This document discusses audience data and targeting. It explains what audience based targeting is and how contextualization and search retargeting work. It discusses the difference between registered or declared data and in-market data. The document provides examples of how audience data can be used to build marketing solutions, such as targeting in-market audiences or building look-alike models.
7. How does search retargeting work? User searches from any search engine on any site, including product searches, i.e. not just Google User clicks through to a landing page from the search results page. If the landing page is on a site is within the Tribal Fusion data network then... ... search referral URL passed to TF for classification and appended to cookie, e.g. http://www.google.com/search?q=toshiba+laptop Serve ads to impressions generated by users within the advertisers search retargeting segment Users that have searched those keywords within our data network create a search retargeting segment for the advertiser Build custom keyword lists with advertiser from their SEM or otherwise for display, e.g. “Toshiba laptop” would be just one of the keywords
12. Scoring Incoming Ad Requests Ad Campaigns 78 22 Use Case 1: Optimisation and targeting engine 93 Rich User Profiles 93 Ads served Clicks 93 Conversations All Auto: In-Market Mini-Van Toyota Sienna All Home & Family Parenting
14. Using audiences to build marketing solutions B2B computer equipment manufacturerpreviously using tech channels to reach target can now target people that work at small businesses for their SMB products Domestic travel provider using location data to power dynamic ads with route and price info. Next step is to ingest audience data in messaging Mobile network operator reducing eCPA through applying interest based audiences as well as usual intent data Auto manufacturer targeting in-market audience for their own and competitive brands FMCG brand building look a like model of visitors to its brand site to extend reach info@tribalfusion-corp.com
In the first half we spoke to a lot of people about an network 4.0. The concept that networks are in their 4th generation of evolution from site rep through blind network through audience network to platforms capable of providing complete solutions to compete with the best of the point solutions now in the market. These are all the capabilities that networks will have to excel at if they are to survive and thrive.
This time we’re going to talk about audience types – how they work and what the challenges are – and how they can be brought together within a data network and then used for advanced targeting purposesWe had originally intended to talk about Insights within this talk but there was too much good content to do the topics justice in the same session
Marketers have traditionally bought specific publisher brands and sites to advertise to the assumed audience of that publisher. This is what they have been doing offline on tv and print for > 50 yearsBuying sites today is still valid – and that shouldn’t go away. There are times when you do want to associated with a specific brand or some specific contentTechnology today allows markets to advertise to specific audiences and to differentiate between lower and upper funnel activity. The ability to buy audiences is what is driving much of the disruptive change in online display today. Buying audiences on demand is an entire topic of its own
4 key types of raw audience data online today and two special casesBrowsing data based on the sites and content you look at. This can be used to infer interests for targeting and in models to predict purchase intent or socio-demographics. The challenges here are building segment scale (and finding those users again on your network) and categorisation. This is observed dataSearch data is a strong indicator of intent (when not navigational). This can be used as a feed into your interests and intent too. The challenge here is scale (unless you are Google!)Registration & social network data provides declared data on socio demographics and interests, with Facebook a great example of how rich this can be. The challenge here again is scale if you’re not Facebook, Yahoo or MSN AND, quite rightly, any privacy concernsOffline data sets are the more recent development where we are trying to bring offline – typically direct mail type – data sets online. Scale and privacy are again the concern hereThe first special case is “in-market” where the content or the keywords are strong indicators of purchase intent. This is great for DR to help you grab share rather than building customer basesThe other special case is retargeting, which is closest to the browsing behaviour and simply targets a user based on a recent visit to a specific publisher or advertiser site (think retargeting Telegraph readers or people recently on Easyjet) This is not technographic (what browser, what ISP, what device) or geographic (based on IP location or declared data)Other than retargeting we’ll talk about each in turn. These data types can be used for optimisation, targeting, modeling and insights
- The alternative route is to do deals with the owners of search engines (e.g. Yahoo, MSN, Google) but also the site search data of media owners and comparison shopping sitesWe see about 3% of queries in the UK through our data network as compared to 4-5% each for MSN and YahooChallenge is to have both scale in the target keywords and the underlying query mix. For example if you’re picking up searches skewed to navigational queries – i.e. People searching for Google from Yahoo – this won’t be very rich. If you don’t have finance sites in your data network then you’re unlikely to get volume in finance queriesAt scale search retargeting is good for intent but often it’s lazy marketing – you’re targeting the people who are already likely to buy your productYou can use this to target keywords that indicate interests and audience types, i.e. Certain keywords are used predominantly by men, or by teenagers, or by people with a certain income BUT performance is not as high
-Facebook is the best example of how registration and declared interest data can be gathered – though Facebook data is available only through Facebook targeted ads-In my Facebook example I’ve provided name, email address, gender, age, school, university, work place, interests and my social graphThere are lots of places where declared data can be gathered. All of these data points can be used for insights, targeting and modeling purposesBUT privacy and privacy regulation is the catch for anyone using this kind of data. PII must be protected. “This is information that, either by itself or in conjunction with other information, can uniquely identify an individual. Examples of personally identifiable information include a name, street address, email address or telephone number. PII is defined in EU directive 95/46/EC.”Creative design/ messaging and target selection have to be used carefully. We can over use this data. Users are happy to get ads when relevant BUT not when it is invasive, creepy and transparent that they’re being targeted based on information that they have revealed onlineWith social and registered data it is all to easy to cross that lineDynamic product retargeting is the current hot topic here but it’s easy to see how much more serious this could be for similar cases around targeting using this data
Offline data is related to our previous discussionbut this time these are profiles built up from info available about you offline rather than your online declared info, e.g. electoral role, credit scoring, census data, households surveys, purchase dataThe idea here is to find the sweet spot of overlap between an offline set of records, an online audience at scale and the right kind of mediaIn the US there are several options in this area with companies like datalogix, TARGUSinfo. In the UK there are similar companies but, with few exceptions, they have struggled to get online, e.g. Acxiom, CACI and Dunnhumby. Nectar and Experian have done a good job with Yahoo and MSN respectively but still not big businessWe built our data architecture to integrate external data from day 1. We recognised that our data wasn’t enough in terms of breadth or depth to meet all our client needs. Today we’re 60% internal data and 40% third party data. It’s easy for us to integrate with these players in the US where the online mapping has already happened. In the UK we’re looking to do that ourselvesScale is the challenge here in a market the size of the UK. How many “rural squires” or people with household income over £100k or carbonated beverage brand switchers can you find on your network for a targeting campaign with scale. Not enough. We use look a like modeling to exted these segments
In-market is built simply by asking whether some content is demonstrating clear purchase intent. There’s a difference between reading an enthusiasts review about an Audi A4 and a new product review of an Audi A4. The first in an interest, the second is an intent.We separate our topics by “content” and “in-market” to recognise just this. You can buy people looking at Audi A4 content, or people looking at Audi A4 product reviews or both audiences to get scale. You can also roll up to all Audi, all luxury cars or all sedans.You can use this data as input variable for dynamic ads, e.g. show an audi ad to people with Audi intent, show a competitive message to those with BMW intent or bring in other data like safety messages for people with family content behaviours. This is also an extension of your retargeting. Audi will find more people that are in-market for their cars than may have visited their site In-market is more direct because you’re not assuming purchase intent from recency and frequency of interest content BUT as with search you’re only catching lower funnel behaviours
Modeling is the final case here. You start out with an audience that meets your criteria of intent or membership of a specific socio-demogrpahic group. Using all the browsing, search, registration and offline data that you have access to you can then build a model of audiences that “look like” the original group. This expands the original audience and gives you the scale to run a campaign. This is reach extension – finding more people that look like the ones that I want. The trade off in the model is that as you increase the reach you decrease the quality of the modelFirst way to do this. Based on behaviours you want then force all users to fit a model. Typically done by linear regression (or neural networks) BUT you don’t know in advance if it is going to work. So companies that do this have repositioned themselves as brand solutions because the success metrics are softerSecond approach tackles this the other way and starts with the behaviours that perform or fit best. We keep including segments until we reach the trade off between reach and model performance that we want, e.g. 5x an audience that is 5x more predictive than the average. Simpler and more powerful
We are typically asked to help marketers “find my audience”