By Brendan Kitts, Chief Scientist, Adap.tv
Conversion attribution on the Internet is ubiquitous, and many methods – including “last click” and cookie-based matching – are used to match a conversion event back to the initial search or banner ad that the user saw or clicked on.
This technology is seamless, automated, and taken for granted: marketers just drop a conversion tracking script onto their websites, and, presto, they have conversion counts and rates for their various ads.
The widespread availability of conversion data has arguably led advertisers to be able to confidently optimize their ads online with a greater degree of granularity than ever before. Different ad creatives, devices, mediums, keywords and times of day can all be used in optimization using conversion data that has been collected on them. Indeed, one could argue that conversion-tracking systems – with their incredibly fine measurability – have been the engine that has powered the online advertising revolution.
In contrast to online, conversion measurement in TV is nothing short of an unsolvable labyrinth for advertisers. When an ad airs on TV, conversion events can happen on the Web, over the phone and even retail stores; and there’s no way to know whether those sales were driven in part to the original TV ad, let alone what particular TV airing caused it.
With the goal of bridging that gap, Adap.tv has been hard at work in aligning television closer to digital within attribution modeling. Today, we are happy to share the news that the US Patent and Trademark Office (USPTO) has granted Adap.tv an important patent for TV conversion tracking.
This patent validates our work in solving a critical problem for marketers: Online advertising systems have a problem with over-attribution. They tend to be very good at taking credit for – well – everything! TV has the opposite problem; TV tends to take no credit at all.
For example, because online advertising systems are often designed to attribute everything that is last click, it is not uncommon to see branded search keywords (eg. Verizon.com) with an amazing cost-per-acquisition in the pennies thousands of conversions. At this kind of incredible cost-per-acquisition, shouldn’t a marketer just shut off all other marketing campaigns on TV, radio, etc. and pour those budgets into online keywords?
Experienced marketers know that such an approach is likely to result not only in a bloated keyword campaign, but could also spell the end of their business by shutting off one of the most important parts of the sales funnel – people who need to learn about and be exposed to the product for the first time.
The reality is the people who typed in the brand’s name into the search box overwhelmingly already know the product’s name and what it does. The paid keywords are being used as navigational links to reach the site. To put it bluntly, keywords can receive credit by simply being in the right place while a user’s navigating to the site. It’d be like a promoter going up to fans waiting in line to get into a concert, giving them a coupon, and then claiming credit for the fans walking through the door. (Recent reports corroborate this phenomenon.)
This dichotomy between over-attribution in digital and lack of quantitative data in television makes it extremely difficult for marketers to calculate how much budget to apply towards TV. Further increasing the difficulty, the lack of granular conversion tracking also makes it impossible to do things like optimize TV media towards the media that is producing the highest number of conversions. In a media environment where every dollar counts, this can be incredibly wasteful.
This patent describes one of Adap.tv’s automated conversion tracking systems for television, designed to go beyond last click-type attribution to credit assignment based on properties of the conversions and media event. US 8,768,770 “System and Method for Attributing Multi-Channel Conversion Events and Subsequent Activity to Multi-Channel Media Sources” describes a process where a machine-learning system is trained to recognize conversions that come from TV, based on cases which are clean enough for deterministic attribution – often about 1% of the cases. It then estimates the probability of conversion for the larger set of TV airings based on those learned patterns between airing and conversion.
The technique makes it possible to probabilistically separate organic Web activity from Web activity driven from TV. This kind of signal source separation is similar to the “cocktail party problem,” where one tries to segment specific conversations out of several conversations happening simultaneously within a room. The system can also report on conversions that are due to marketing events that were not TV – For example it would be equally bad to over-assign credit to TV for understanding which TV airings have been more effective – in order to do this, not assigning credit is just as important for optimization purposes as assigning credit.
Adap.tv has worked diligently to disambiguate, measure, and report on TV’s impact and make it possible to budget TV in a rational way. Adap.tv and Convertro, the leading multi-touch attribution platform recently acquired by AOL, have years of experience running television campaigns and measuring television sales response. In general, our results suggest that TV effects are extremely large, distributed across multiple channels, extended in time, and are woefully under-reported.
With better measurement techniques, we believe that it will be possible to bring hard ROI measurement to TV and put it onto a similar footing as online advertising. That’s good for everyone, not only in television, but also in online advertising too. Optimizing your marketing campaign using keywords only is equivalent to looking for your keys by only searching under a street lamp – just because there is some light there, doesn’t mean that’s where you should be focusing.
The objective should be to cast light on all parts of your marketing campaign.
Read the patent contents here: http://bit.ly/1oCejXA
A scientific paper on the algorithm was published at IEEE International Conference on Data Mining, along with attribution results from various television advertisers. You can read that paper here: http://bit.ly/1zpDXEE
Conversion attribution on television, radio, offline, and digital channels, is performed by Convertro who use a multi-touch attribution (MTA) algorithm to measure effects across channels. You can learn more about Convertro’s technology at www.convertro.com.