SPLIT TEST ACCELERATOR
 


Free Services
Free Accelerated Split Testing Course

Split Test Statistical Confidence Tool

Taguchi Advertising Forum

Products
Split Test Accelerator Software

Taguchi Landing Page Consulting (coming soon)

Please see if your question is here before sending Jim an email message (link at bottom of page). If you don't see your question here, feel free to contact Jim.


Split Test Accelerator Frequently Asked Questions



What Is "STA"?

1.   What is the Split Test Accelerator?
2.   How can the Split Test Accelerator help me?
3.   What is split testing?
4.   What is Taguchi testing?
5.   What other kinds of testing are out there?
6.   What are the advantages of Taguchi testing?
7.   How many combinations can you test with the Split Test Accelerator?
8.   How do I learn the art of Accelerated Split Testing?
9.   How can I purchase the Split Test Accelerator?
10.   What are the advantages of STA over Google Website Optimizer?


Requirements

11.   What are the requirements for using the Split Test Accelerator?
12.   How can I find out if my web server meets these requirements?
13.   How many different domains can I use the Split Test Accelerator on?
14.   Do I need to install the Split Test Accelerator on each domain?


What Special Features Does the Split Test Accelerator have?

15.   Does STA use first party cookies?
16.   I get a lot of "junk" traffic from spiders and "hits" sites. Will this dilute my results?
17.   Which combination will a user see if they view the page a second time?
18.   Does STA allow me to see which pages work best for different sources of traffic?
19.   Can I do this by keyword?
20.   Can I do this by referral domain?
21.   Can I do this by custom ad codes?
22.   Can I optimize different things for the same landing page(e.g., optin-rate AND sales conversion rate)?
23.   Can I string tests together along my whole sales funnel?


Setting Up Tests

24.   What's the best way to start testing?
25.   Which test design should I choose?
26.   Do I have to use all the factors in a test?
27.   Do I have to use all the options for a factor?
28.   Can I have more than one test lead to the same "thankyou" page?
29.   What will users see if the server can't access the database?


Video Tutorials

30.   Do you have any videos showing use of the software?







What Is "STA"?


1.   What is the Split Test Accelerator?

Answer: The Split Test Accelerator is a software program that you install on your web server. It allows you to improve your conversion rates through a kind of multi-variate testing called Taguchi testing.   top

Answer: The Split Test Accelerator can help you increase the conversion rate of your landing page, which will increase your profits.   top

Answer: Let's say you want to know which headline will motivate people more to buy your product. You can create two headlines and show half of your visitors one headline, and half the other, and see which headline produces the most sales.   top

Answer: Taguchi testing is a kind of multi-variate split testing. It is a "fractional factorial" multi-variate test. And it uses "orthogonal arrays" to make sure every pair of options shows up an even number of times across your test. This means that you can accelerate the process. There are some trade-offs. You don't get to actually see how every combination fares in the market, but this is offset by the fact that you can project which combination is best, and you can often run your tests 10x faster (or more). Once you get used to designing tests you can see improvments much more quickly than with any other kind of testing.   top

Answer: There is simple split testing where you test one factor at a time. There is a kind of fractional factorial testing where orthogonal arrays are not used. There is also what is called "full factorial" multivariate testing. Google's new tool "Website Optimizer" is a "full factorial" multivariate split tester. With full factorial testing you have fewer combinations in your test, but you show every combination.   top

Answer: Taguchi testing is faster than simple split testing, because you can test multiple factors on your page at the same time, and with the same amount of traffic. So instead of spending a month testing just your headline, you can spend the same month, with that same traffic, and test 10 factors all at the same time (such as your headline, your offer, your price, your style, an image, your first paragraph of copy, your order button, etc.)

Taguchi testing is also faster than full factorial testing, by roughly the same scale.

Taguchi testing gives better statistical results than fractional factorial tests that don't use orthogonal arrays.   top

Answer: The Split Test Accelerator allows you to test up to 4 Million combinations at the same time in the same test. (With the same amount of traffic you need for a simple A/B split test).

By contrast Website Optimizer (as of 7/21/2007) allows only 1000 combinations. And Google will suggest you use over 100 times more traffic to test those 1000 combinations as you will need to test 4 million combinations with STA. Again, there are some trade-offs. You have to learn how to design tests with important factors and wildly different options, or your results will be minimal. But once you learn how to run Taguchi tests, you can improve your pages very quickly.   
top

Answer: You can sign up for my free course. OR (and also?), when you purchase STA, you get a document titled "The Art of Accelerated Split Testing".   top

Answer: You can purchase by clicking here.   top

Answer: Google has a multi-variate testing tool that's pretty good for small tests.  When they first launched it in late 2006, we worried a bit, and thought, "Oh great, now we're competing with Google, and they're giving their tool away for free."  

But after using their tool, we realized that GWO is not really a split test "ACCELERATOR."  It's actually pretty slow if you follow their instructions (slower than A/B split testing in some cases).  

You can make it a little faster if you know what you're doing.

But you've got to realize that Google will be using your tests to collect data for their own purposes.  

As far as they're concerned, the MORE data they can collect per test, the better!  

As far as you're concerned, the LESS data it takes to improve your landing pages the better.

If you're on a tight budget, using GWO is better than not testing at all -- by far.  In fact, if you're not testing at all, and you're not ready to purchase STA, I would encourage you to start running tests with GWO.  However, STA really is a big step up.  Here's why:



  top



Requirements


11.   What are the requirements for using the Split Test Accelerator?

Answer:

  • You need a server that can run PHP
  • You need a server that can run MySQL.
  • Integration with 3rd party shopping carts is NOT supported when the landing page is generated out of a database, or when either the landing page or thankyou page is part of a templated system (such as Smarty). In short, you must be able to paste STA's php snippets on these pages in a straightforward manner.
  • Remote hosting is NOT supported. Your landing page and thankyou pages should be on the same domain. (But you can still use STA on any domain you own)
  • Integration with Smarty or other templating systems is NOT supported.
There are probably workarounds possible with the last three requirements. In fact, I know there are. I've done some myself. But there are just way too many contingencies involved for me to support such workarounds in general.

If you know what you're doing, or have a good programmer, and want to try to integrate STA with other 3rd party solutions yourself, you will have the source code for STA, so feel free to try, but with the usual caveats.

Also, it's very easy to set up a very cheap hosting account for less than $10/month to use as a testing platform, and STA will work flawlessly with most. I would recommend going with a Linux/Apache/MySql/PHP (LAMP) account for best results.
  
top

12.   How can I find out if my web server meets these requirements?

Answer: You can ask your web host. A majority of hosting accounts are Linux/Apache/Mysql/PHP (LAMP--cute, eh?), and can run STA without any additional configuration. If your host uses Windows or another kind of server, you will have to make sure you can run PHP and MySQL on it. In many cases you can, or they can configure it so you can without any hassle.   top

Answer: As long as you own all the domains, you can use it on as many as you want.   top

Answer: Yes, most users must install STA on each domain they use it on. It is possible to set STA up for remote hosting, but unless you have experience setting up databases for remote access, and configuring the software accordingly, you should create a unique installation for each domain. Remote hosting is not currently supported.   top



What Special Features Does the Split Test Accelerator have?


15.   Does STA use first party cookies?

Answer: Yes. This is important. Third party cookies are often blocked, and, if not blocked, are cleared out by anti-spyware programs. First party cookies are accepted (in my testing) by about 95% of users (plus or minus 4%), and this gives you a great base of test data for your tests.   top

Answer: New with STA 4.0 you have the ability to exclude traffic from your test based on ip address, user agent, or domain.

This gives you the ability to make sure your tests are run with "clean" data.  
top

Answer: If the user accepts cookies, she will see the same combination she saw the first time she visited. This is important. This makes sure the right options get credit for each factor. If a single visitor saw multiple combinations it would not be possible to determine which options should get credit in case of a successful outcome.

The roughly 5% of visitors who don't accept cookies will receive a random combination, but they are not part of the test data, so there is no problem with assigning credit to the proper options.   top

Answer: Yes. You can optimize your page for overall robust performance. But if you want to use STA to help you create customized sales funnels for different sources of traffic, you can get a ton of information when you filter your results in any number of ways to get results for subsets of your traffic. This will give you valuable clues for creating different landing pages for different market segments or traffic sources.   top

Answer: Yes. You can see what are the optimal results for each specific keyword.   top

Answer: Yes, the referring domain is stored. You can see if Google traffic converts better than Yahoo!. And you can see which combinations work best for each domain if you want to customize futher.   top

Answer: Yes. You can tag individual ads with a simple tag, and STA will then allow you to see your results for each ad code. See how your different ads convert, and see which landing page combination converts best for each ad.

Note: making firm decisions about which landing page works best for a subset of your traffic will depend on getting enough traffic from each ad to be statistically significant. Also, subsets of the data won't be orthogonal. But it still gives you tons of valuable clues, and can save you from mistakes that keep people from particular market segments from buying.   top

Answer: Yes. You can test up to three different outcomes for each landing page. You can test for clicks to order page, sales, and opt-ins all at the same time. Get clues about how to optimize each part of your process.   top

Answer: Yes, you can have a test running on every page in your funnel all at the same time if you want.

If you set up the biggest test on every page of a 5 page path, you will be testing about a decillion combinations simultaneously. That's a 1 followed by 33 zeroes.  
top



Setting Up Tests


24.   What's the best way to start testing?

Answer: Someone asked me this question recently. This person had several sites that got a lot of traffic. Here was my answer:

----------------------

If I were you, I would start with the T3x4 array, and just pick a couple sites. That way you get to know the system with smaller designs.

If you are like most people, your first attempt will proabably be too timid. That means that the options you design for each factor will probably not be different enough from each other, and you will probably see a small improvement. However, you will also get a feel for how the process works, and what the results look like, how the numbers fluctuate, and so on.

Once you get a test or two under your belt, and you start getting used to designing wildly different options for your test, I would advise you to go after one of the bigger arrays.

The T6333333 is a popular one, and was one of Dr. Kowalick's favorites when I asked him that question. But there are bigger arrays, with the T632x18 being the biggest. If you can fill that test out with good options, you will almost certainly find 4 or 5 really significant changes you can make to improve your outcome.

As for traffic . . . It's funny. People invest in a Taguchi product, because they know it will "accelerate" the testing process, but then they don't believe how little traffic you need for a Taguchi test.

Here's the bottom line. If you design your test well, you need the same amount of traffic for a Taguchi test as you would need for a single split test.

So if you typically use 2000-6000 impressions to conduct an A/B/C split test, you will need the same amount for even the biggest Taguchi test. That's why, as long as you design it well, it really is "the bigger the better".

I would say that 6000 impressions will typically be sufficient. A lot depends on how the variables are behaving, though. If you design a timid test, you might not see much variation, and might need 10,000 impressions to see significant differences, and then the differences will be relatively small. If you design a bold test, with wildly different, good options, then you might need only 2,000 impressions, and the differences will be relatively large.

So the traffic you need depends on how well you design your test. If you design it well, you need less traffic, and you will see much better results. If you are too timid, you will need much more traffic to get much weaker results.  top

Answer: Make sure you choose a design large enough to contain all the factors and options you want to test. Many people start with the T3x4, because they want to learn how to choose good factors, and they think 4 factors is enough to start with. With a good brainstorming session, however, you should have no trouble filling out a larger test -- perhaps up to 20 factors.

For the most part the larger tests can be run with the same amount of traffic as the smaller tests. In practice there is more potential for interaction effects with a larger test, so you might have to run them a bit longer.

Perhaps it's best to first brainstorm with others about what changes you want to try, and then pick the test that will accommodate all your factors.

Eventually you will probably come to have your favorite test design, and will use that most of the time.   top

Answer: No. All factors are optional. You can use a 20 factor test and use only one factor if you want. Some users will want to split test a single idea, but will choose a larger test design, because they want to leave open the option of adding more factors later.

If you add more factors later, you should restart your test, or record the state of the test at the time you added the factor, so you can subtract out the "old" results for the new factor by hand.   top

Answer: Yes! Every option will be rotated, so you have to specify content for all options for every factor that you use.

So the rule is that all factors are optional, but once you include a factor, all options are required.

If you leave an option blank for a live factor, no content will be shown for that factor when that option is on the page. Only do this if you want to test no content for that factor (which is actually often a good idea.)

You may "double up" content in a factor, though. Let's say you only have 4 headlines, but you have 6 options for the headline factor. It's alright (though not ideal) to repeat headlines, as long as you remember to combine the stats by hand for the identical headlines.

A much better idea is to spend a little more time brainstorming so you can have 6 headlines in the test in that case.   top

Answer: Yes. Just stack the thankyou page snippets in any order.   top

Answer: Users are encouraged to specify default code and include it on the page with their factor code snippets. There is a place to specify default code on the project creation page. That way, if your server can't access your database for some reason, your visitor will still see a viable page.   top



Video Tutorials


30.   Do you have any videos showing use of the software?

Answer: Yes, see below. These are getting dated (many new features have been added since I made these), and I'm not a professional video producer, so they aren't high in production value, but they should give you an idea about how to set up tests, and I will be putting up new videos at some point.

Setting Up Your Database With A CPanel Interface -- This video will help you set up your database and configuration file for both the free split test software, and the Split Test Accelerator. If you want to set up the most recent version of the Split Test Accelerator, the configuration file will have a different name, but that shouldn't be a problem.

Using STA with Glyphius

Overview Of Setting Up A Test   top

Do You Have Any Other Questions?

NOTE: If you need a reply, use my contact form (link below). This form is for collecting more questions for the FAQ page.
Are there any questions you have that did not appear above?









To sign up for the free course, click here.

To see the long-form sales letter, click here.

If you have any additional questions, Here is an email contact form.

Thank you.

Sincerely,

Jim Stone, Ph.D.

(c) Copyright 2004-2012 by Jim Stone Enterprises. - All Rights Reserved
20027 13th Drive SE Bothell, WA 98012
Disclaimer¦ Contact ¦ Free Adwords Accelerator Spreadsheet ¦ Free Multivariate Testing Course ¦ Free Confidence Measuring tool ¦ Split Test Accelerator ¦ Blog ¦