Privacy concerns have grown around using cookies to track consumer behavior. This poses a particular problem for firms, like Nielsen, for which tracking consumer behavior is their core business.

But the firm has developed a new, cookie-less strategy for tracking consumers without invading their privacy. Nielsen CDO Mainak Mazumdar talked last week to TVTech about the new method:

“If the industry has learned anything since the rise of cookies, it’s that digital media measurement must remain scalable, flexible and useful,” said Mainak Mazumdar, chief data officer at Nielsen. “Nielsen’s new cookieless measurement approach will further position the company to deliver deduplication across linear and digital as part of Nielsen One. Our new approach to measuring authenticated and unauthenticated digital traffic will enable us to scale across channels and platforms to ensure a comprehensive view of success and uncover areas for optimization.”

In laying out its plans for measuring audiences in a cookless world, Nielsen noted that digital traffic will ultimately move into two distinct categories for measurement–authenticated and unauthenticated–as the deprecation of digital identifiers continues.

To measure authenticated traffic, Nielsen said it will leverage all available identifiers and first-party data from participating clients, such as hashed email addresses, Unified ID 2.0 and select, verified self-reported demographic labels. This will ensure interoperability in the ad ecosystem, including with walled gardens, and simplify measurement for clients by reducing reliance on third parties.

To measure unauthenticated traffic, logged out traffic or traffic on properties that do not have logged in environments or where no alternative identifiers can be provided, Nielsen said it has has developed a machine learning technique with additional contextual data signals including time, browser, content and device information.

Nielsen also explained that the model will be validated against the panel for accuracy. Demographics  of unauthenticated behavior are also modeled and validated with panel observations for both representation and accuracy.