±±¾©Èü³µpk10

This is clearly good news for the online industry, but if the web is to take its place alongside TV, print and other traditional media as a proper, grown-up communications channel, then advertisers need to feel confident that the statistics on which their media decisions are based are robust and accurate.

Yet this is often far from the case, according to eCRM company RedEye.

Last year, RedEye undertook a study of the traffic on two of its clients' websites, williamhill.co.uk and asda.com, over a 28-day period.

The majority of page requests on these two sites are made by existing, logged-in customers, so RedEye was able to use this data to compare the actual number of visitors to the sites with the number of visitors counted by the websites' IP (Internet Protocol) and cookie-based server logs.

RedEye found the IP logs inflated visitor numbers by 660% and cookies by 128%. This, the company says, could lead a website to underestimate its conversion ratio of visitors to customers by a factor of 7.6 times.

Thus a site converting 10% of visitors to customers could be reported as 4.3% by a cookie-based system and 1.3% by an IP-based system. Given that most sites use one of the two approaches, it is likely they are more effective at driving sales than they are being given credit for.

There are good reasons why both approaches are prone to a degree of error.

IP-based logs rely on the IP addresses assigned to each visitor to a website.

Since internet service providers (ISPs) assign these addresses dynamically, the same person can be picked up as three separate visitors on three separate visits, or even as multiple visitors during a single session, if they are assigned with new IP addresses during that period. Cookie-based logs are more accurate, but these can be skewed by web users who delete cookies.

Once they do so, they are identified as a new user, even if it is a site they use regularly.

'The research has put a definite number on a problem a lot of people were aware of,' says RedEye chief executive Paul Cook. 'It also illustrates the differences between the objectives of companies' commercial and technical teams. The majority of companies still use an IP-based system for measuring traffic to their sites because a lot of marketers do not understand or feel the need to care about the methodology.'

Efficiency needs

Most online practitioners seem aware that there are flaws in the methods used to measure web traffic. Sites that require a log-in can measure the number of visitors most accurately, as the identity of the user can be checked, and in effect, 'de-dupe' the logs. But relatively few sites operate such a device.

'The industry has always known that there's a problem identifying visits and visitors,' says Mat Morrison, consultant at design agency Tonic. But, he adds: 'We hardly ever need the most accurate and exhaustive data to run efficient businesses; bricks-and-mortar retailers don't measure with any precision the number of visitors through their doors, and only a few bother to track repeat custom. Off-line publishers, studios and broadcasters can make efficient business decisions using nothing more than sample data.'

As long as online professionals understand the limitations of the technology and plan what they want to measure and why, Morrison argues, they can still make informed decisions.

Conrad Bennett, technical service manager for EMEA at web analytics company Webtrends, concedes that online measurement is an inexact science. 'Anyone looking for absolute values will in most cases be disappointed,' he says.Beyond the numbers, he adds, firms look to online metrics to tell them not just how many people are using the site, but factors such as what they do when they get there and how they landed there.

Peter Matthews, managing director of new media agency Nucleus, says the key is to decide beforehand what these factors are. 'Key performance indicators for a client in travel will be different from a retailer, so when designing a site, you have to decide the indicators by which the site will be measured, and build them in from day one.'

Pursuing precision

Richard Foan, managing director of ABC Electronic, the online arm of the Audit Bureau of Circulations, points out that the RedEye research is just one of many audits conducted over the years. RedEye, he adds, is a member of the Internet Technical Group (ITG), a body that meets bi-monthly to discuss and set online measurement standards. The ITG, together with the Joint Industry Committee for Web Standards (Jicwebs), has been working on online metrics issues for several years.

'We have about seven or eight years' experience of applying these metrics across thousands of audits,' says Foan. 'As more people understand and make use of these standards, so they can be part of the debate and the evolution going forward.'

Danny Meadows-Klue, chief executive of the Interactive Advertising Bureau (IAB), presents a robust defence of online metrics. 'We need to do more work on IP and server-based logs, but if you take a step back and ask how it compares, there is nothing like this in other channels,' he says.

'We are streets ahead of other media in accuracy, but we are all a bit geeky in the online industry and try to bring precision to new levels.'

At Clickthrough Marketing, managing director Phil Robinson says there is increasing interest from companies in using its ClickMetrics web analytics software on an application service provider (ASP) basis. Rather than buying a licence for everyone in the company who needs access to the data, companies access their web metrics online via a secure, password-protected site, enabling them to see a wide range of statistics, including referring web sites, search words in rank order, campaign conversion rates and most popular pages.

The service costs about £35 a month, according to Robinson, who says: 'The beauty of something like this on an ASP service is that there are no licensing issues and no waiting around for reports. Anyone in the company who needs to see the stats can do so and get real-time visitor tracking.'

Final analysis

Nielsen//NetRatings is working on a system to combine its demographic approach with the census-based method of cookie-based server logs. The company measures and analyses internet audiences by means of user panels - it has a 6000-strong 'at home' panel, and a smaller 'at both' (home and work) panel - with a known demographic make-up. While this is good for demographic profiles of site visitors, it falls down when panellists access sites outside the home or work. Also, it does not provide the absolute numbers that website owners and media sales houses like to use to sell their site to advertisers.

To counter this, the company is working on a system that will combine the panel information with what Day calls 'site-centric' statistics, which will use a cookie-based tagging system to count the number of visitors to web pages. Because the panel members are known to the company, it will be able to remove the skews caused by cookie deletion.

By doing so, Day believes, the company will be able to combine the 'publisher viewpoint', which renders the biggest numbers but does not take account of duplication, with the media planning viewpoint, which is more concerned with the reach of a campaign among a demographic profile. 'The only way to unite those two schools of thought is to link the audience information provided by the panel and the census information from the site-centric statistics,' says Day. The methodology, called NetViews Plus, is already in use in Australia and Sweden, with plans to introduce it to the UK early next year.

As the RedEye research makes clear, the methods used to measure website traffic are by no means perfect, but they are no more imperfect than those used in other media, and the online industry seems driven by the need to improve.

'The bigger players in the market are aware that it's up to us to raise the bar and improve standards,' says Richard Dunmall, sales director at advertising sales house AdLink. 'We still only represent 5% of overall marketing spend. If we really want to punch our weight, we have to continue to put pressure on ourselves to raise standards and move things forward.'

THE INDUSTRY BODY

Danny Meadows-Klue chief executive, Interactive Advertising Bureau

'IP address and server log-based systems need more work, but there is nothing like this level of measurement in other channels. We have a sample size for online metrics that is of greater magnitude than BARB's panel.

'If you look at web publishers, such as Guardian Unlimited, there is a far greater degree of accuracy in what they can say about their medium than other channels can provide. There is a long-term need to carry on exploring how counting can take place on a global basis. We want to go further because we know can.'

THE CLIENT

Michael Smith head of digital advertising, Central Office of Information (COI)

'The COI runs 70-80 bursts of activity each year, across about 40 campaigns. Some are broad messages, others are niche. We rely on our media planning agency to recommend sites to us, but we also "sense-check" them.

'Page impression and unique users are important, but once a campaign is running, we look at it against our objectives. If a site has lots of unique users but no one is clicking through to our site, we may ask questions. But for awareness-raising campaigns, there may be no need to get people to our site.'

THE MEDIA AGENCY

Matthew Pitt managing partner, i-level Generator

'Too many media owners are fixated on page impressions or unique users. This is short-sighted and they would do better to understand the metrics that matter.

'There is a lack of willingness to spend money on bespoke customer insight research, which would tell a media owner more about the people visiting their site, their motivations for doing so, and so on. This is because media owners think buyers are only interested in impressions, and in many cases they're right. But the trick is to reach the people who count, rather than counting the number of people reached.'

ONLINE MEASUREMENT

Methodologies IP/web server logs

These keep a record of referrers, key search words and so on, but tend not to include persistent cookie values, meaning unique visitors can't be measured. Pages stored in cache memory are not measured and ISPs' practice of dynamically assigning IP addresses to individual users can mean the log counts the same user as multiple ones. Groups accessing a site via a proxy server can look like a single user.

IP + user agent

Like IP logs, but they attempt to distinguish individuals within corporate user groups by identifying different PC operating systems and web browsers. But users accessing a site via a corporate internet connection are likely to share PC and browser specs.

Tagged pages/cookie-based logs

Use HTML code inside web pages to send a request to a stats server when a page is viewed. The code can discover web browser and computer settings, and tell the stats server about any web referrer, and the search words used. Persistent cookies can be reliably sent to users to measure when they return to a site.

Panel-based statistics

Deliver measurements using a panel with a known demographic make-up. Good for identifying the types of users visiting a site, but does not provide census-type counts or register web access from points outside that for which the panellist is registered.