Tag Archives: analytics

Measuring Site Performance (Part 3)

Popularity metrics are a set of yardsticks by which you can judge the relative popularity of your site over time. The primary metrics are:

  • Unique Visitors
  • Visits
  • Page Views (Impressions)
  • Average Visit Length

Your web server archives the information needed to generate these numbers and many others. The raw data is stored on the server in what is known as a log file. The statistics referenced above are best accumulated through the use of a log analysis program to convert your hard-to-read server log files into an understandable format. The most popular of these programs is WebTrends (www.webtrends.com).

Let’s take a quick look at each of these popularity metrics. The number of Unique Visitors is perhaps the most vital statistic as it counts the visitors to your site and then factors out double counting. (Note that this statistic is far more meaningful to you than the oft-referred to Hits statistic.  Hits simply tells you the number of files transferred from the server to the visitors’ computers. While this initially sounds good, it falls apart when you learn that a single web page can contain a large number of individual files, each of which is counted and contributes to the total Hits count. Hence the Hits number can be easily manipulated by site owners by varying the number of individual files on any page.)

Be aware that there are limiting factors in the counting. The primary impact comes from what are known as Masked IP Addresses, that is, networks that automatically give all their users the same IP Address.

An IP Address is an identifying number given to each computer connecting over the Internet. Servers use IP addresses as a convenient way to track visitors. This brings us to an important point: Unique Visitors does not count people, it counts computers, and that is the root of the problem.

The biggest Masked IP Address villain is AOL. All AOL users share the same IP address,  so when 35 people from AOL visit your site in a week, your number of Unique Visitors stat will count only 1 visitor.

Another limiting factor is multiple users on one computer. If all four members of my family visit your site one week, you only see one visitor in the number of Unique Visitors stat.

The Visits statistics gives the total number of visits to your website during the reporting period. It is a useful metric when used to temper the Unique Visitors stat for purposes of arriving at an accurate picture of the trend of activity on your site. Remember of course to factor out double counting as one person visiting the site 25 times in a week will show 25 visits. In other words, remember this is Total Visits, not Total Visitors.

Page Views (also called “Impressions”) tells you the total number of pages viewed by site visitors during the reporting period. So, if Visitor A looks at just the Home Page, but Visitor B explores the site, visiting 9 pages before leaving, Visitors A & B would be collectively responsible for 10 Page Views (1 + 9).

The Page Views number is also susceptible to a degree of miscounting, as cache files cause undercounting and search engine robots can cause over counting.  The cache files problem is very hard to detect and according to some sources causes up to 30% undercounting. The severity of the problem depends largely on both the way your site is built and how your server is configured. This is a complex problem to solve and if it is key to your efforts it should be discussed with your IT team or vendor prior to construction of the site.

Search engine robots (or “spiders”) can be easily factored out if you use a program such as WebTrends. WebTrends maintains a separate count of visits by robots, allowing you to adjust for them with an acceptable degree of accuracy.

The final primary popularity metric is Average Visit Length. Ever wondered if your site is sticky? This is the key indicator. This should be tracked across time for a trend. Content-heavy sites, subscription sites and sites relying on ad revenues obsess on this number as it indicates pretty clearly the success of their efforts to draw and hold an audience to their site.

The numbers above are primary metrics — key indicators. You can easily go beyond these numbers for more information, but the numbers above should the first stop for inquiries relating to how the traffic on your site is moving across time. Next column, we’ll look at a variety of eBusiness metrics.

Measuring Site Performance (Part 2)

In this column we will continue with our examination of website metrics. Last column introduced the idea of performance metrics and the basics of what to measure. In this column I would like to go a little further into discussing the implications of those metrics.

From a business perspective, performance metrics are not an exercise in technical esoterics, but rather an attempt to gauge your site’s performance as it affects users of your website. For some firms, performance is mission critical. If, for example, your site includes ecommerce functionality, it is necessary to maintain an always up, always accessible website. A slow site or downtime translates into lost sales. Similarly, though of perhaps a less critical nature, if your site offers online customer service the performance of those customer facing systems is critical to maintaining customer satisfaction. But the list of those affected by poor site performance doesn’t stop there; poor performance will have a negative effect on any website. Web users tend to be impatient, and as the cliché goes, your competition is only a click away.

I am sometimes amazed by companies who tolerate under-performing web properties. Would you mail to your customers a brochure that they cannot read? Would you air a television commercial that contains only half a picture and garbled sound? Obviously not, but companies still produce similarly flawed websites — and then discuss cynically how the Web doesn’t deliver on its promises.

The web is not the problem, the sites are. While we here in Thailand must suffer on in a bandwidth desert, it doesn’t mean that Internet connections as a whole are problematic. Moreover, while we cannot do anything about our bandwidth situation, we can make sure that the sites we produce perform well, regardless of the users’ environmental constraints. We must strive for world class standards and to the extent possible offset the peculiarities of the local market with smart design.

Last week we talked about key indicators that would allow you to assess your site’s performance. I mentioned:

* system uptime
* server response times
* page download speeds
* server errors
* failed hits
* form failures
* effective bandwidth per user

What do those numbers tell you?

System Uptime tells you the percentage of time the server was available and delivering web pages to site visitors.  Down time is what we want to avoid.  Down time is an inescapable reality as servers require re-starts and re-boots due to a variety of factors, including perfectly normal operations like software upgrades. What we need to avoid is excessive frequency and long duration of any one event. This data is generally supplied to you as part of your hosting support package. To improve service in this area, look for a web host partner who gives you an uptime guarantee.  If this is mission critical to your firm, build redundancy in drives (e.g., RAID 5), in processors (multi-processor machines), and in machines (clustering).

Server Response Times tell you how quickly the server responded to a request from a site visitor. Every time a site visitor clicks on a link a request is made to the server. We want the server to respond quickly, regardless of load.  This statistic is commonly mined from stress tests in order to determine how response time varies under load. To improve this metric you must look a variety of items, including whether your code set is optimized to deliver the fewest requests possible and to retrieve the data quickly (the latter is often a database optimization issue). Another key factor is the hardware itself — is it up to the job?

Page Download Speeds will tell you, on average, how long it takes for a page to load completely. This is a metric you can benchmark yourself (click and count!) but remember that your analysis will be skewed by your local Internet connection and the browser you use. To get accurate metrics, sample from a variety of locations. The key factors here are the size of the page and the performance of dynamic content (database-driven content) on the page, if any.

Server Errors, Failed Hits and Form Errors are your key error indicators. This data can be found in your log files (or your WebTrends report, as discussed previously). Note that all errors are not created equal and that means you need to look at more than a raw count; you need also to look at the type of error generated. The type of error will help you identify server performance problems.

Form Errors data is critical as this indicates how the forms on your site are performing.  A form that fails a disproportionate percentage of the time can indicate problems with either your code on the form or with some problem in your system architecture.

The last factor mentioned, Effective Bandwidth Per User, tells you how much bandwidth is available to each site visitor and is a good indicator of whether your current hosting set up provides adequate bandwidth to support the site’s traffic. This data can be found by running a server stress test. If you find your bandwidth plateaus at unacceptably low levels, you should be looking at a bandwidth package that gives you guarantees of peak usage availability (bandwidth on demand).

That’s it for this week. Next article we move on to Popularity Metrics.

  • view Part 1 of this Artilce
  • view Part 3 of this Article

Measuring Site Performance (Part 1)

Establishing a set of reliable metrics for measuring the performance of your web site in the real world is a key success factor. In the next few articles, we will explore what can be measured, how to do it, and how to turn that data into some useful intelligence for your business.

Web site metrics can be roughly classed into three categories: performance measures, popularity measures and e-business measures. Today we will examine the first category _ performance measures.

Performance measurement is all about how well your code and hosting setup perform in actual use. Responsiveness and reliability are your goals and achieving those goals will require management of three key variables: the code set, the hardware, and the bandwidth.

In order to assess whether you’ve put together the optimal combination of the three factors, you will need benchmarks and objective indicators of the sites performance. Key indicators include:

  • system uptime
  • server response times
  • page download speeds
  • server errors
  • failed hits
  • form failures
  • effective bandwidth per user

The technophiles will wax rhapsodical about terms like throughput and latency, and certainly they are key technical indicators, but from a management perspective, the metrics are more understandable when expressed in the terms given above. That is, when expressed in terms that relate to the actual end-user experience.

Important sources of data include your Web Trends reports (or other log files), Web Check, server stress tests, and third-party services like Keynote. Server log files give you insight into server errors, form performance and the like, but will tell you nothing about download speeds, uptime, or effective bandwidth.

The industry standard for log file analysis is Web Trends (www.netiq.com/webtrends/). The Web Trends products take your log file data, organise it into useful categories, and display it in graphical formats that can be interpreted by laypeople. To get Web Trends, contact your web host, as they likely offer the service for a monthly fee. It is very affordable _ I have seen it offered for as little as US$2 per month.

Alternatively, if you own your own server, you will need to purchase a license or subscribe to Web Trends’ web-based service.

For more technical statistics, you will have to seek out the assistance of someone with the IT skills to run the tests and interpret the results. The WebCheck system is offered by CompuWare (www.compuware.com) and is a good benchmarking tool for checking a variety of data relating to the integrity of your site. It will help identify slow pages and errors in your link structures. The reports also give basic recommendations for handling problems.

CompuWare licenses aren’t cheap and as a result few firms outside of the IT arena maintain them. Try contacting your IT vendor about WebCheck, as they are likely have a licence for testing purposes.

Web server stress testing tools are another invaluable aid to measuring your performance. A server stress test will simulate loads on your server and provide analysis of effective response times, errors, and bandwidth per user. These tools tend to be quite technical but are excellent for assessing the robustness and scalability of your site. If you are planning a promotional campaign or a web-based event, a server stress tool allows you to simulate in advance load scenarios in order that you may determine whether your site is up to the job.

This year’s Superbowl provided a great lesson on the necessity of projecting loads and testing server capacity in advance of major events. Cadillac, Philip Morris and Universal Pictures were among the 17 advertisers who premiered new ads during the SuperBowl. Unfortunately for the companies, their commercials were too successful _ traffic to their web sites jumped dramatically after the commercials aired and their sites slowed to a crawl, becoming basically unavailable for the duration of the game.

In contrast, Sony, McDonalds and Levi Strauss all anticipated the load spikes and their sites remained accessible throughout the game. No executive ever wants to hear the phrase “your site’s down again,” so test it before the game!

A number of firms specialise in providing independent testing results for sites. Perhaps the best known is Keynote. The company has a large network of testing facilities globally and as a result they are able to produce snapshots of site performance and send you alerts when performance levels fall below a certain point. For a quick look at the metrics, try this link to Keynote indices:www.keynote.com/solutions/solutionspmperformanceindicestpl.html.

In the next Article I will expand on this topic as we delve further into measuring success online.

Tools for Tracking Buzz on Twitter

Twitter is one of the most popular expressions of the social media boom. Though a very limited “micro-blogging” format — where speakers are limited to messages that are no more than 140 characters in length — it has become extremely popular, particularly with early adopters and people who spend a great deal of time online. As someone interested in marketing, I find Twitter to be a way to tap into the buzz from social-minded, technologically-friendly Internet users. Gauging that buzz can be a challenge.

In the course of preparing this year’s Open Source CMS Market Share Report, I’ve spent a fair amount of time researching Twitter as a means of judging buzz and popularity. As a result of that I have bookmarked quite a few tools that I found useful. In this post I share what I’ve found — perhaps it will save you some time.

There are 23 tools here, organized topically as follows:



The message data contained at Twitter.com can be analyzed directly through either search of the Tweets or by looking at the use of hashtags, that is, tags users have associated with particular Tweets.

Twitter Search


You should be aware of the default Twitter search functionality, as it does show basic trending information, but more importantly, you can tailor your searches to extract trend data by using either the advanced search interface or the Twitter Search Operators in your queries. While this may not produce the prettiest charts and graphs, the tools allows you to search by date, user, location, sentiment and other variables, thereby giving you the best control over the raw data set — direct from the source, as it were. The results display allows you to filter by language. Use the RSS feed to save your query and keep up with the Tweets as they occur. One disappointing aspect of this tool: it lacks result counts, export, and charting of the result set.



This site accumulates a directory of hashtags and lets you search for particular tags and track their activity for the previous 30 days. The default homepage allows you to view the most popular hashtags for the past day, week or month. One handy option allows you to get an RSS feed for a hashtag.


Want to find out what’s buzzing in a particular place? Geo-tracking of Twitter users is one way to do it.

Nearby Tweets


Geo-location filtering for Tweets. Want to find out every Tweet coming from within 100 miles of New York City or Paris or Hong Kong? You can do it here.



This application is more distraction than serious research but it is most certainly good fun. It takes Tweets and maps them on the world map as they happen.


When you have to know what’s happening across the Twitter-verse as it happens.



Monitor up to three terms or hashtags in real time. Great for tracking breaking stories and top Twitter trends



A real time Twitter monitoring tool. Twendz shows you the most recent Tweets on a topic, along with the most popular subtopics. It does very little, but the ability to isolate the incoming Tweets by sentiment is somewhat useful (thought not terribly accurate!)


When one user retweets a message from another user, the message is assumed to have some special value — at least to the user that retweeted it. When a message is retweeted many times by many users, you have a trend worthy of further examination.



Find out who and/or what is being re-tweeted now.

Retweet Radar


This is a limited tool that shows you a tag cloud related to the most popular topics on Twitter, judged by what is being retweeted. An archive lets you view the most recent days’ activity.


There is a wide selection of tools for tracking what’s hot — both in terms of topics and URLs.



A tool for tracking backlinks — in this case, links posted on Twitter. The system is able to identify both raw and compressed URLs. An advanced search option lets you filter by user and date. Useful options allow you to receive the information remotely, either by setting up e-mail alerts or by dragging the search bookmarklet to your browser toolbar. Perfect for tracking reputation or getting clues to the reach of the URLs you have tweeted. For a variation on this service, see this group’s other website service: backtype.)

Breaking Tweets


This site aggregates news, photos and other current events type data from Tweets. Stories are categorized by region and topic. It is searchable and you can extract the data into an RSS feed.



Set up email alerts for Tweets – sort of a Google Alerts-type service for Twitter. The free version sends either daily or hourly updates for up to 10 alert topics; the premium version costs $20 a month, gives you up to 200 alert topics and a 15 minute alert window. The tool includes a useful set of filters, allowing you to narrow searches by place, attitude, or speaker. Limited Boolean searches are also supported. Alerts can also be retrieved via RSS. (Note: At the time this was written the system was suffering from failures.)



A finger on the pulse of the Twitter masses. TweetingTrends tracks top Twitter trends and then uses Twitter to deliver the results. Follow their Twitter account to receive notifications via Twitter each time a new topic trending on Twitter enters the top 10.



Tracks the top URLs being discussed on Twitter. This is perhaps the most complete of the various URL tracking services. It goes so far as to organize popular topics into channels and presents the information in a variety of ways — including for those who want to grab a hold of the information firehose — a live Tweet stream. The system supports searching as well. Tweetmeme maintains a large number of different Twitter accounts, enabling you to follow particular channels for the most popular stories.



This cool one page website enables you to type in a word, phrase, URL or hashtag and get a snapshot of the reach of that term or phrase. It does this by finding all mentions, then reporting back to you the sources and the reach achieved by that user when they post a message. The system calculates total viewers and views. The tool is useful and interesting, but to get the most from it you will need to use the Twitter Search Operators in your query. Unfortunately there are significant limitations to the tool. The free version is limited to only 50 tweets. You can, for a fee of US$ 20, order a more complete report listing up to 1500 tweets across the last 7 days.



Twist lets you track trends in Twitter mentions on a daily, weekly or monthly basis. Output is provided in graphical form, as shown below, along with a list of the most recent items. Charts can be embedded in your local website.


> above: Twist stats showing activity for search terms\ across 30 days.



This site requires you to log in with your Twitter I.D. Once you have logged in you can monitor trends by tag cloud, or by searching for trends. The options allow you to graph activity on one or more topics by last 6 hours, 1 day or 3 day increments, as shown below.


> above: Part of the Twitscoop display.

Twitter Power Search


This basic tool displays the Top Twitter Trends by day, week, or in near real time. You also have the option to enter the term of your choice to see Tweets in real time.



Another URL popularity tool, this time showing the most popular URLs of the moment. What distinguishes this tool is that it also displays thumbnails of the most popular pictures and videos.



Tracks most popular URLs on Twitter. Shows by default a list of the top URLs, but you can also search for the terms or URLs you wish to track. They maintain a Twitter site that automatically displays the current top 3 most popular URLs.


Several tools allow you to assess the reach and influence of particular users. You can also gain insight into their interests and various activity levels.



A great way to run numbers of Twitter user influence. Use the tool to graph Twitter activity by user name. Enter the Twitter username and the system will graph a wide variety about the user’s Twitter posting activity. Stats include, frequency, time, retweets, tool used, and aggregate daily and weekly totals. Graph output quality, as seen below, is also a step above many of the other tools on this list. Note if the user has protected their updates, the tool will not work — the user’s Tweets must be publicly accessible.


> above: Some of the many graphs produced by TweetStats.



By default this search tool shows you what’s hot right now, but the best feature here is the ability to run a search and get results displayed both by the authority of the poster and by most recent. Great way to find out who is talking about a topic. They also maintain a Twitter account that automatically Tweets the most popular trends.

Twitter Analyzer


Twitter Analyzer claims to be “the most advanced Twitter analytic system in the world.” While it is certainly powerful and slick (see below), it is focused purely on users, rather than on topics, trends, etc. Great if you are looking to assess a user’s influence, or want to find out more about a particular user’s Twitter usage patterns.

Twitter Analyzer

> above: Part of the Twitter Analyzer’s user profile dashboard.



Xefer is a Twitter activity monitor. Enter a user’s Twitter ID and the system produces a nice little heat map (see below, of the user’s historical activity. The output will also list the user’s replies, with a list of to whom they were directed and the frequency. Note if the user has protected their updates, the tool will not work — the user’s Tweets must be publicly accessible.


> above: xefer profile of a Twitter user.

This article originally appeared on RicShreves.net. Used by permission.