A Google Data Analytics capstone should let another person follow your reasoning from a business question to a defensible recommendation. Choose a question the available data can answer, document cleaning decisions, show the result, and state what remains unknown. A complicated dashboard cannot compensate for an undefined metric or a missing comparison group.
The official capstone course emphasizes case studies and communicating analytical work. In the current Coursera series it follows the Python introduction; the ninth listed course adds AI-assisted job-search support. Follow the instructions in your enrolled version for submission requirements. The project plan below is independent practice, not an official marking rubric.
How do you choose a question that is small enough?
Start with one decision, one population, and one time window. “Analyze bicycle rentals” is a topic. “Should the operator investigate weekend availability at Station A during September?” is a decision. A trip log can show completed rentals and duration; it may not show customers who could not find a bicycle. That difference limits the recommendation before you calculate anything.
Use this fictional brief: a rental operator wants to know whether weekend use at Station A justifies an availability check. You have start and end timestamps, station IDs, ride IDs, and membership category for completed rides. You do not have inventory snapshots, weather, marketing exposure, or records of unsuccessful rental attempts.
Your first deliverable is a question sheet: decision owner, requested action, measure, comparison, source coverage, and missing evidence. Ask whether average rides per day or total monthly rides answers the decision. Weekday and weekend totals use different numbers of days, so comparing totals alone can reverse the apparent pattern.
What should you record about the source?
Record who published the data, its download date, license, covered dates, time zone, row grain, and available fields. Define one row as one completed ride. Check whether a ride ID is unique and whether station codes changed during the period. If records cover multiple months, verify that column names and units remain compatible before combining them.
Keep the original files unchanged. In a public case study, link to the source and give reproduction instructions rather than republishing a dataset whose license does not permit redistribution. Do not treat a publicly reachable file as automatic permission to publish personal data. Prefer a dataset whose documentation makes reuse clear.
How do you document the cleaning?
Create a rules table before removing rows. For the fictional rental file, possible checks include duplicate ride IDs, end time before start time, missing start station, and timestamps outside the analysis window. A suspiciously long ride is a flag for investigation, not automatically an error. A legitimate overnight rental can look extreme without being invalid.
For each rule, record the affected count and reason. Suppose the synthetic export has 1,010 rows: ten are confirmed duplicate exports of existing rides, leaving 1,000 unique rides. If 20 of those lack a start station, retain them in an exceptions file and state that the station comparison uses 980 rides. These counts are an example, not findings from a real operator.
Validate that the filtered dataset still covers the intended dates. A parser that reads day/month dates as month/day dates can silently shift the sample. Spot-check raw and parsed timestamps, and inspect a known end-of-month record. Document whether the weekday label is calculated in local time or UTC.
What does a defensible comparison look like?
Suppose Station A records 440 weekday rides over 22 weekdays and 240 weekend rides over eight weekend days. Weekday use averages 440 / 22 = 20 rides per day. Weekend use averages 240 / 8 = 30 rides per day. Total weekday rides are higher, but weekend rides per day are higher.
The relative difference is (30 − 20) / 20 × 100 = 50%. State it as “In this fictional September sample, completed rides per day were 50% higher on weekends.” Do not say weekend demand rose by 50%: you measured completed rentals, not unmet demand, and you did not compare with a previous period.
Next compare individual days. A single festival weekend might account for much of the difference. Look at the distribution, missing dates, and station closures before presenting an average as typical. If the pattern is unstable, a qualified result is better than a dramatic unsupported conclusion.
Which charts should you include?
Use a bar chart for weekday versus weekend rides per day and a date-level chart to show variation across the month. Label the first chart with rides/day, include a zero baseline for the bars, and show the denominator in a caption. The second chart helps the reader assess whether the average hides exceptional days.
There is no universal requirement for three charts or a 20-slide deck in this independent plan. Include the views needed to support the decision and follow the actual course rubric where it differs. Two well-explained charts can communicate more than six views that repeat the same total. Add a text summary so the conclusion remains available without reading colors.
How far can the recommendation go?
The evidence supports investigating weekend availability at Station A. It does not yet establish that adding bicycles will increase revenue. A useful next step is to collect station inventory or unsuccessful-rental data during comparable weekends, then compare availability and completed rentals under a clearly defined trial.
Write the action with its uncertainty: “Check bicycle availability at Station A on weekends before reallocating inventory. Completed rides per day are higher in this sample, but inventory and unmet demand are unobserved.” This demonstrates judgment while giving the decision owner something concrete to do.
What files make the project reproducible?
- A README with the business question, source, period, tools, main result, and limitations.
- A data dictionary explaining identifiers, units, missing-value codes, and row grain.
- Cleaning SQL, spreadsheet formulas, or a notebook with execution order and assumptions.
- A reconciliation log connecting raw rows to the final analytical population.
- Charts and a short decision memo with accessible captions.
Run the project from a fresh copy of the documented inputs before sharing it. Remove local file paths that only work on your computer and replace them with instructions. If you used an AI assistant, check its calculations and explain the steps yourself. Never submit invented observations or expose confidential data to a public tool.
Can you explain the project without the slides?
Practice a two-minute account: the operator's question, the source limitation, the duplicate rule, the per-day calculation, and the qualified recommendation. Then answer “What evidence would change your mind?” For this example, inventory data showing no weekend shortages would weaken the case for reallocating bicycles even if completed rentals remain higher.
Verified October 14, 2026: capstone purpose and current sequence checked against Coursera. All rental records, counts, decisions, and review criteria above are original illustrative material.
Practice next
Use the Google Data Analytics Certificate overview to place this topic in the current course sequence. Use the cleaning-log example before analysis, then the portfolio publishing checklist to present the finished case study.
The Google Data Analytics study guide PDF includes capstone planning and portfolio templates. The 242-page expanded edition includes current Python labs, 120 multiple-choice questions, 69 applied prompts, and companion practice files. It is an independent supplement to the official coursework. For questions as you practice, explore the separate SimpuTech Google Data Analytics study coach.