Day 1 of SES Chicago kicked off with a keynote from Google's Analytics guru - Avinash Kaushik. Not one to mince words Avinash started out by saying that "People who don’t get search have a lame view of the world". Then he asked the question "What the heck is search marketing?", with the answer being that it's more than just the main search engines, and in his view is comprised of 5 main components
- Keyword discovery – if you’re not good at this you’ll suck at everything else
- Keyword Management & Analytics- who bids on what? When?
- Keyword Bidding & optimization – ad groups, keywords & phrases, text & creative, bids, match types, time, position, SEO, multivariate testing, analytics
- Website & landing pages – landing page management, LP optimizations, testing, ‘behavior’ targeting, advanced SEM analysis, analytics
- Business Outcomes – Direct revenue, profit & margin, online, offline, Non revenue impact - You made love to the search engines, what came out of it?
The way to improve your search marketing efforts is to look at what you do well at, and see where you should focus. Then follow the 4 rules:
- Don’t obsess about tools
- Understand the true landscape
- Identify gaps / opportunities
- Execute like crazy
Smarter search and analytics
If you underestimate the amount of data / user experience, you’ll end up focusing on the top head terms. The magic in search is not from the head, it’s in obsessing over the tail. If you have, for example, 50,000 rows of data, you can't physically see it all to compare, so how do you understand how it all fits together? Avinash recommends 4 ways to do just that.
Use logical filters – i.e. show all brand keywords where there’s an above average ROI. Look at what's working well for you and replicate it, look at what's not working well for you and fix it.
Use an algorithm to eliminate the humans element i.e. look for high bounce rate keywords, find the ‘sucking losers’. When using the regular sort you'll get lost in the weeds (those keywords that have only 1 or 2 visits), so you need to use intelligence. “Weighted sort” is a new feature in GA that shows the most interesting rows of data when sorted, rather than just the raw sort functionality. Then you can focus on these 'interesting' rows.
3. Tag Clouds
Use tag clouds to identify what your site is all about, make sure that you’re writing about lots of things, not focused on 1 or a few items. A site that Avinash recommends to use is Wordle.net. An example he showed of someone doing a good job with their tag cloud was LDS, with their tag cloud showing that they write about a variety of topics that they want to rank for, whereas Blackberry had a tag cloud that showed that they primarily focused on the term Blackberry, not really on anything else.
4. Keyword Trees
The Juice analytics plugin allows you to understand the relationship between keywords. It's a good way to visualize data to see what works & what doesn’t work
"Your competitors are not doing these 4 things, because they’re lame" - Avinash
The top 10 rows of your dashboard never change, that’s why no-one cares about it after a few weeks (for many sites), so you have to figure out how to live life beyond those top 10 rows.
After talking for 10 minutes about how he worked to convince his wife that he was an important person in the industry, he finished out the presentation with a discussion on attribution models.
The first step in attribution modeling is to determine whether you have an issue with attribution. Take a look in your analytics to see how many visits it takes for someone to convert, if the majority of conversions occur in 1-2 visits, then you shouldn't care about attribution. If, however, it takes 9-12 clicks for a conversion, then it's time to either cry or work out the right attribution model.
- Should the first touch point be the one that gets the credit? No, that’s like giving your first girlfriend the credit for marrying your wife.
- Should it be spread equally around all touch points? Life is not a participation event, so no.
- The MCU (Make Crap Up) model randomly (or seemingly randomly) attributes percentages across touch points…
- The decaying credit model spreads the credit backwards across the touch points (which may result in earlier ones receive 0% credit), potentially looking at the timeframe.
- Other models include giving the last touch point a much higher percentage, then applying a decaying model backwards across other touch points
While you may decide to go with any of these models, Avinash's advice was to "Be thoughtful, skeptical, & objective when applying models".