Yandex company blog

Not All Your Clients Look Alike – unless we handpick them for you

We have been precision-targeting our advertising audiences for quite a while. First, we fed the information about online behavior of millions of our users to our online behavior analytics technology Crypta, which used the unique capabilities of our machine learning method MatrixNet to learn how to tell men from women and one age group from another and show them only the relevant banner ads.

Then, we partnered with digital communications group VivaKi to create a tool for targeting display ads to web users based on the amount of time they spend watching TV.

And now, we took it a notch further by covering all kinds of demographics and audiences with one universal clickthrough-oriented targeting tool, which shows banner ads to web users based on similarity of their online behaviour to the behaviour of those web users who have already visited or performed specific actions on the advertiser's website.

Our new Crypta-powered targeting technology finds patterns in a website visitor behaviour and then tracks down those who exhibit similar behaviour online. Combined with the information about users' actions on a website, provided by our website analytics tool Yandex.Metrica, the new targeting allows advertisers to show their ads to a very narrow audience – those who have already made a purchase on their website, for instance. Or, conversely, those who put an item in the basket, but didn't pay for it. Any website owner with at least 15,000 visitors per week – the minimal sample size for the technology to find patterns – can use this targeting.

The look-alike targeting works only for those users, whose behaviour matches to a large degree – for up to 80%, for instance. The behaviour match threshold varies from audience to audience and depends on advertisers' goals.

Field tests showed impressive results. Using our look-alike targeting, clothing company Quelle managed to increase clickthrough rates from their banner ads by 300%.

We also experimented with manual campaign optimisation and boosted first results for Quelle by almost 100% in the fourth week. Our next step is to offer automated campaign optimisation individually to each of our advertisers.

Besides increasing conversion rates at virtually no cost, look-alike targeting gives advertisers some generous windfall in the form of additional visitors who delayed their conversion. KIA's advertising campaign with look-alike targeting showed 145% increase in delayed conversions compared with conversion rates from the same banner ads without look-alike targeting.