Sramana: As you were going through this process, you likely found there was a multivariate optimization algorithm that could apply. How do you feed that algorithm? How do you decide which keywords should even be considered in the beginning?
Mike Mothner: That stage of the process involves creating a net of keywords. We call that keyword research. We look at all the products that you sell, and all the ways that people could search for those products. Wine.com is a client of ours that sells 3,000 different bottles of wine. The keyword research for Wine.com is massive because of all the wineries they sell, types of wines they sell, and so on. They can also sell only to certain states, so it becomes a massively complex campaign.
To help us create that keyword net, we have built tools to help us find all of the different permutations of keywords. A keyword is rarely one word, it is more often a key phrase. In the wine example, a user may type in ‘buy cabernet wine,’ or ‘cabernet wine reviews.’ Those are all variations on the keyword ‘cabernet.’ During our keyword research process, we will generate thousands of permutations and phrases just for cabernet. ‘Cabernet wine review’ may turn out to be far more valuable than ‘cabernet wine.’
We have to go through our process always assuming that we do not know the answer. We think we need to test and find data to support the conclusion. We don’t like hypothesis because we have the ability to test and find the right answer. We like to know that ‘cabernet review’ is 30% more valuable as a key phrase than ‘cabernet wine’. We can actually answer that question.
Sramana: How do you do get to that answer?
Mike Mothner: By answering on both. We spend $100 on clicks for each keyword or key phrase. We can then say that the second key phrase brought in $300 more revenue. As a result we would lower our bid on the less effective keyword and raise it on the keyword that performed better. The ultimate goal is sales maximization given a fixed input. It is a fairly complex challenge to solve.
Sramana: You test various campaigns to see which keywords yield as a first step. I recently spoke with 1000Bulbs.com, and their keyword list has 14,000 phrases. How can you test 14,000 phrases?
Mike Mothner: We would have 14,000 keywords and look at the daily, weekly, and monthly data on those keywords to determine the relative effectiveness of keywords. When you look at 14,000 keywords, you are dealing with a long tail concept. The idea behind the long tail is that the internet brought about mass customization.
The long tail theory is that after the 10 or so best sellers, the heads, and the body of the next 100 or so, then the rest is the long tail which will have far greater volume. If you can effectively reach that long tail by customizing the message then you don’t have to do a Super Bowl ad, you can do a unique ad for a unique person. Tiny niche businesses can now do well by reaching a tiny, unique audience. The closest thing to a long tail effect is the Yellow Pages.
1000Bulbs.com is a great example of the long tail because it would be very challenging to be an industry leader in the light bulb category without the Internet. People would just get their light bulbs at the store and probably pay a bit more to get them there. Now there is a company that can only sell light bulbs and do a lot of business by reaching all of the long tail customers with unique light bulb needs.