Web Metrics Tested

How To Convert Your Metrics Into Smart Marketing Decisions | Part 2, Section 2 (Analysis)


In this section, we move from data to analysis, from theory to practice. We focus on just two key spreadsheets, and two key lists.

How can you keep from miscalculating your ROI — and seriously losing money? (6 Common Mistakes)

Consider this an “intervention.” It is all too easy to fall under the giddy influence of promising (but inflated) numbers, make incorrect marketing assumptions, and subsequently waste vulgar amounts of money.

Here are six mistakes you do not want to make when calculating your ROI:

  1. Do not use average order amounts; use exact order amounts. Campaigns change often.
  2. Double-check your math. Here, in the hallowed halls of MEC, we often make mistakes. The more important the calculation, the more care we take in verifying its accuracy.
  3. Monitor (constantly) the QUALITY of your visits. The sources of traffic from within the pay-per-click engines often change. There are always developing new search partnerships and sometimes a sudden burst of visits is accompanied by a sudden drop in conversion — costs can go up while sales can go down.
  4. Track the annual value of your customers. Remember, (in most cases) you cannot calculate an accurate ROI solely on the first-order. You need to know how many customers are coming back to your site and how much they are spending.
  5. Do not use average cost per click. Divide total sales by total clicks.
  6. Do not estimate your margins. As we saw in earlier examples, this can distort your analysis.

How can you protect yourself against reporting errors caused by your metrics program? (5 Common Problems)

Users beware! No matter how much care you put into your analysis, sometimes the reporting tools skew your conclusions. Here are five potential dangers:

  1. Beware of double-counted sales. At times, your metrics tools will report the same sale twice. Here are some signs of trouble:
    • Watch for two orders at the same price.
    • Watch for two orders from the same IP address.
    • Watch for two orders at the same time.
  2. Beware of misdirected URLs. Sometimes your tracking program will confuse which sales belong to which campaigns. AND sometimes the search engines send your traffic to the wrong landing page. If you notice any sudden changes in your campaign’s performance, go to the “front side” of the engine and click on your own links.
  3. Beware of recording delays. Be certain to have the complete results in for a given period before you calculate that period’s ROI (there is always a lag).

    And here is a related problem experienced by one of our readers:

    “I have used WebTrends for years, and consistently find that it underreports the final interval of a measurement period.”

  4. Beware of double counted clicks. Pay-per-click search engines, such as Google, count those visitors who use the back button and then click again to revisit a site. This is a common action for shoppers who are comparing offerings. But while your tracking software will most often ignore such revisits, the pay-per-click engine will surely count them (and charge you for them).
  5. Beware of Yahoo’s metrics tools. They are not even close to accurate. See this excerpt from a previous report.
    • Yahoo! Store offers one of the best hosted e-commerce platforms on the Internet, but their site metrics are sorely lacking. There are at least seven problems:
    • Yahoo! does not grant access to its log files.
    • The referral link section, entitled “References,” is absolutely unreliable.
    • Yahoo! is an affiliate of Overture, and you cannot determine if your orders are coming from the Yahoo! network or from a separate Overture campaign.

      NOTE: Our research partners report that they are (sometimes) charged a Yahoo! Shopping Commission for orders that came through their pay-per-click campaigns.

    • Yahoo! will not let you place the necessary code on the order confirmation page to set up a simple tracking system.
    • There is no easy way to view, download, and analyze a complete list of your Yahoo! store orders.
    • Yahoo! often displays orders with no referring link and no entry point.
    • Yahoo’s technical support is outsourced and unable to assist you with important questions about your metrics.


It seems appropriate to quote (again) the Marquis: “A Prince who will not undergo the difficulty of understanding must undergo the danger of trusting.”

If your whole concept of web metrics is based on vague understanding of log files, it may be worth your while to invest the next few minutes.

We have endeavored to create a succinct, 688 word course…

How do you understand your log files?

Log files are big and ugly.

A log file is a record of interaction between your web server and a client machine. Every time a client machine connects to your web server, the server writes a line to the log file.

This is why log files are big:

(1) If your server delivers one million pages, (2) if each page is made up of ten files, and (3) if each file is about 20 kilobytes — your server will have to find, read and send 200 terabytes of data.

One terabyte equals one trillion bytes. This is the equivalent of all the starts in our galaxy — multiplied by ten.

This is why log files are ugly:


“2002-06-25 00:07:49 W3SVC201 GET/tabletools/ showprod.cfm DID=6&User_ID=1182745&st=8702&st2=-69508442&st3 =73707243&CATID=5&Object Group_ID=34 200 40930 747 80 Mozilla/4.0+(compatible;+MSIE+6.0;+Windows+NT+5.1) http://www.example.com/file.cfm”

Do you really need to understand your raw log files?


Still, you can learn the basics in the next 180 seconds, and this newfound knowledge could improve your social standing, your annual salary, and your conversational prowess at parties.

Here is a sample extract from the above entry:

2002-06-25 00:07:49 – This is the date and time (Greenwich MeanTime) that the visitor requested a file from your server. – This is the DNS of the person who asked for the file. You can use this number to find their domain name.

GET /tabletools/showprod.cfm – This is the page (file) they requested from your server.

40930 – This tells you the server sent back 40,930 bytes.

&DID through to ID=34 – This (painfully long segment) tells us the user ID, query, etc.

200 – This means the page request was successful. IF the server had been unable to find the file, this number might be at 404.

Mozilla/4.0+ (compatible;+MSIE+6.0;+Windows+NT+5.1) – This shows us the page they were on when they made the file request. (It is the referrer.) If they had typed in the address, instead of clicking on a link, this URL would be replaced by a “-“.

In the Olden days, men of valor attacked these log files with ponderous, dull-edged spreadsheets. But today, there are sharp, efficient software programs to help you cut through the confusion.

Why is it so difficult to interpret your log files?

As you may have deduced from the previous section, it is tough to capture accurate metrics.

But why? Why is it so difficult to get reliable numbers?

There are several reasons. Here are just three:

  1. Cache files – A cache file is (basically) a temporary copy of a file. It can be stored on your ISP’s computer or on your own. In either case, viewing this temporary file is faster than viewing the original version.

    Cache files increase speed. Cache files also confuse web metrics programs.

    When you click on a link, your computer searches your ISP’s cache and then yours to see if a temporary copy already exists. If it does, it will display this copy instead of the original.

    If it displays the copy, it bypasses the original. If it bypasses the original, it also bypasses the log file for the original.

    The net result (pun intended) is that your visit to the page is invisible.

    Some advertisers say up to 75 percent of page views are from cached versions. AOL claims that page view reports may be short by up to 30 percent.

  2. Spiders and Robots – The digital efficiency of a search spider looking at every single page and clicking every single link on your site can generate line after line of deceptive entries in your web log.
  3. IP duplication – We cannot assume that a single IP address equals a single visitor. When you dial into your local ISP, you are assigned in Internet protocol (IP) address. That number is drawn from a pool of numbers maintained by the ISP.

    If someone else dials up, is assigned the same IP address, and goes to the same site, the server logs for that site would indicate that the two of you were the same person.


ProLinkz Link Tracking Script:

Ezine Promotion: Tracking the Results of Your Efforts:

Ezine Readership Measurement Using ProLinkz:

Does a Perfect Web Metrics Tool Exist?, Part 1

Does a Perfect Web Metrics Tool Exist?, Part 2:

How to Interpret Web Metrics:

One Metric Can Tell the Tale: Visitation Frequency:

E-Commerce Metrics: Drowning In Your Own Data:

Metrics Identify Problems, Not Solve Them:

Click-Through Rate, R.I.P.:

E-Newsletter Metrics:

More E-Newsletter Metrics:

Metrics Don’t Replace Marketing Judgment:

Monitoring Visitor Conversion Using WebTrends:

Measuring the Value of Visitors with WebTrends:

Endnotes for the Report

(*1) Marquee of Halifax (1633-1695) “Of Princes,” Political, Moral and Miscellaneous Reflections, 1750.

(*2) This number is still somewhat arbitrary, as the true margin is yet to be fully discerned.

(*3) If averaging your cost per click is dangerous, so is averaging your revenue per order. You cannot assume that the last period’s average order will be the same for this period. The only safe way to calculate this number is to divide the total revenue by the total orders for the period you are measuring.

(*4) This number is an estimate based on the last informal survey taken more than 12 months ago. Still, it is a useful indicator.

::Back to Section 1::

You might also like

Leave A Reply

Your email address will not be published.