On our final day of refresher week I explored Will Ockenden’s phone metadata and tried to answer a set of practical questions using Power BI:
- Where does Will spend most of his time? Where does he likely live and work, and where might family be based?
- Can we see routines and weekly patterns?
- Who does he contact most, and does he prefer calls or texts with certain people?
- When is he away from usual locations (e.g., travel)?
Approach & Dashboard
I built a Power BI report with four sections:
1) Where he spends time
Slicers let you focus by day of week, cell-tower state, and communication type. A location view shows the towers contacted, with size scaled by the number of interactions in the selected period. This makes it easy to separate weekday (likely work) from weekends (likely home/family).
2) Routines and activity
A heatmap (hour × weekday) reveals when Will is most active. I also added a line graph to display yearly trends. Together, these visuals surface habitual windows of activity and any breaks in routine.
3) Who he contacts
A bar chart ranks the top contacts (by identifier). A second heatmap shows when those interactions happen and a slicer splits call vs SMS, which helps spot preferences (e.g., texts during work hours, calls in the evening).
4) Away from usual locations
Filtering the location view against the activity timeline highlights dates where new states/regions appear (e.g., Tasmania, Victoria) and ties them back to the time series to infer trips.
What an outsider could infer (sensitive insights)
From metadata alone (no message content), an outsider could still infer a lot:
- Home & work areas: regular night/weekend activity vs weekday daytime clusters.
- Routine timetable: typical wake/sleep windows, lunch breaks, commute times.
- Social graph (at least the “top nodes”): who he interacts with most, and when.
- Travel dates and destinations: new towers/states outside the norm; gaps between pings.
- Lifestyle hints: favoured evenings out, weekly events (e.g., sports, classes), quiet days that might signal illness/holidays.
- Device behaviour: heavy data vs call/SMS patterns that suggest app usage rather than person-to-person contact.
These are inferences, but when combined they form a surprisingly detailed insight of a person’s life.
How this data could be misused
- Law enforcement:
- Legitimate uses include timeline reconstruction, placing a device near incidents, and validating alibis.
- Risks include over-confidence in location accuracy (cell sectors can be broad) or function creep beyond the original purpose.
- Marketers:
- Hyper-local targeting (home/work postcode look-alikes), time-of-day ads, audience building based on routines.
- Risk of profiling without meaningful consent, and cross-linking with other datasets (the mosaic effect).
- Criminals / bad actors:
- Stalking or predicting when someone is away from home.
- Social engineering: calling/texting at known active times or impersonating top contacts.
- Targeted theft around travel dates.
Digital data retention & privacy
Keeping granular metadata for long periods magnifies risk:
- Longevity = sensitivity: what looks harmless for a week becomes revealing over months.
- Secondary use: data collected for one purpose tends to be reused for others.
- Breach exposure: the more retained, the more attractive to attackers.
- Fairness & transparency: people rarely understand how much behavioural insight metadata enables.
Good practice: data minimisation, shorter retention windows, aggregation where possible, strict access controls, and meaningful consent.
Limitations - what you can’t learn from metadata alone
- No content or intent: you see that someone interacted, not why or what was said.
- Location is approximate: towers/segments can cover large areas; phones hand off between cells; indoor coverage can mislead.
- Device ≠ person: shared devices, dual-SIMs, or a phone left at home break assumptions.
- Background noise: app pings and system traffic inflate “activity” without human action.
- Sampling gaps: phone off, airplane mode, or out-of-coverage creates blind spots.
- Causality: patterns suggest correlation, not cause. You need corroborating sources for firm conclusions.
Takeaways
- With a few simple visuals, metadata paints a coherent story: likely home/work zones, daily rhythm, key contacts, and trips.
- Those same insights are highly sensitive. Responsible analysis means recognising the limits of accuracy, minimising identifiability, and handling retention with care.