1. Getting to the People
Behind the Keywords
How to Use Semantic Meaning to Identify and Optimise
for Different Intentions
Carmen Mardiros
@carmenmardiros
Sunday, 17 February 13
2. How Many of You Hate
Keyword Analysis?
@carmenmardiros
Sunday, 17 February 13
• Why I hate keyword analysis:
‣ Tedious, grunt work
‣ “Aha” moments get lost, fit in the big picture - Constant danger of drowning in data
• Why I love keyword analysis:
‣ Unparalelled insight into the business and its audience
‣ Voice of Customer at large scale
‣ Search is top channel for most websites so insight into a huge chunk of your audience
3. Common Keyword Analysis
Techniques
• Long Tail vs Head
• Branded vs Generic
• SEO: Navigational, Informational,
Transactional
• Highest volume keywords
• Top converting keywords
“So what?”
@carmenmardiros
Sunday, 17 February 13
• No difference between Long/head or there is a difference but that requires more investigation
‣ Meaning and intent trumps Long Tail classification.
• So what if they’re brand aware/unaware? What does this mean about what they’re here to do?
• “Top keywords” leads to obsession over metrics rather than visitor satisfaction and doesn’t say
anything about topic/interest coverage
4. There Must Be a Better Way
Isn’t there...?
@carmenmardiros
Sunday, 17 February 13
5. Classify Keywords by Intent
“where do I find reviews for
communication apps ipad for kids [brandname]”
@carmenmardiros Brand name replaced for client protection
Sunday, 17 February 13
• “I’ll just break the keywords into groups of intent based on words they contain”.
• All discussions I found online suggest you do this.
• It makes sense
6. Classify Keywords by Intent
“where do I find reviews for
communication apps ipad for kids [brandname]”
Brand name
@carmenmardiros Brand name replaced for client protection
Sunday, 17 February 13
7. Classify Keywords by Intent
“where do I find reviews for
communication apps ipad for kids [brandname]”
Core Product Brand name
interest
@carmenmardiros Brand name replaced for client protection
Sunday, 17 February 13
8. Classify Keywords by Intent
Intent to buy
“where do I find reviews for
communication apps ipad for kids [brandname]”
Core Product Brand name
interest
@carmenmardiros Brand name replaced for client protection
Sunday, 17 February 13
9. Classify Keywords by Intent
Informational Intent to buy
“where do I find reviews for
communication apps ipad for kids [brandname]”
Core Product Brand name
Criteria
interest Qualifier
Logic is solid, putting it into practice for 1000‘s keywords is hard
@carmenmardiros Brand name replaced for client protection
Sunday, 17 February 13
• Keywords easily fall into different categories depending on how you look at it.
• Breaking keyword portfolios into logical groups (segments) gets messy and complicated.
10. Step 1
Step Away From The Keywords
and focus on the People
@carmenmardiros
Sunday, 17 February 13
11. People-focused Keyword
Analysis
1. Identify expected search behaviour based on
Intent and Lifecycle stage.
2. Build way to segment data.
3. Establish target behaviour (the goal).
4. Analyse actual behaviour (the reality).
@carmenmardiros
Sunday, 17 February 13
• Use gap between 3 and 4 (what I hope is happening when these Personas visit the website and
what is actually happening to identify areas requiring optimisation)
12. Identify Lifecycle Events that
Mark Crucial Behaviour Shifts
Becomes
Brand Aware
Loyal
Unaware ..................................................... Customer
Becomes
Customer
@carmenmardiros
Sunday, 17 February 13
• These lifecycle events translate across industries and businesses.
• Indicate different interactions with the website, familiarity with the website, different reasons for
visiting.
• Those attitude changes translate into data: recency, frequency, engagement, conversion rates are
different
13. Map Search Behaviour to
Customer Lifecycle Stages
What keywords are these people likely to search for?
Not Yet Customers Existing Customers
?
‘Brand Repeat
HelpMes
Aware’ Buyers
Queries
?
‘Brand
Unaware’ N/A
Queries
‣ What is the Potential Value to become customers?
‣ Brand Unaware = “fresh blood” into the business
@carmenmardiros
Sunday, 17 February 13
• “Not yet customers” likely to form the bulk of the traffic
• Will include a variety of stages, different overlapping types of intent, varying degrees of Potential
Value to become customers
• Potential more important in Brand Unaware as that’s fresh blood, category keywords
14. Does Interest Match What Business Offers?
Potential Value is determined by
Interest match and Interest strength
Not Yet Customers Existing Customers
‘Brand Vague High Potential Repeat
N/A HelpMes
Aware’ Potential Buyers
Queries
‘Brand No or Low Vague High Potential
Unaware’ N/A
Potential Potential
Queries
No Interest Match Interest Match
= =
No potential value How much value?
@carmenmardiros
Sunday, 17 February 13
• Interest match alone is not enough, it’s just the first step
• So what? -
‣ If the No/Low potential segment is high it skews the data (also indicates crap marketing)
‣ Some Potential and High Potential have different profiles and respond to messaging
differently
15. Brand Unaware
Determine Interest Match and Strength
Broad Interest Match: Product Interest Match:
“App for Ipad”
“Language development”
“software”
“Learning difficulties”
Irrelevant
Stumblers
Definite
Possible Convertibles
Convertibles
@carmenmardiros
Sunday, 17 February 13
• The point: you have at least 2 subgroups of people based on their potential.You would reasonably
expect them to respond differently to different types of content and to have different
conversion rate.
‣ First query type only: possibly target audience, touches on what the business has to offer but
is not a great fit because it doesn’t express interest in product
‣ Second query type only: irrelevant for the business, not a good match at all.
16. Brand Unaware
Determine Interest Match and Strength
Broad Interest Match: Product Interest Match:
“App for Ipad”
“Language development”
“software”
“Learning difficulties”
Irrelevant
Stumblers
Definite
Possible Convertibles
Convertibles No or Low
Potential
Vague
Broad Interest Match+
Potential
Product Interest Match
=
@carmenmardiros
High Potential
Sunday, 17 February 13
17. Classify the Brand Aware by Potential Value
Navigational queries =
Vague Potential
Unknowns
HelpMes
Definite
Convertibles
Repeat
Buyers
High Potential =
Consideration/Evaluation
Intent to Buy or Try
@carmenmardiros
Sunday, 17 February 13
• No Irrelevant Stumblers, if they used brand name they have at least some interest
• Definite convertibles - evaluation, trial purchase etc
• Unknowns - Branded queries without any additional qualifiers.
‣ You cannot determine their intent from keyphrases used, you can only determine it based on
website interaction.
‣ Their behaviour is likely to be different: more pages seen, quicker browsing as they intended
to navigate a known website to perform a task.
18. What Behaviour Do You Expect
and Target From Each Persona?
Not Yet Customers Existing Customers
‘Brand Unknowns Definite Repeat
N/A HelpMes
Aware’ Convertibles Buyers
Queries
‘Brand Irrelevant Possible Definite
Unaware’ N/A
Stumblers Convertibles Convertibles
Queries
‣ Refine further by proximity to conversion
(Consideration, Intent to Buy, etc)
@carmenmardiros
Sunday, 17 February 13
• Why I split my keyphrases by Brand - Different kinds of personas and potential found in Brand
Aware and Brand Unaware
• Reasonably expect different behaviour from the Possible Convertibles compared to Definite
Convertibles
19. Step 2
How to Apply This to Large
Sets of Data
(the good stuff)
@carmenmardiros
Sunday, 17 February 13
• I tried keyword classification first and I drowned in my segments, time for another approach
20. Early Grunt Work
• Get (clean) list of all keywords, create frequency tables
for all 1, 2, 3 words combos*
• Google Adwords Keyword Tool
• Group them into broad buckets (building blocks):
‣ Brand terms
‣ Interest match and strength - Broad interest, Product interest terms
‣ Proximity to conversion - Qualifiers, Evaluation, Buy/Try terms, etc
‣ Existing customers - HelpMes, Repeat Buyers terms
*Text editor+Regex+Excel or
http://www.hermetic.ch/wfca/wfca.htm (thanks to Charles Meaden for the suggestion)
@carmenmardiros
Sunday, 17 February 13
• Hard work but part of it gets reusable for different clients, industries etc - You’re building up a
reusable framework which you can reuse and improve over time.
21. Create Custom Dimensions
(NextAnalytics)
1. Create copy of Keyword column
@carmenmardiros Brand name blurred for client protection
Sunday, 17 February 13
• My workflow involves NextAnalytics to pull data via the API from Google Analytics
• Their “fix” feature is gold for creating custom dimensions. I use it for any dimension I may want
to apply a higher level grouping for.
22. Create Custom Dimensions
(NextAnalytics)
2. Create Custom Dimensions Rules using Regex
@carmenmardiros
Sunday, 17 February 13
• If column includes one of the predefined terms, the value in that cell gets replaced with
“Product Interest”.
• If there is no match then cell left blank.
23. Create Custom Dimensions
(NextAnalytics)
3. Run Regex Search and Replace on Keyword Columns
@carmenmardiros Brand name blurred for client protection
Sunday, 17 February 13
24. Create Custom Dimensions
(NextAnalytics)
4. Aggregate the Data Using Pivot Tables
Without logic and connections between keyword groups
the above is just a bunch of labels
@carmenmardiros
Sunday, 17 February 13
25. Create Custom Dimensions
(NextAnalytics)
5. Create Conditional Custom Dimensions
IF Broad Interest Match AND Product Interest Match
THEN Definite Convertibles
• I started with how I expected people to use Search
based on Intent and Lifecycle stage
• The data corroborated my theory
@carmenmardiros
Sunday, 17 February 13
26. Create Custom Dimensions
(NextAnalytics)
6. Use Slicers to Peel The Onion Further
IF NOT (Active Use OR Evaluation OR Purchase)
THEN ?
@carmenmardiros Brand name blurred for client protection
Sunday, 17 February 13
• Helps determine other intents you may have missed by excluding the already defined intents
from your data. Slicers and pivot tables make it amazingly easy.
• For Brand Aware queries I keep going until I end up with navigational queries (brand name
without any other qualifiers)
27. Summary
• What search behaviour do you expect
based on intent and lifecycle stage?
• Use tools to apply that logic at large scale,
and improve it over time.
How Do You Go About It?
@carmenmardiros
Sunday, 17 February 13
• Contributions from #measurecamp audience:
‣ @james_cornwall - Add internal search data as well into Persona definition
‣ @yalisassoon - Identify behaviour of Personas and then use machine learning to identify
other visitors with very similar behaviour which may fall into the same category
‣ Validate the assumptions with qualitative Voice of Customer to complete the picture
28. Thank You
Carmen Mardiros
Please get in touch with questions and comments:
carmen@mardiros.net
@carmenmardiros
Sunday, 17 February 13