Day 1 Keynote – #SESLondon
I’m sat at the front of the auditorium, in the rather comfortable press row, with @jaamit. It’s nice not to have my laptop burn a hole in my lap as we have tables. Ooooh!
Today’s keynote is from Avinash Kaushik @avinashkaushik, Google’s Analytics Evangelist and author of Web Analytics: An hour a Day.
Introduced by Global VP of Incisive Media as a “true force of nature” Avinash jokes “it can only go downhill from here”.
We’re starting by talking about the origins of the book. Avinash was surprised when asked to do it, that people would pay for information that is essentially free. However the books have been a great joy to produce (though a tough process), and of course very successful.
Avinash puts the success of the books down to passion. “It’s amazing that you can acheive success doing something you are passionate about”. Today, the hat he is wearning is the hat of someone who is enourmously passionate about the web and the underlying data. We’re doing this as a four act play.
Act I
Why are we not extremely hooked into data? Because we’re seeing data with our eyes. We’re therefore limited in a way, by field of vision. We naturally tend to focus on the keywords we see right in front of us.
Analytics allows us to apply conditions that repurpose the way we view our data, and filter our data. Which of course makes it much more manageable.
Applying “winning” filters such as goal value etc, can also be balanced by applying losing filters, which can aslo be valuable.
Avinash is also a fan of tag clouds. I’ve yet to meet anyone who is, so this is rather refreshing. I guess because this gives a graphic conceptualisation of keyword patterns, we can’t get from lists. Pretty obvious really, if we have one big, fat, dominating KW we’ve got a pretty unbalanced and narrow means to pull in visitors.
In summary, looking at data in visual ways, adds a layer of value you can’t get from excel.
Looking at relationships between KW using relational mapping tools, showing paths through keywords, related keywords is very valuable. Step away fro the top 10!
Act II – Outcomes Baby!
We come up against a lot of bitching in this industry. getting clients and brands to understand web data and the importance of prioritising web data.
Example: Convincing the boss.
Avinash only gets time to blog at midnight (kids… I hear ya). Of course the wife is telling him to go to bed. In order to convince the boss, Avinash tries the “I’m kind of a big deal” agrument. Such a piece of data is of course pretty meaningless to ‘the boss’.
Second attempt… shows her a visitor by geo graph which shows thousands of visitors including Somalian pirates. But still she says, “go to bed”.
This is not making a difference because this data is not of consequence to the person we’re trying to convince.
Data and the way we present data must be of impact and consequence to the audience. As an example quantifying “goals” does not have to be about conversions or actions. Avinash has classified blog post types into goals, and therefore specific impressions on posts, add to a cumulative illustration of performance against objectives.
Lesson: Understand “what is success here?” and create goals accordingly.
We’re looking at tripadvisor.com to understand macro and micro conversions. Macro would be bookings, micro would be ad clicks, newsletter sign-ups, etc. With a site like tripadvisor.com many people choose this site because of the value of reviews; the newsletters, the brochures. All this additional rich content and actions around them can and should be measured.
Act III – Think Looooooooooooooong Tail
We obsess about brand terms, but in reality people look for things in very different ways. “If you’re not getting half of your traffic or more through search, then you are doing something wrong!”
Why? Because of the looooooooong tail.
13 Head terms driving 5000 odd visits on his blog; however…
26,124 Tail terms are driving 35,534.
That’s worth something right?
Yes! The loooooong tail is full of “impression virigins” as they haven’t made up their minds. They are doing research and informational queries. GET THEM FIRST.
Example: HMV
Search “HMV” and we get 5 page 1 entries.
If I’m an “impression virgin” how do I get to HMV? Example, I search for “princess and the frog” – where is HMV now?
What about “best ipod offers”? Nope. “Michael Jackson”? Nope!
HMV is not appearing in search for ‘tail’ terms i.e. products and services. HMV are invisible for tail terms. How will our impression virgins find them?
Why such a problem? Because the media that people who are “brand aware” are using, is going away. How do we adapt to this and how do we learn? Google’s New Keyword Ideas Tool takes site data and content, and show keywords and also the PAGES on which these KW are most prevalant. Also the most important data surfaced here is Ad/Search share, which shows (of total queries), the percentage share of voice to your site.
Moving onto “Intelligent Attribution”. Attribution that is, of visit origin and resulting behaviour. Looking at number of visits to purchase, not just traffic source origin. Finding out that visit 6 – 14, might be the most likely to convert visit number bracket, is incredibly useful.
Why do most purchases occur after 15 visits for example? Is this even a problem?
This is kind of like the old display “post-impression” conversion problem, where most recent referral source wins! Avinash liken this to giving his ex-girlfriend the credit because he married his wife. It’s a bit cack-handed really.
Avinash jokes his favourite model is “lets make crap up”. (Normally favoured by agencies.) There’s a lot of random and inferred behavior applied as ‘rule” based on all sorts of crap. E.g. why did you credit the email channel as 20%? A – “Because we have a very large email marketing team”.
Avoid this arbitrary approach to weighting value like the plague.
Instead, a sensible model could look at e.g.
75% credit to last click
Remaining 25% of credit
1. Apply ‘decay’ functin for 3 – 5 clicks back
2. All within last 7 – 30 days.
Complexities – nuance, impulse, specificity of business organisation.
Lesson: Be thoughtful, be skeptical, be objective.
Attribution analysis is a very real problem right now, but Avinash tells us… “incremental attribution analysis models are the future.”
Moving on…
“The biggest gift of the internet is the ability to be proven wrong, FAST!”
Working with a particular brand, they had a particular opinion about a particular behaviour. This was purely an opinion and the behaviour must be tested.
Hypothesis, test, hypothesis, test. The bedrock of all analytics must not be forgotten. (That was a total paraphrase by the way.)
Q: from the audience: attribution with decay theory – what about the difficulties with word of mouth?
A: Sometimes you simply can’t. However, use additional strategies using many different ways of collecting data, more qualitative e.g. exit surveys, market research, focus groups.
Avinash has a couple of posts he recommends on his blog.