Good practices

Realytics' scoring, or how to identify TV visitors with precision

  • Home
  • »
  • Realytics' scoring, or how to identify TV visitors with precision

"The scoring", what is that? Definition

For Realytics, the scoring is the analysis of all the sessions of one cohort, in the minutes following the diffusion of a spot, in order to construct an algorithm that can find the most likely TV-exposed sessions.

We can see you from here, all skeptical and frowning your eyebrows ("whaaat?").

Let's take it from beginning. Like we told you in our article about our baseline, in order to compute our baseline, an analysis window of a few minutes opens up right after the diffusion of a TV spot. We call these few minutes "cohort" because it represents all the persons who watched the same spot at the same moment, during which we've identified all the visits on the advertiser's website (or the call to the call center/app downloads... TV is not only drive to web media!).

It is afterwards the special moment where our scoring algorithm comes into play. A score is given to every identified session during this window of impact and those chosen then become "scored TV" users.

 

Realytics' scoring and the "scored TV users"

Are you scored 1 or scored 0?

All the persons that are identified during the cohort are named scored TV users; the scored TV "1" are more likely coming from TV when the scored TV "0" would have come anyway, without any spot TV; they're then part of the baseline.

How can we differentiate them, will you ask? Realytics' smart algorithm takes into account the operating system (Windows, iOs...), the device (cellphone, laptop, tablet...), the web browser, the country, the application system (web site, mobile app..),etc.. Around 10 criteria are analyzed in order to qualify the TV users as best as we can.

If you should guess, where would you say the users coming from TV are more likely situated?

Realytics has a specific way of knowing it...

As you may notice it on the graph up there, the traffic we can observe right after a spot is broadcasted is pictured in blue. We've divided the cohort in different segments, and we estimate that the segment which is more likely coming from TV would be the one with the ratio impact / traffic the more important.

Then, we compare the rest of the cohort to this segment most likely coming from TV, to which we've given the grade "1".

TV scored users are the qualified like this because they show the most criteria in common with users whom behavior show they're probably coming from TV.

 

Training our algorithm

Like every high level athlete (kinda like our Head of Data), the algorithm is trained on every segments of the session that have the same entry point (web, app, sms...) and the same device (mobile phone, laptop...).

Overall, a model of scoring, specific to every client, is trained again after 5 days, then 25, 125, 625... so it can keep on providing the most precise data possible.

 

Realytics' scoring: what's it for?

Why qualify those users and give them a score? For different reasons... (and actually, different solutions).

Digital Follow-up and the scoring

You already know it, but Digital Follow-up allows advertisers to drive the customer journey from TV to digital.

Retargeting and scoring

With our retargeting solution, advertisers can expose TV-engaged viewers again to their ads during their web navigation. Like we've shown with Just Eat, it is possible to think of retargeting a TV audience on Facebook. In this scenario, the advertiser observed +200% conversion compared to his Facebook campaigns as a whole.

 

Scoring and cohorts

We are following cohorts during 30 days, differentiating those scored 1 and 0, following the metrics chosen by the advertiser.

At the end of the 30 days, our team compares the results of the metrics followed by the 2 groups of TV scored users. It helps us define if the TV scored react better or not than the other users from the baseline and gives advertisers precious insights.

 

Scoring and lookalike

We know it by now, the scoring allows us to give a score to the users who came on a website following a spot on TV (the cohort). What if we could compare them to every profile coming outside of the cohort, allowing us to extend the audience reached thanks to TV?

In order to do so, we just gave to compare the web customer experience, the device, the session duration and find similar profiles to the ones already scored 1 and identified within the cohort. Afterwards, we define the probability there is to find some similar users. For example, we can say we'd consider as similar only the profiles that are more than 70% similar to TV scored 1.

 

Brand Effect and our scoring

Clearly brand-oriented, Brand Effect allows advertisers to measure the global impact of their TV campaigns so they can see if their campaign really shines upon their brand. The brand is then at the core of the advertiser's media strategy, who can follow the evolution of his branding KPIs.

 

 

Mots clés : Good practices