A mobile provider enters into marketing sharing agreements with credit card companies. It extracts housing information from local property and tax records. It enters into marketing sharing agreements with retailers, payment processors like ADP. Same with license plate reading companies, loan companies, banks, professional organizations, etc.

It fills its data lakes with the vectorization and down tilt data that it collects every day. It uses federated batched Hadoop tasks to join the above data lakes into one large data lake. Mid-PB in size.

Then it looks for mobile phones that travel to the 400 block at night and stay there, that are buying dentist stuff from Walmart, travel to a dentist office every workday, have an income over $120k, and are a member of the local dentist society. Maybe look for someone with dentist student loans, graduated with a dental degree.

None of those data points can identify an individual. Taken together they can ID just about anybody.

But maybe there is a chance that you ID their wife/husband. So maybe include/exclude people that regularly visit OBGYN offices.

Back in the day we could link cell numbers to credit card purchases in locations to the point of being to identify the name of the person and what they purchased and where it was purchased. For all people in a metro area that were using credit cards and physically visiting stores.