a small usability exercise and report observations instead of claiming causal lift.
A strong final memo can say "not enough evidence yet." Include the analysis, assumptions, and the next test you would run. Keep simulated results and measured results visibly separate.
How to describe it on your resume
Analyzed a [synthetic/real] onboarding funnel for an AI product, identified [drop-off], and designed an activation experiment with [primary metric] and [guardrail].
Use only after completing the work. Replace brackets with your own evidence. Label independent, prototype, and synthetic-data work accurately; do not invent customers or business impact.