Patterns In The Surf: In an era of media fragmentation, the need for mixed media planning is paramount. But internet audience behaviour has hitherto been measured separately from other media, if at all. Peter Callius, Anders Lithner and Stefan Svanfeldt outline a solution from Sweden.
ORVESTO Internet: its structure and components Described schematically (see diagram below), the postal survey ORVESTO Consumer with a sample of 50,000 respondents is used to recruit an internet panel. Each of the panel members is then asked to accept a simple cookie file from the RealMedia traffic measurement InsightXE (in turn linked with site-centric banner systems) on each of the computers that he or she uses to access the internet.
Since the respondents of the postal ORVESTO Consumer keep their identification number all along, from the postal survey, to the panel and all the way into the little cookie file placed on their computers, we are able to identify the panel members in the site-centric traffic measurement and attach all the information from the postal survey to the electronically monitored traffic patterns of the panel. This way internet audience behavior can be analyzed single source with other media consumption and on the background of rich target group information. All this is made possible by the Sesame analysis platform.
InsightXE, being a total traffic measurement and not a measurement built on a statistical sample, reports the number of unique web browsers visiting any site during a certain period, as well as the number of visits, page views and a whole range of other key values. This data is of great importance to the individual websites, since it gives a lot of information, in real-time, on matters such as what sections are the most visited, how the visitor navigates, how the site performs electronically and so forth.
The basic figures from the traffic measurement are also made available in the Sesame software, where they are presented pretty much as the press circulation figures which are published alongside the reach figures of print media titles.
The traffic measurement figures are also used to produce a frequently-published top list and since the traffic measurement of websites at the media title level is closely linked with corresponding systems from RealMedia to measure banner performance, the plan is to 1) give TGI data on banner level; and 2) to use the banner system to collect internet advertising spend data.
ORVESTO Consumer: the base studyAs mentioned earlier, one of the two pillars on which the measurement rests is the postal survey ORVESTO Consumer. The survey is carried out three times a year and has served for many years as the print media industry standard in Sweden. Alongside the print media reach questions, the questionnaire also contains a vast range of TGI data as well as questions on cinema, outdoor and direct mail consumption. As a result of a second interview on the same sample the database made available to the market also contains reach and frequency data for radio and television.
Some 5,000 respondents complete the ORVESTO Consumer questionnaire each year and the data is delivered in the Sesame planning software – a platform used by just about all key players on the Swedish media market (see below). That means that, unlike many other countries, a single-source cross-media measurement already exists in Sweden.
Since the release of the Sesame Multi-Media Module in 2004 ORVESTO is becoming more and more used for cross-media analysis, but until today the internet figures have been based on recency questions in the postal questionnaire. For obvious reasons that is not the most accurate way to measure internet use. It works for the top level reach of large websites with strong brands and no blurry alliances with content providers, but it gets too rough on subsite level and in all cases where the website brand and the website URL differs from each other. That is why the postal recall data is now replaced by electronic panel measurement.
The user-centric internet panelThe panel used for the internet measurement described in this paper is recruited from those ORVESTO respondents that do not actively disagree (by checking such a box in the questionnaire) to participate in further surveys from Research International (or actually from SiFo, which due to its uncontested public recognition as the “official” provider of opinion, media and market statistics is the brand used by Research International Sweden when communicating with survey participants). Today, some 17,000 panel members have been recruited this way and approximately 8,000 of them have activated their computers in the measurement.
The e-mail address of the panel members is taken straight off the questionnaire if the respondent has filled it in; otherwise it is gathered with the help of an additional telephone interview to everyone that uses the internet and does not actively disagree. As soon as a panel member is recruited all communication is taken care of by e-mail unless the respondent chooses to call the support phone number.
It should be said that the representation of the panel is remarkably good. When we compare the unweighted panel with the respondents in ORVESTO Consumer that claim they use the internet regularly, we find no significant biases at all when it comes to gender, age, region, income, education or even internet use. However, when compared in terms of softer properties such as interests and lifestyle indicators we find small biases that call for weighting procedures that will be described later on.
The electronic measurement stands and falls, however, with the representation of the universe which is not just of the sample of individual panel members, but also of the sample of measured computers.
To make sure we measure all computers used by the panel members, but only the computers used by panel members – when they themselves are using them – we need to have a good picture of their internet environment. Knowing this also helps us statistically correct the data in the cases where we find ourselves measuring too few computers or computers used by someone else than a panel member (for a further discussion of this, see below).
The incentive system is partly based on how many computers a single panellist activates in the measurement and in order not to tempt anyone to over- or understate anything; the information of the panellists’ computer setup is gathered before we tell them that we like them to be part of an ongoing measurement. Therefore, prior to telling the panel members what we are about to do, we find out:
- The number of computers used by each panel member.
- The number of persons sharing each computer.
- The percentage of the usage on each computer that is done by the panel member.
- The percentage of the panel member’s total use that is done on each computer.
- The location of each computer (home, work, portable, other).
After collecting this data we ask the panel members to accept a simple cookie file on each of their computers. The cookie file is sent to their computers via a click on a link – it takes no installation, in fact the panel members do not even notice the cookie file being sent to them.
With the help of various reminders and incentives we make sure that the panellist accepts the cookie file on all the computers that he or she uses and not just the computer from which the initial survey is answered. In this process we also make sure that all computers that are used by more than one person has our cookie sending page as its browser start page. The start page is used to separate the panel member from other users of the computer (see below).
When comparing the number of work and home computers that are activated with a cookie with the number of home and work computers that the Swedish internet users claim to use in questionnaire surveys, it turns out that both kinds of computers are represented at accurate levels, with no statistically significant bias at all. The main reason for this is that no installation is required on the client side and hence no corporate policy or public suspicion about foreign software is there to reduce the number of activated work computers.
All that is needed is the sort of cookie file that any computer – home or work – receives in the dozens when just surfing the net. The single source connection with other media and TGI currencies left aside, this is the biggest difference between this measurement and other attempts to measure internet by electronically monitoring the behavior of a panel. For the first time all of the use is mirrored and not just the use from home.
Given the fact that roughly speaking a third of all internet time in Sweden is spent at work and significantly more in some target groups this is an absolute necessity for any advertiser or media planner who wishes to fully understand how the internet works and benefit from it.
The site-centric traffic measurement: InsightXEThe traffic measurement has already been described in some detail. It is a browser- or cookie-based traffic measurement operating with the double aim to 1) give the electronic and editorial departments of clients some insights about the visitor’s behaviour (in that sense InsightXE is a content management system); and 2) give the market department reliable figures to communicate to partners, buyers of advertisement space and the public at large.
An important aspect of InsightXE is that the sub-sites of a large website are measured separately as well as on an aggregated level and that the sub-sites are separated and labeled the same way as in the banner system. That way the measurement measures the exact same sections that are sold as advertisement space.
The customers get access to their own figures, at a very detailed level, in real-time in an online interface that is protected by a password. Only the key figures are published in publicly accessible platforms such as the weekly top list and the Sesame software.
By placing an InsightXE cookie on the computers used by panel members, modified to include the ORVESTO identification number, we are able to use the InsightXE data capturing to monitor the surf patterns of a statistical sample (the panel) about whom we know a lot of other things.
Banner measurements and optimizationAlongside the InsightXE traffic measurement RealMedia also offers banner management systems for the websites – the selling side of the process (OpenAdstream) – and for the agencies, the buying side (OpenAdvertiser). These systems too are cookie-based and the work to integrate them and turn them into one single platform has come a long way. In fact InsightXE and OpenAdstream are already integrated in a way allowing for behavioral targeting. The way this works can be illustrated by an example. Say a website has sold out the advertisement space in the Economy section. The website could then group the visitors (cookies) that have visited the Economy section at least X times the past X periods and then direct banners to that group wherever they are on the site. This way, advertisers can find the same target group as on the Economy section without actually placing a banner in the sold out section. This concept is called InsightACT.
As the different RealMedia systems converge they will automatically pick up the traffic of panel members. That means we will be able to provide TGI data on actual banner level.
The Sesame media planning platformThe media planning software, Sesame, is already being used by all media categories in Sweden. It was therefore an obvious decision to also add internet into the same software package. Planners can now plan internet cinema, print, television, radio and direct mail in the same software package and on a single source database. It should be stressed that Sesame first and foremost is a planning tool – not a tool for postevaluation.
By averaging, for example, four weeks to produce average weekly figures we are producing more stable figures. And when using data about historic events in drawing conclusions about the future, stable data are in every way preferable. But this also means that dramatic changes in audience size between one week and another will be smoothed. Since, however, the traffic figures are published simultaneously without any averaging, the short-term changes needed for post-evaluation are reflected, though without TGI information.
As mentioned, the panel is recruited from the ORVESTO Consumer respondents. But for obvious reasons all panel members do not originate from the most recent wave of ORVESTO Consumer. To make multimedia analysis possible between internet and other media on the currency level (the most recent survey wave), Sesame is using a rather complex weighting, ascription and calibration routine to match the panel data with the most recent measurement of the other media. This, however, does not significantly change any patterns, since the panel and the most recent ORVESTO Consumer respondents all share the same TGI data. The weighting, ascription and calibration can be done with a very high precision.
The internet figures are presented in Sesame in the same fashion as other media. The planner is given a great degree of freedom when analyzing reach and frequency during different time spans and in different target groups. Sesame allows for everything from simple cross-tabulations to complex planning based on OTS and with net and gross reach figures published side by side with the total campaign costs.
Concept summaryAll in all, ORVESTO Internet is a very complex concept. It builds, however, solely on known and well-tested techniques. A low-tech postal survey is used to recruit a panel. The data capture is based on simple cookie file transactions and the reporting is done in a tool long since well established on the Swedish market. In developing and marketing the concept it has been a key ambition to make this really simple for respondents, as graspable as possible to the market, but as advanced as it gets in its production details.
Source: Atticus - Volume 12 - Page 44, published in 2006