Has anyone every experimented with the Taguchi ad optimization method for Google Adwords (or email campaigns)? If so, was there a significant increase in performance (clickthrough, conversions, etc.)?
I've been developing something that's similar but not exactly the Taguchi method. (and yes in beta testing of my utility I've seen a 20% boost in profit with a drop in advertising cost). Seems like the Taguchi method has too many variables and would require too large of a data set in an ever changing market place.
I need to brush up on the Taguchi method but you should be able to use it on a small number of variables like 8 or less. If you can use it to determine something about the Adwords algorhythm it would be well worth the effort. I'd look at factors daily budget, content vs. search, top rank vs. lower rank, and maybe some quasi-measure of substance in your ad message (are you presenting steak or sizzle?). I can probably supply a good size dataset if you have a design worked out.
@KernelKlink: Are you integrating directly with the Adwords API to automate your testing? @tbarr60: They have an *interrupter* scale with 7 factors that to help get attention.
I used it to optimze the ctr on the adsense ads on my site. I saw an increase in ctr although not that huge. I guess my site was already optmized
I changed the ad colors: border, background, etc. I used this script that I made for that purpouse http://www.adsense-toolbox.com/color3_1.php I was thinking on making a version of the script to test the ad position on the page but there seemed to be little interest on the subject from the webmaster community so I dropped the idea. I think that most people still don't know about taguchi.
I would think that position would be one of the most critical factors to test. I have a colleague who attended the Taguchi 3-day seminar. Apparently, they have special techniques, specifically for AdWords in place - Unfortunately, I do not have the inside scoop on that one.
Was the seminar put on by the Taguchi organization? I learned Taguchi method while involved in manufacturing. To see it applied to online marketing is quite a jump from welding, bending, milling and drilling but it is design of experiments so can be applied to anything.
There is a 2-year-old Cringely article on Taguchi here: http://www.pbs.org/cringely/pulpit/pulpit20030925.html I'd love to find more info about it, as applied to AdWords et al.
Aside from some oft-disputed-by-western-statisticians principals (i.e. "noise factor"), there's really nothing special about Taguchi versus fractional factorial design. Unless you have more than ~12 total combinations of creative/copy/text/color/whatever, just test each unique combination and see what gives you the best results. ("results" as in higher profit, lower cost, better CTR, better leadflow, etc.) Nothing too hard about that. For many combinations of changes, Taguchi gives you some leverage to test fewer things and still be able to determine the main effects of each type of change (i.e. color vs. text). Another way to look at it is that you can test the same stuff with a smaller sample size (= fewer visitors to your site, etc.) But what you gain in being able to get reliable results from a smaller sample, you lose in resolution of test results... if you have the traffic to test all combinations (non-Taguchi), you should always do that instead. LC
Is part of the benefit of Taguchi the test arrays they developed? "fractional factorial design"? In 30 words or less.
Yes, in few words , if you want to test combinations of things -- for example: 3 x font sizes, 2 x background color, 4 x text (copy/words) -- then you have two choices: (a) full factorial, which tests every combination of things (in this example, 3 x 2 x 4 = 24 total "sets" of things to test), or (b) fractional factorial, which tests a subset of things (you could test 6 total sets and still get some useful data.) When it comes to advertising/marketing testing, Taguchi really just boils down to fractional factorial, aka "Taguchi array". There's nothing special about a Taguchi array other than a cleverly-designed "sweet spot" of which sets can be tested to still yield some coarse results. There are other elements of a "true" Taguchi design experiments, but they deal with noise factors and other social/societal considerations, and western statisticians generally do not accept them as applicable beyond very specific manufacturing/industrial concerns. HTH LC
Ok, now I understand the concept of "fraction factorial design". Thank-you for the excellent (and brief explanation). The Excel spreadsheet Taguchi issues does have a "noise factor" as a part of the calculations. They originally were selling software called MR2 that did a 12-set test. But I think they stopped doing that in lieu of the more valuable consulting and seminar business model.
Offermatica and Optimost are two startups that are, I believe, offering technology for multivariate A/B testing & which employs the Taguchi method. Our clients have used both, and most have seen worthwhile improvements from them. There are other companies as well (Kefta is one), but Offermatica & Optimost seem to be the industry leaders.
I wonder if the Taguchi or a related methodology could be effectively applied to a series of websites to test which factors are most significant regarding SEO for the major search engines. Does anyone have any thoughts on this?
A few others to check out are SiteSpect, as well as ATG (the latter has a special module for conducting A/B testing within e-commerce sites.) HTH LC88
quite interesting! however because i dont have that amount of money, i just follow what amazon, google or yahoo are doing depending on which site. If they change the color background, put brighter/smaller/bigger photos, increase/reduce/change size/font or ads, etc i just do it as well. After all, who has more money than these 3 websites to pay on reseach to determine what s the best way to attract people and the way a website should look? just follow a famous website that is doing what you r doing, and follow their path just adding small new details. this way works for me.