Tech viewpoint on predictive modelling
A view from Nimeshh Patel

Tech viewpoint on predictive modelling

Tracking users across devices is complex and, as tech giants such as Apple bring out new devices and Amazon reportedly makes its way into the market of the Internet of Things, the challenge is only set to increase.

A myth exists in the industry that advertisers or ad technology providers need to be 100 per cent certain of a user’s identity, even anonymously, to be able to accurately track, target and attribute user conversations across devices. Unified logins – such as a social network or e-mail client – are one way to access this. However, Google and Facebook have seen recent backlash from consumers, and now the ad-free and hugely popular Ello has arrived. This highlights concerns around the use of single-login environments for ad-targeting.

However, there are now advanced predictive models that can be used to create a portrait of a consumer using device and browser data that comes with much lower privacy concerns. These techniques, which match up various devices to a certain user, can be up to 85 per cent accurate at identifying users without the need for any personal information. This levels the playing field when it comes to brands successfully identifying audiences across devices – something that could previously only be done effectively by the Googles and Facebooks of our industry.

Probabilistic models can show a user's path to purchase with high accuracy, no privacy concerns

It is often thought that predictive modelling technology is an unscaleable method of tracking or targeting users. The fact is that several companies have now developed machine-learning models that gather information from browsers, apps and devices and can connect the dots between millions of consumers.

This means that it is now possible to understand the journey if a user receives an ad on Facebook for a clothing brand on their mobile, goes home to browse on their tablet in front of the TV, then makes a purchase the next day via desktop. It’s also possible to understand what ads worked on which consumers and which device. This allows advertisers to serve relevant ads without the restrictions of the "walled gardens" of social media or other login-based platforms.

Ultimately, the question of accurately tracking and talking to users comes down to one of identity and creating a 360-degree view of today’s always-on consumers. Unified logins have near-100 per cent accurate identities but are at the risk of individuals’ private data, while probabilistic models can show a user’s path to purchase across devices with high accuracy, no privacy concerns and, in most cases, at a much larger scale. We’re moving away from a model where the ability to track and attribute effectively across devices is held by those with single-login environments and/or dependent on encroaching on consumer privacy. This is a positive for both our industry and consumers.

Nimeshh Patel is the vice-president, strategic partnerships, EMEA at Drawbridge