Questions about owner records

The method on this site runs three matching passes. Pass one matches exactly on a field that should be unique, such as a phone number stored one way on every record. Pass two standardizes phones, emails, addresses and names, then matches exactly again. Pass three scores near matches with Soundex or edit distance, and a person reviews every uncertain pair before it merges. Each merge keeps the original records, so it can be undone. The duplicates guide sets this beside a Census Bureau list built with software and clerical review.

A random sample of the list in hand measures its own duplicate rate. An invented example: 20 of 200 records drawn at random each have a twin somewhere in the full file, a rate of 10 percent, with a 95 percent range of about 7 to 15 percent. Combining sources pushes the rate up. When the Census Bureau drew six million records from 12 sources for its 1992 Census of Agriculture mailing list, software and clerks together marked 20.9 percent as duplicates, a case the duplicates guide tells in full.

A match only means the provider returned something. A matched record can hold a disconnected number, a relative's number or an old address, and only the buyer's own calls and mail show which contacts reached the owner. Cost per usable contact, the total spend divided by the contacts that reached an owner, puts providers side by side on what reached owners. In an invented two-provider test in the skip tracing guide, the provider with the lower match rate comes out cheaper, at 40 cents per usable contact against 44.

The date each number was last confirmed answers it. A random sample of up to 50 from each age group shows the share that is dead or reaches the wrong person, and the age where that share passes a line set in advance is the re-check age. Numbers change hands at scale: in its December 2018 order the Federal Communications Commission estimated that about 35 million were disconnected and made available for reassignment each year. How fast owner records go stale cites that order and works a full curve.

One format on every record, built on E.164, the International Telecommunication Union's numbering plan: a country code of one to three digits, then the national number, at most 15 digits in all. Written with a leading plus sign and digits only, a United States number reads +14065550147. With one stored format, exact matching finds two copies of the same number. A second field can hold the number as the owner wrote it, for display, and the duplicates guide links the ITU's own text.

Two Postal Service standards cover it. Publication 28 sets the abbreviations for street suffixes such as Street and Avenue and for units such as Apartment and Suite. For a whole list, address software certified under the Postal Service's CASS program assigns ZIP + 4, carrier route and delivery point codes, and the Postal Service tests that software for accuracy before certifying it. The original address stays in its own field, so every change can be seen. Both standards are linked in the duplicates guide.

Waterfall enrichment sends each record to one data provider after another, starting with the one that has the lowest cost per usable contact, and passes only the misses to the next. Returned data is verified between tiers, so a disconnected number counts as a miss. The waterfall stops at the tier where one more usable contact costs more than it is worth to the buyer. Each tier is measured on its own, because it sees only the records the earlier tiers missed; the skip tracing guide numbers the four steps.

Both values stay, and a written rule marks one as current. The order this site uses for owner records: a value a person entered after talking to the owner outranks an appended value; for who owns a property, the public record wins; for how to reach someone, the value with the latest confirmation date wins. When no rule settles it, neither value is marked current and the record goes on the re-check list. A note of which value won, and why, lets anyone reverse the choice later.

A holdout answers it. Before the append, a random share of the records, one in five for example, is set aside and keeps its old numbers. Both groups are called the same way over the same weeks, and the share of calls that reach the owner is counted in each. The difference between the two shares is what the appended numbers added. Without a holdout, the provider's work cannot be told apart from a good calling month. The skip tracing guide sets out how a usable contact is counted.

Licensed full-service providers get 48 months of the Postal Service's change-of-address records, and other licensees get 18 months. The data, called NCOALink, holds about 160 million permanent changes of address. A provider compares a list with it and returns new addresses for people who filed a change. Someone who moved without filing is not in it, so an address the match leaves unchanged is still unconfirmed. The Postal Service's NCOALink page is linked in how fast owner records go stale.

Yes. The Reassigned Numbers Database, set up under the Federal Communications Commission's 2018 order, answers whether a number has been permanently disconnected since a date the caller gives, with yes, no or no data. It answers for dates from January 27, 2021 onward, and access is sold as prepaid subscriptions. A yes means the number no longer belongs to the person it was recorded for. The skip tracing guide links the database's own questions page and records a yes as a dead number, with its date.

A spreadsheet holds up while one person edits the list and the file stays well inside its limits; Google Sheets allows up to 20 million cells (Google Drive Help). A database takes over when several people edit at once, or when the list needs a history of who changed what. Matching outgrows a sheet as well: a table of every pair in 10,000 records would need 49,995,000 rows, and a file of 20 million cells cannot hold them.