Ever wondered what your phone gives away even if no one reads your messages? In 2015, reporter Will Ockenden shared his own mobile metadata including timestamps, towers, calls and texts (no content).
Today I ran through that dataset and, with a few simple charts, pulled out a daily rhythm, likely home/work spots, who he talks to (call vs text), plus odd spikes and trips.
The takeaway: metadata tells a surprisingly rich story—useful, but risky if it’s kept too long or used the wrong way.
What his day looks like
- A strong evening peak (≈ 18:00–20:00), with a gentler late-morning/early-afternoon rise.
- Weekends show a similar shape but shift later and lower, consistent with leisure time.
Where he spends time
- Clear Sydney focus with a strong cluster on the Lower North Shore/Northern Beaches (e.g., BALGOWLAH HAYES ST / FAIRLIGHT), and CBD/South-CBD towers (e.g., CHIPPENDALE / HAYMARKET / REDFERN / SUNDERLAND ST).
- Persistent night/weekend dominance hints at home vicinity; repeated weekday 9–5 dominance hints at work vicinity.
- Time windows, not guesses: “In-office” and “Out-of-office” windows surface likely places probabilistically.
Who he talks to and how
- Channel preference changes over the day: calls rise in work/early-evening windows; texts dominate late evening.
- Different contacts cluster at different hours (e.g., some are work-hour contacts, others are evening-only).
What could an outsider infer?
- Home & work area: Towers most hit overnight vs weekdays 9–5.
- Routine & commute: Typical wake/sleep times, rush hours, lunch, regular routes.
- Social graph (partial): Top contacts and call vs text preference by person.
- Trips/holidays: Runs of activity from rare/out-of-state towers; gaps at usual spots.
How could this be misused? What’s the privacy impact?
- Law enforcement: Helpful for missing persons or corroborating timelines
- Criminals: Stalking, burglary timing (when you’re away), blackmail from sensitive-place inferences
- Privacy impact: Metadata is behavioural measures. Longer retention = stronger inferences, greater breach risk
Self reflection on use of PowerBi
- I built a 24×7 matrix heatmap, a weekday vs weekend hourly line, Top-N bars, and 100% stacked splits.
- I used visual-level Top N, conditional formatting, and simple Count of Id aggregations to keep things lightweight.
- I added date context with a Between slicer and From/To cards, and I controlled Edit interactions so the daily trend filtered other visuals.
- I applied ArcGIS heatmaps for context and split time windows (night vs weekday-day) to infer likely home vs work.
- I set up a page navigator and synced slicers to present the analysis as clear scenes.
- In terms of improvements I could use a clearer visual hierarchy (one hero chart per section) on a consistent grid, with standardized colors/sorting, lighter “ink” (fewer gridlines/labels), and maps kept small for context.
- Also add concise annotations, dynamic date subtitles, and accessible typography/color-blind-safe palettes so insights read faster.