What is Testing?
Internet Marketing Dictionary |
Testing is the process of changing the
content of web pages, recording the
results and implementing those changes that improve
The testing process needs to be statistically
valid and takes some discipline. Testing without recording
results is a waste of time, money and effort.
The principles underlying testing are that
changes to your site, and especially copywriting changes, will
have an impact on how well your site converts visitors into
subscribers or customers. Some changes will make your
conversion rates smaller. Some changes will improve conversion
It follows then that making changes and
recording the results will result in an overall improvement in
your site's conversion rate if you delete the poorer performing
versions and stick with the better ones.
If you continue with this process you can get a
page to "evolve" into a highly effective sales tool.
To make testing work properly though you need
to make the tests as unbiased as you can. This means taking out
of the test any factors that may twist the results. For
example, it is rarely sensible to test one page on a Friday and
another on a Saturday and compare the results because search
traffic is different at weekends.
The best way of avoiding problems like this is
to split your web traffic evenly so that it
alternates between two different pages under test. This removes
as many time and location biases as possible. This process is
also known as A/B testing.
To get the best value from test results
you need to apply statistical theory. Statistics can be very
complex but there are some key concepts worth understanding
such as confidence intervals, normal distribution and standard
deviation. The key issue is knowing whether your results are
sufficiently robust for you to be able to make reasonably
accurate predictions based on them.
A simple example illustrates the point.
Suppose you toss a coin and it comes up heads.
Can you assume therefore that it will come up heads every time
you toss it.
The answer is clearly no because you do not
have enough data to make a statistically valid judgement. If
however it had come up heads 45 times in a row you can be much
more confident in your prediction and can assume farly safely
that it is heavily biased.