Data & identity
Probabilistic matching
Probabilistic matching links devices or identifiers to the same person or household using statistical inference from signals like shared IP address, location, timing and device traits, with some uncertainty.
When there is no login to prove a link, probabilistic matching makes an educated guess. If a phone and a laptop are both regularly on the same home IP address, in the same place, at complementary times, with similar browsing patterns, a model might conclude they probably belong to the same household or person, with a confidence score.
This approach greatly extends the reach of an identity graph beyond logged-in users, and it is used for cross-device targeting, CTV household matching and measurement. It relies on signals like IP address, user agent, location and behaviour.
The drawbacks are accuracy and privacy. Shared networks (offices, cafés, mobile carriers using shared IPs) create false links, and probabilistic links built from device signals can resemble device fingerprinting, which browser makers and Apple's App Store rules restrict. Regulators treat such inferred links as personal data. Buyers should ask vendors for accuracy metrics validated against deterministic truth sets. Compare deterministic matching.
Think of it like this
Probabilistic matching is like guessing two people are siblings because they share a surname, an address and a laugh: probably right, but not guaranteed.
An example
A CTV vendor links a smart TV and two phones in a Madrid home because they share an IP for 90% of evening hours; its model gives the link 85% confidence.
Related terms
Deterministic matching
Deterministic matching links identifiers to the same person using definite, shared information, such as the same login email or phone number, rather than statistical inference.
Identity graph
An identity graph is a database that links the many identifiers belonging to the same person or household, such as emails, device IDs, cookies and IP addresses, into one profile.
IP address (in advertising)
In advertising, an IP address is the network address a device uses online, used for rough geolocation, household matching, frequency capping and fraud detection, and treated as personal data in many laws.
Device fingerprinting
Device fingerprinting is identifying or characterising a device from the combination of its technical attributes, such as browser, fonts, screen, graphics and network traits, instead of a stored cookie or ID.
Household ID
A household ID is an identifier representing a home and the devices in it, rather than one person, used mainly in CTV and addressable TV for targeting, frequency capping and measurement.
Sources: W3C: Mitigating Browser Fingerprinting in Web Specifications, Apple: User privacy and data use