Organizations need to select the most appropriate tools for collecting and reporting metrics which meet requirements such as reporting of marketing performance, accuracy, analysis and visualization tools, integration with other marketing information systems (import, export and integration using XML standards), ease of use configuration (e.g. creation of custom dashboards and email alerts), support quality, cost of purchase, configuration and ongoing support.
Techniques to collect metrics include the collection of site visitor activity data such as that stored in web analytics systems and in site log files, the collection of metrics about outcomes such as questionnaires and focus groups, which collect information on the customer’s experience on the website.
Site-visitor activity data captured in web analytics systems records the number of visitors on the site and the paths or clickstreams they take through the site as they visit different content.
In the early days of internet marketing, in the mid 1990s, this information we typically collected using log files. This server-based log file is added to every time a user downloads a piece of information (a hit) and is analyzed using a log file analyzer.
Despite their wide use in the media, hits are not a useful measure of website effectiveness since if a page consists of 10 graphics, plus text, this is recorded as 11 hits. Page impression or page views and unique visitors are better measures of site activity. Auditing companies such as ABC electronic (abce.org.uk), which audit sites for the purpose of providing a number of visitors to a site to advertisers, use unique visitors and page impression as the main measures.
Information giving detailed knowledge of customer behavior that can be reported by web analytics include:
Page per visit (PPV) – the average number of pages viewed per visitor to a site (this is indicative of engagement with a site since the longer a visitor stays on a ‘sticky site’, the higher this value will be). PPV is a more accurate indication of stickiness than duration on a site in minutes since this figure is skewed upwards by visitors who arrive on site and are inactive before their session times out at 30 minutes.
Visits per (unique) visitor (VPV) – this suggests the frequency of site visits. Readers will realize that this value is dependent on the period that data are collected over. These data are reported for a month during which time one would not expect many returning visitors. So it is often more relevant to represent these data across a quarter or year.
- Top pages
- Entry and exit pages
- Path of clickstream analysis showing the sequence of pages viewed
- Country of visitor origin (dependent on the location of their ISP)
- Browser and operating system used
- Referring URL and domain (where the visitor came from)
- Greater accuracy than server-based analysis
- Counts all users, cf. panel approach
- Relatively expensive method
- Similar weaknesses to server-based technique apart from accuracy
- Limited demographic information
Measurement is often highlighted as an issue once the first version of the site has been ‘up and running’ for a few months, and employees start to ask questions such as ‘How many customers are visiting our site, how many sales are we achieving as a result of our site and how can we improve the site to achieve a return on investment’? The consequence of this is that performance measurement is something that is often built into an online presence retrospectively. Preferable is if a technique known as design for analysis (DFA) is designed into the site so companies can understand the types of audience and their decision points. For example, for Dell, the primary navigation on the homepage is by customer type. This is simple example of DFA since it enables Dell to estimate the proportion of different audiences to their site and, at the same time, connect with relevant content. Other examples of DFA include:
- Breaking up long page of form into different parts so you can see which parts people are interested in.
- A URL policy used to recommend entry pages for printed material.
- Group content by audience type or buying decision and setting up content groups of related content within web analytics system.
- Measure attrition at different points in a customer journey e.g. exit points on a five-page buying cycle.
- A single exit page to linked sites.
Often site owners and marketers reviewing the effectiveness of a site will disagree and the only method to be certain of the best-performing design or creative alternatives is through designing and running experiments to evaluate the best to use.
AB testing and multivariate testing are two measurement techniques that can be used to review design effectiveness to improve results.
In its simplest form, A/B testing refers to testing two different versions of a page or a page element such as heading, image or button. Some members of the site are served alternatively, with the visitors to the page randomly split between the two pages. Hence, it is sometimes called ‘live split testing’. The goal is to increase page or site effectiveness against key performance indicators including click-through rate, conversion rates and revenue per visit.
When completing AB testing it is important to identify a realistic baseline or control page (or audience sample) to compare against. This will typically be an existing landing page. Two new alternatives can be compared to previous control which is known as an ABC test. Different variables are then applied. For example: Test 1: original page and new headlines, existing button, existing body copy. Test 2: original page and existing headlines, new button and existing body copy. Test 3: original page and existing headline, existing button, new body copy.
An example of the power of AB testing is an experiment Skype performed on their main top bar navigation, where they found that changing the main menu options ‘Call Phones’ to ‘Skype Credit’ and ‘Shop’ to ‘Accessories’ gave an increase of 18.75% revenue per visit. That’s significant when you have hundreds of millions of visitors! It also shows the importance of being direct with navigation and simply describing the offer available rethan than the activity.
Multivariate testing is a more sophisticated form of AB testing which enables simultaneous testing of pages for different combinations of page elements that are being tested. This enables selection of the most effective combination of design elements to achieve the desired goal.
- Structured experiments to review influence of on page variables (e.g. messaging and buttons) to improve conversion from a website
- Often requires cost of separate tools or module from standard web analytics package
- Content management systems or page templates may not support AB/multivariate testing
Clickstream analysis refers to detailed analysis of visitor behaviour in order to identify improvements to the site. Each web analytics tool differs slightly in its reports and terminology, but all provide similar reports to help companies assess visitor behaviour and diagnose problems and opportunities.
Aggregate clickstreams are usually known within web analytics software as forward or reverse paths. This is fairly advanced form of analytics, but the principle is straightforward – you seek to learn from the most popular paths.
Viewed at an aggregate level across the site through ‘top paths’ type reports, this form of clickstream analysis often doesn’t appear that useful. It highlights typically paths which are expected and can’t really be influenced. The top paths are often:
- Homepage » Exit
- Homepage » Contact us » Exit
- Newspage » Exit
Clickstream analysis becomes more actionable when the analyst reviews clickstreams in the context of a single page – this is forward analysis or reverse path analysis.
On-site search is another crucial part of clickstream analysis on many sites since it is a key way of finding content, so a detailed search analysis will pay dividends. Key search metrics to consider are:
- Number of searches
- Average number of searches returning to zero results
- % of searches returning to zero results
- % of site exists from search results
- % of returned searches clicked
- % of returned searches resulting in conversion to sale or other outcome
- Most popular search terms – individual keywords and keyphases
Segmentation is a fundamental marketing approach, but it is often difficult within web analytics to relate customer segments to web behaviour because the web analytics data isn’t integrated with customer or purchase data.
However, all analytics systems have a capability for a different, but valuable form of segmentation where it is possible to create specific filters or profiles to help understand one type of visitor behaviour. Examples of segments include:
- First time visitors or returning visitors
- Visitors from different referred types including:
- Google natural
- Google paid
- Strategic search keyphrases, brand keyphrases etc.
- Display advertising
- Converters against non-converters
- Geographic segmentation by country or region (based on IP addresses)
- Type of content accessed, e.g. are some segments more likely to convert? For example:
- Saw search results
- Saw quote
- Saw payment details
- Provides competitor comparisons
- Gives demographic profiling representative
- Avoids undercounting and over-counting
- Depends on extrapolation from data-limited sample that may not be representative
There has been consolidation of web analytics tools, such that there is now basic choice of a free service such as Google Analytics or Yahoo! Analytics or paid services from the main providers such as Omniture (owned by Adobe Systems), Coremetrics (owned by IBM) and WebTrends which may cost hundreds of thousands of dollars a year for a popular site. All will report similar measures for digital marketing activity. Often the selection of best system will depend on factors as such:
- Integration with other data sources (for example, social media marketing, customer data and financial reporting). Types of data that need to be integrated include: operational data, tactical and strategic data.
- Accuracy. Potential sources of inaccuracy are: traditional log file analysis and more common browser-based or tag-based measurement system that record access to web pages every time a page is loaded into a user’s web browser through running a short script, program or tag inserted into the web page. The key benefit if the browser-based approach is that is it potentially more accurate than server-based approaches. The free version of Google Analytics uses sampling on large sites which can decrease inaccuracy.
- Media attribution. ‘Last-click wins’ model of attributing a referral source to sale is inaccurate. Weighted models based on the whole customer journey are more accurate. The capability of analytics system to display this is important for companies investing a lot in online media.
- Visualisation. How data are displayed through reports and alerts. Vendors continuously introduce new features in this area.
- Customisation facilities. For creating and distributing new reports and alerts.
- Support services. For configuration of data feeds and reports or consulting to assist in auctioning the results. In 2011 the free service of Google Analytics introduced a premium version for large corporate customers which included account management.
- Privacy considerations. Web analytics systems store personal data. It is important that data collection and disclosure about the method for collecting by the system follow the latest laws about use of cookies.
Performance management systems for senior managers will give the big picture presented as scorecards or dashboards showing trends in contribution of digital channels to the organisation in terms of sales, revenue and profitability for different products.
Chaffey, D. and Ellis-Chadwick, F., 2012. Digital marketing: strategy, implementation and practice (Vol. 5). Harlow: Pearson.