Tips & Tricks for Getting Things Done Using Analytics Data
1. Tips & Tricks:
Doing the right
stuff with
Analytics
bit.ly/AnalyticsDone
@OptimiseOrDie @CharlesMeaden
2. “A very common theme that runs through all our
investments is really around data and the merger of
human and machine. How do we process that data in
order to make the world a better place? To make people
happier, to enrich their lives, to provide better products
and services. It doesn’t matter if it’s to enable drug
discovery or trying to make food delivery more efficient,
data runs across almost every one of our companies.”
Jeffrey Housenbold, Softbank
(1 Trillion Dollar fund)
5. WHY ASK US?
• 40 years working with Analytics
• Nearly 500 Analytics setups
• Mistakes are good!
• Failure is a part of analytics work
• Our pain is your gain
@OptimiseOrDie @CharlesMeaden
6. TECHNOLOGY AGNOSTIC
• This isn’t a Google specific talk
• Actionable information
• But not about geeky reports (aww!)
• It’s about Analytics Strategy
• Competitive Advantage
• Employee Productivity
@OptimiseOrDie @CharlesMeaden
7. EXAMPLES ARE REAL
• We care about handling client data
• We can’t always show real data
• We will show you models or concepts
@OptimiseOrDie @CharlesMeaden
8. The Right Stuff:
1. Data Collection (CS)
2. Data Skew (CM)
3. Data Pollution (CS)
4. Data Enrichment (CM)
5. Data Automation (CS)
6. Data Presentation (CM)
7. Data Ownership (CS)
8. Training & Investment (CS)
10. “Out of the entire lifecycle of
analytics measurement, there is
only one place guaranteed to
pollute everywhere else and that’s
the data collection layer.”
Simo Ahava
12. 480 client configurations
95% had high priority issues
100% had medium priority
3/480 had no tracking issues
<1% had solid analytics
Analysis of Site Audits
@OptimiseOrDie
14. @OptimiseOrDie
You wouldn’t do out of the box for….
• Ecommerce software
• Credit Card Processing backend
• Security systems, Tills, Wifi in retail
• Computer Networking
• Accounting Software
• Stock Management
• Nuclear power station management software...
15. @OptimiseOrDie
Even FREE tools are not FREE of effort:
• FREE simply means that you get a basic system
• It does not mean that it RUNS for free
• A car needs correct setup, oil, fuel and maintenance
• It doesn’t run on air and neither does analytics
• Investment in analytics is #1 weakness in my clients
• How can you race with others, if your dash is broken?
• How can measure self performance if it’s flawed?
• How can your team do their jobs, if they can’t measure too?
17. @OptimiseOrDie
Knackered Tills = Crap Data
• Would they work out of the box?
• What would you need to do?
• Set up Categories
• Set up all the buttons
• Refunds?
• Lunch Money?
• Gift Certificates?
• Supplier cash payouts?
• Till operator signon?
• Cash, Credit, Account payments?
• Fraud detection?
In other words, the till is an empty
vessel until wired into the context of
your business operations and
reporting!
19. @OptimiseOrDie
Get Data Collection Nailed
• Over 50% of setups are ‘plain vanilla’ (GA)
• Even Free Analytics is NOT free – it needs work
• If you screw up collection, you need a Time Machine
• Problems either exist now (Audit catches them) or are introduced (poor
release process for Analytics)
• Always have a TEST and LIVE data view (TEST is for staging/QA/any
non live testing)
• Invest in automation (e.g. link/campaign tracking tools to reduce
tracking errors)
• Invest directly into supporting data collection as part of the project
outcome!
20. Data Collection Summary
Get a regular baseline audit of tracking
Set up TEST & LIVE data views
Provide ‘tracking’ support for all stories / tasks
Test all new changes on staging (before
implementing)
Have a ‘signoff’ process to validate
Listen to Simo about Agile Integration
Don’t do a bolt on – data is part of everything
22. • Where your data is correct, but
needs to cleaned and untangled,
before you perform any analysis
• Time taken to clean the data
means less time for analysis!
• People start to lose trust in the
data if you don’t fix this…
@OptimiseOrDie @CharlesMeaden
What is Skewed data then?
23. SECTION TITLE
SANITY CHECKS FOR DATA HYGIENE
1. Does the data I’m working with make sense to me?
2. Does it make sense to someone else?
3. How clean is the data collection & processing?
4. How much work have I got to fix this?
25. You have one landing page on your site called:
www.website.com/index.html
But in your analytics system, you have three hundred pages like:
www.website.com/index.html?partnerid=123
Why is this a problem?
@OptimiseOrDie @CharlesMeaden
Data Skew – Too Many URL’s
26. • Subdomain / Cross domain tracking
• Payment Gateway / 3rd party redirects
• Campaign Tracking Misconfigured
• Pages with mixed cases (e.g. /Index and /index.html)
@OptimiseOrDie @CharlesMeaden
Data Skew – Common Sources:
27. • Google Analytics shows only the top
50,000 items per table
• Everything else goes into a bucket
marked “(other)”
Our client asked 3 questions:
(1) Are we recording everything?
(2) Are we recording too much data?
(3) How do just record what we need?
• Investigations showed 111 different
query strings appearing in the URLs
@OptimiseOrDie @CharlesMeaden
Data Skew – Too Many URL’s
28. • Working with the client, we determined which
query strings:
1. Were useful for analysing user
behaviour and should be kept
2. Had no analytical value and should be
removed
• 150 Google Analytics filters were configured to
clean and remove the unnecessary query
strings
• Where possible, useful data passed is via the
Data Layer instead of on URLs!
• The end result - a clean and usable analytics
system without page fragmentation! @OptimiseOrDie @CharlesMeaden
Data Skew Filtering Solution
29. Data Skew Summary
Do the page names and events make sense (to someone
else other than you)?
Are you capturing data that you don't need for analysis?
Does your analytics match up with your other systems?
How have you grouped the data in your systems?
How many views on your data exists?
When was the last time you audited your analytics?
Test every single change that you make!
31. Your data contains extra stuff (more data) that shouldn’t be there
A 3rd party is adding the data (bot, spam) or you’ve done something wrong!
@OptimiseOrDie @CharlesMeaden
What is data pollution (vs. skew)?
32. Our Conversion Rate is Dropping:
Prospects
Customer
Logins
Internal
Logins
Dev & Staging
Philippines
Data Pollution Example
Automated
@OptimiseOrDie @CharlesMeaden
34. • Not filtering out 3rd parties (agencies, dev, partners,
employees, contractors)
• Bots and automated services
• Site Performance Tools
• Internal Testing (Dev)
• Internal Monitoring (Dev)
• External Monitoring (Uptime)
• Misconfigured tags
• Double firing tags
@OptimiseOrDie @CharlesMeaden
Common Data Pollution Sources:
35. Data Pollution Summary
Very common problem but until you check, it’s
usually invisible, unless you look carefully
External or internal can screw data hygiene
Impacts trust in data (visible or invisible)
Worst case we’ve seen was 60% of traffic
Corrupts key site metrics (CR, Bounce etc.)
Filter out to ONLY what you explicitly need!
@OptimiseOrDie @CharlesMeaden
37. 37
Plug and Play Analytics?
Data quality isn’t acquired – it’s earned!
Thanks to
@Simoahava
38. SECTION TITLE
• Some ‘Out of the box’ analytics definitions are flawed
• You will always need to tune these to fit how your business,
product or service actually works
• Customising, tweaking and enhancing the vanilla analytics
configuration = enriched data
• What can you do to capture more of the important
interactions and events as part of the user journey?
39. • Client had lots of useful information –
which was added to the query strings on
the page URL
• When they launched their new site, they
implemented analytics with ‘clean urls’
and a vanilla analytics implementation
• All the useful data that used to be there,
was no longer being recorded
@OptimiseOrDie @CharlesMeaden
Data Enrichment – Data Going Missing
40. • We created a plan to explicitly review all
the data required and identified which
stuff was useful
• We then made sure the important
signals that were really needed by the
client, were put back!
• Instead of being on every URL, the data
is pushed to GA 360 using Custom
Dimensions & Events
@OptimiseOrDie @CharlesMeaden
Data Enrichment – Capturing The Useful Stuff
41. • Departure and arrival airports
• Dates of travel
• Size of group
• Hotels viewed
• Whether search used
• Quote created
• Availability for each hotel
• Existing customer or prospect
@OptimiseOrDie @CharlesMeaden
Example Custom Dimensions
44. Data Enrichment Summary
Are you capturing all the key information you need to
analyse the customer journey
Maximise use of your analytics tool to store additional
information – you’ll be glad that you did in a years
time (love note!)
Think about what you need to capture every short and
long term customer goal reached
Investigate how to integrate or blend additional tool
data (e.g. VOC, Session replay) with your analytics tool
@OptimiseOrDie @CharlesMeaden
46. • If it takes longer than 15 mins/week
• AUTOMATE IT
• Do you know how much time is wasted?
• Not without doing user research!
• Do you know the life of data?
• Why don’t you follow it sometime?
• Remove Report Monkey Wastage
Data Automation Foundations:
47. • Removal of manual reporting tasks
• Making it 1 click, auto email or dynamic
• Saving multiple repeat efforts
• Allows for pushback / redefinition
• Reduces ‘ad hoc’ report processing
• Needs OVER resourcing to work
• Won’t save time – saves PRODUCTIVITY
Data Automation Definition:
48. • SQL extracts
• Spreadsheet or Google Sheets pull
• Automation / Reporting / Analysis tools
• Analytics Package Auto Reporting
• Dashboards
• Emailed Reports
• Work back from outcomes!
Data Automation Execution:
49. @OptimiseOrDie
Craig’s Favourite Models
A. 1 Hour Research
B. Bounce Layer
C. Landing & Next Page Path
D. Traditional Funnel
E. Ring Model
F. Horizontal Funnel
G. Top of Funnel Drivers
H. Page Groups
I. Intent Model
J. Multi Goal Groups
K. Temporal or Lifecycle
L. Ecommerce Grid
M. Dimensional Grid Pull
50. Funnel success
Funnel Step 2
Funnel Step 1
Sales
Ecommerce
Behaviour
@OptimiseOrDie
Traffic
Get a funnel for all these:
• New & Returning Visitors
• Channel breakdown
• Country breakdown
• iPhone Models
• Android Models
• Windows Browsers
• Mac Browsers
• Device type & Category
• Landing page(s)
• Content page(s) viewed
• Demographics
• Days to conversion
• Path Length
Want something? Get in touch!
Dimension Grid Pull
51. @OptimiseOrDie
• Rapid Blockage Identification
• Landing Layer Issue
• Site Depth Engagement
• Funnel Loss
• Bugs
• Measurement issues
• This one grid cube on the left
started 1 hour of work that
nailed a 2M per month issue
Dimension Grid Pull (app.profitgrid.io)
54. Grid Tool app.ProfitGrid.io
Download Deck bit.ly/2nySVXd
XDO – Full Article bit.ly/XDOfull
Build your own device lab bit.ly/MobileDeviceLab
Google Analytics Templates bit.ly/XDOTemplates
Mobile Checklist for XDO bit.ly/MobiCheck
Excel Grid and Dimensions bit.ly/TheGridLayout
Feedback & Suggestions sullivac@gmail.com
pieter@zetom.nl
@OptimiseOrDie
4L Grid Resource Pack
55. @OptimiseOrDie
Automating Models and Data Integration:
Dimensions, Metrics, Query Explorer:
Very useful for forming queries and getting to know the API:
https://developers.google.com/analytics/devguides/reporting/core/dimsmets
https://ga-dev-tools.appspot.com/query-explorer/
Guide to getting started with the API:
https://www.optimizesmart.com/how-to-use-google-analytics-api-without-any-coding/
Free Google Sheets Addon:
Google Sheets Addon:
https://developers.google.com/analytics/solutions/google-analytics-spreadsheet-add-on
56. @OptimiseOrDie
Automating Models and Data Integration:
Automating Google Sheets:
https://medium.com/@jev/how-to-import-data-from-google-analytics-into-google-sheets-with-
google-apps-script-5174a10b24d8
ProfitGrid:
Simple tool to pull a multi-dimensional grid from any GA setup:
http://app.profitgrid.io
Analytics Edge:
Very flexible excel tool for pulling API data. Free version available. Bristling with options:
http://www.analyticsedge.com/simply-free/
Supermetrics:
Free GA plugin for Excel and Google Sheets. Multiple data sources available:
https://supermetrics.com/product/supermetrics-data-grabber/
https://supermetrics.com/product/supermetrics-for-google-drive/
57. @OptimiseOrDie
Automating Models and Data Integration:
PowerBI:
Allows you to pull GA data directly - was PowerPivot. If you want a multi-dimensional API pulled
horizontal funnel you can segment on-the-fly, this is what you need. A very powerful way of
viewing and pivoting cubes of GA data:
https://powerpivotpro.com/get-the-software/
Next Analytics:
Free and paid options available:
https://www.nextanalytics.com/excel-google-analytics/
Analytics Canvas
Integration, extracts, multiple accounts, automation and scripting. Paid only.
http://analyticscanvas.com/google-analytics/
Scitylana:
Integration, Hit level datastream, Extracts. Free option available.
https://www.scitylana.com/
58. @OptimiseOrDie
Automating Models and Data Integration:
Integrating GA with R:
http://code.markedmondson.me/googleAnalyticsR/
http://cran.r-project.org/web/packages/RGoogleAnalytics/
https://github.com/skardhamar/rga
http://cran.r-project.org/web/packages/RGA/index.html
R for Analysts:
http://www.eanalytica.com/r-for-web-analysts/
Simo Ahava Recommends:
"I wrote these two a while ago. First is for validating a GA account setup, with focus on
Custom Dims too, and the second is for mass updating Custom Dimensions"
https://chrome.google.com/webstore/detail/google-analytics-
validato/nmjiiaaejkhpegmcpfaehmbijgoilimo?utm_source=permalink
https://chrome.google.com/webstore/detail/google-analytics-custom-
d/ogcaloflfbimfnpkkfpfddocaegdmgkk?utm_source=permalink
59. @OptimiseOrDie
• Learn how to craft useful and reliable models from data directly
• Make your own ‘on the fly’ ‘User’ or ‘Session’ funnels for anything!
• Understand how the Analytics Package data model works
• Understand your data collection – how it works
• Learn about common collection and configuration biases
• Improve Event and Interaction tracking
• Learn how to integrate or transform data sources
• Ability to target/analyse cohorts rather than just ‘average’ or ‘mixed’ traffic
• Automating and iterating models, grids and funnels
Core Skillz & Benefitz
60. @OptimiseOrDie
• Browsers & Devices – how do they work (for analytics)
• Tagging and Javascript – how it works (and tool proficiency)
• Filters and Regex – for quickly hacking or grouping models
• Data Model – familiarity with how GA collects and stores data
• Advanced Segments – a stepping stone to mastering API pulls
• Modelling – imposing your abstraction upon a dataset
• Audit & Configuration – knowing exactly what’s broken or reliable
• Visualisation – Presenting actionable, useful and simple truths
• Automation – API pulls, Integration, GTM, Machine Learning
Core Skills – MUST have
61. @OptimiseOrDie
Automated Ecommerce Models
A. Customer Segmentation
B. New vs. Existing customer
C. Basket analysis (various)
D. Returns analysis
E. Site Search analysis
F. Boston Grid Quadrant
G. Profit Analysis
H. Yield Analysis
I. Product Interaction Layer analysis
J. Out of stock analysis
K. First purchase / anchors
L. Discount impact model
M. Cart abandonment analysis
N. Live chat or textual analysis
O. Merchandising reporting (various)
62. Data Automation Summary
To save, you have to spend
One hour of automation = Days saved
Increases Data Literacy and ‘art of the possible’
You don’t need to know how to code
Or create SQL queries
If you can use Excel, you can Automate
Restructure analytics work around removing
duplicated or senseless work
90/10% monkey/thinking vs. 10/90
65. SECTION TITLE
ASK THESE 3 QUESTIONS WHEN PRESENTING
1. Can people clearly see surfaced and actionable issues?
2. Do people know what actions they can or should take?
3. What does your presented data cause your audience to think?
66. • Ecommerce client had suspicions
that their site search tool wasn’t
working effectively
• The person responsible for the
tool also managed 4 other tools
• Data was presented as a table
containing thousands of rows
@OptimiseOrDie @CharlesMeaden
Data Presentation – High Cognitive Load
69. Data Presentation Summary
Find out what jobs people actually do and then figure out the
information they need to do their job better (or just ask
them)
Don’t assume that everyone needs to know everything
The insight should jump out at you, along with either a
followup question or a concrete action
Watch – “How to Present to Get Results” by Lea Pica
Buy - Storytelling with Data: A Data Visualization Guide for
Business Professionals
72. Monthly reports which lack relevance, are rife with generic suggestions unsupported by research
within the context of your business and simply regurgitate last month’s points, failing to show
any new actionable data.
Thanks to
@Simoahava
73. Ugly and ineffective hacks to get tracking working at the last minute.
A lack of coordination or standards for how business critical tracking should be delivered. No
data layer or cross silo approach.
Thanks to
@Simoahava
74. Analytics changes are given low priority and deployed infrequently
Fix Funnel
Tracking
Thanks to
@Simoahava
Deploy
new
funnel
75. Analytics tracking is often ‘added later’ or seen as a ‘Bolt On’ to existing project work because
ownership is unevenly distributed!
77. How to Integrate Analytics & Agile
Agile Analytics:
www.youtube.com/watch?v=yIJp9s46CF8
Meaningful Data:
www.youtube.com/watch?v=rMrB0bKdOtc
79. Data Ownership
A complicated area but vital
Dedicated analytics inside project teams
No Analytics or Report Monkey Silos
Watch the Simo talks on Agile Integration
Every Project or Task must include Analytics
If it doesn’t need any, then great
Done does not mean Done – without tracking!
@OptimiseOrDie @CharlesMeaden
81. “Nobody ever admits being unable to use
Google Analytics properly. We’d all admit to
being no good at Quantum Physics but when it
comes to analytics, most people either don’t
know or they’re lying. ”
Craig Sullivan
83. @OptimiseOrDie
Train Your Team so they….
• Can be better at their job, through data
• Know how everything works in analytics
• Get data out with less work
• Achieve self servicing for data needs
• Automate repetitive tasks
• Get more proficient at finding value
• Data Noodling!
A Ferrari mechanic without engine training?
85. @OptimiseOrDie
Invest in Analytics
Even if tools are free, you should:
• Get an Independent Audit
• Invest in a proper release architecture (QA/DEV/STAGING)
• Make tracking part of every project
• Don’t just tread water – improve tracking
• Put >5% of dev time & analytics budget on EXTENDING
• Things like -> Data Collection, Training, Fixing Stuff,
Automation, Enrichment, Tracking Tools
• Validate training – through research (UX)
• Write a love note to yourself…
86. @OptimiseOrDie
Write a Love Note to your Future Self:
Hey Craig,
All we are is dust in the wind dude. You’re like my future self, so I had
to tell you. I thought it would be most excellent to give you clean
data, so I fixed it, like all the things.
You’ve got like least 3 months of clean stuff and like reports and stuff
like totally makes sense now! Whoa!
Love you, Dude!
Craig from the Past, Dude
87. Investment & Training
Analytics Strategy needs Training Strategy
Training people does not mean they leave
Add 5-10% dev & software budget!
Don’t just tread water – continual improvement
Over investing leads to automation (explain)
Automation leads to standardisation (explain)
Standardisation leads to Productivity (explain)
Hiring tips – ask us in Q&A
@OptimiseOrDie @CharlesMeaden
89. 480 client configurations
95% had high priority issues
100% had medium priority
3/480 had no tracking issues
<1% had solid analytics
The Dark Side
@OptimiseOrDie
92. THANK YOU!
• WRITE THAT LOVE NOTE!
• WANT AN AUDIT?
• SKILLS TRANSFER?
• GET IN TOUCH!
CHARLES : bit.ly/2HeHAt6
CRAIG : linkd.in/pvrg14
Optimising the
Optimisation
@OptimiseOrDie
@CharlesMeaden
26TH MARCH 2019
GOOGLE CONVERSIONS, DUBLIN
bit.ly/AnalyticsDone
Editor's Notes
This is the title of my talk today.
Wonderful picture, isn’t it?
It’s from an IBM advert from 1951 and is a great piece of work, especially the copywriting. The whole message here is “Buying an IBM computer gets you the same power as 150 extra engineers”. And not a feature in sight – the trick is they’re not selling the computer, they’re selling what the computer will do for your business and your life.
And what am I talking about today? Well the fact that most split tests being run these days are just bullshit – the slide rules don’t add up for a lot of companies.
Many C level execs I’ve spoken to complain about the variability of return or success on this kind of testing.
There’s a reason for this -
This is what we’ll cover today in the talk – what the big problem is that I’m trying to solve and some basic groundwork you should do before you start. I also show you how to pull the data for your company with a 15 minute technique and how to use this to test the right browsers and devices on your website or product.
I’ll wrap up with showing you some examples and how to completely automate everything we’ve covered today, so nobody whines later on ;-)
Not everyone will necessary know that the data is skewed or what steps need to be taken to clean the data
Then looking at the analytics system that you’re working with
Your data may look like this piece of data – shattered into tens, hundreds or thousands of pieces of data
The URL’s are available via the Google Analytics API, but not everyone has access to this
Not everyone will necessary know that the data is skewed or what steps need to be taken to clean the data
Query strings in the URL’s not pretty, but it worked as they had a process to extract then
The analytics team hadn’t been involved in the site rebuild
Is there such a thing as plug and play analytics?
Data enrichment is where we go beyond the standard out of the box analytics implementations and make your analytics systems a lot more powerful
Analytics are flawed because they are essentially a one size fits all model
Query strings in the URL’s not pretty, but it worked as they had a process to extract then
The analytics team hadn’t been involved in the site rebuild
Client is on 360 so made full use of the 200 custom dimensions
Events
Think about what data you need to capture both your short term and long term goals
No point in spending time and effort analysing your data, if it ends up looking like this
When data is presented, we ask 3 questions
On a weekly basis, this would contain thousands of terms
The results show was use to determine which action to take
Anything less than 15 result can indicate a stock issue
Created a taxonomy to classify the search terms and create useful grouping
Split the searches into successful and failed searches
Used the taxonomies to show types, materials
Low stock queries are where less than 5 items were returned