CLICK
TO
CONTINUE
DATA DRIVEN
INTERNAL
LINKING WITH
BOTIFY
JAMES BARDSLEY
1
ENTERPRISE DIGITAL MARKETING & ANALYTICS |
JAMES
JAMES BARDSLEY
HEAD OF DATA & ENGINEERING
HI MY NAME IS JAMES
● LEADER OF THE IMWT NERD
SQUAD
● BASED IN MELBOURNE
● BEER, RUNNING, PETTING CATS
● WAS UNABLE TO GROW A BEARD
UNTIL THE AGE OF 36
CLICK
TO
CONTINUE
2
DIGITAL MARKETING & ANALYTICS |
ENTERPRISE DIGITAL MARKETING & ANALYTICS |
I’M ALSO A
DEVELOPER
3
DIGITAL MARKETING & ANALYTICS |
ENTERPRISE DIGITAL MARKETING & ANALYTICS |
HOW BOTIFY TOOK
MY JOB
HOW BOTIFY TOOK
MY JOB MADE MY
JOB MORE
INTERESTING
4
DIGITAL MARKETING & ANALYTICS |
ENTERPRISE DIGITAL MARKETING & ANALYTICS |
DATA DRIVEN
INTERNAL LINKING
OPTIMISATION…
WITH BOTIFY AS
THE DATA
PROVIDER
5
I like to make
your life easier
so I brought you
this data!
Wow, thanks
Botify! I love
you!
DIGITAL MARKETING & ANALYTICS |
ENTERPRISE DIGITAL MARKETING & ANALYTICS |
3 DIFFERENT USE
CASES…
6
1.Identify quick wins and tidyup
work.
2.Post hoc selection of test and
baseline groups after launching
experiments.
3.Data supply for machine learning
models.
ENTERPRISE DIGITAL MARKETING & ANALYTICS | 7
1. FINDING
QUICK WINS
DIGITAL MARKETING & ANALYTICS |
ENTERPRISE DIGITAL MARKETING & ANALYTICS |
HAVING WELL
TARGETED
INTERNAL LINKS IS
POINTLESS IF THE
SITE IS BROKEN
8
DIGITAL MARKETING & ANALYTICS |
ENTERPRISE DIGITAL MARKETING & ANALYTICS |
FINDING QUICK
WINS
9
The Botify Action Board tells us where internal linking
needs to be tightened up before we begin to work on
targeting links better.
DIGITAL MARKETING & ANALYTICS |
ENTERPRISE DIGITAL MARKETING & ANALYTICS |
EXAMPLE: CUTTING
DOWN HUGE SITES
10
Discoveries:
● 280M pages known by Google (that’s 11 pages per
every human in Australia).
● With only 100M crawled, and less than 1% of the pages
crawled producing traffic in a given period.
Actions:
● We dropped 60% of the 'crap' pages for Google.
● Increased 'the good URLs' by 5M by redirecting crawl
resources.
Results:
● 47% increase in traffic.
● ADDED 1.6M NON-BRANDED VISITS PER MONTH.
DIGITAL MARKETING & ANALYTICS |
ENTERPRISE DIGITAL MARKETING & ANALYTICS |
IDENTIFYING AND
FIXING THESE ISSUES
GIVES AS A SOLID
FOUNDATION TO
BUILD FUN STUFF
ON…
11
ENTERPRISE DIGITAL MARKETING & ANALYTICS | 12
2. POST-HOC
SELECTION
TESTS AND
BASELINES
DIGITAL MARKETING & ANALYTICS |
ENTERPRISE DIGITAL MARKETING & ANALYTICS |
YOU SHOULD
ALWAYS DESIGN
YOUR EXPERIMENTS
WELL BEFORE YOU
LAUNCH THEM…
13
DIGITAL MARKETING & ANALYTICS |
ENTERPRISE DIGITAL MARKETING & ANALYTICS |
BUT SOMETIMES
THINGS GO AWRY…
14
DIGITAL MARKETING & ANALYTICS |
ENTERPRISE DIGITAL MARKETING & ANALYTICS |
OR SOMETIMES YOU
FIND YOURSELF IN
THE MIDST OF AN
ACCIDENTAL TEST
15
“One sometimes finds, what one is not
looking for. When I woke up just after
dawn on September 28, 1928, I
certainly didn't plan to revolutionize all
medicine by discovering the world's
first antibiotic, or bacteria killer. But I
suppose that was exactly what I did.”
— Alexander Fleming
DIGITAL MARKETING & ANALYTICS |
ENTERPRISE DIGITAL MARKETING & ANALYTICS |
WE’VE EXPERIENCED
BOTH OF THESE
SCENARIOS. BOTIFY
TO THE RESCUE!
16
DIGITAL MARKETING & ANALYTICS |
ENTERPRISE DIGITAL MARKETING & ANALYTICS |
SPLITTING
TEST/BASELINE
GROUPS FOR
INTERNAL LINKING
17
Filter ourselves a test group
Then filter ourselves a control group
DIGITAL MARKETING & ANALYTICS |
ENTERPRISE DIGITAL MARKETING & ANALYTICS |
WE CAN THEN RUN
CAUSAL IMPACT
ANALYSES ON THESE
RESULTS
19
…the positive effect observed during the intervention period is
statistically significant and unlikely to be due to random fluctuations. It
should be noted, however, that the question of whether this increase also
bears substantive significance can only be answered by comparing the
absolute effect (2.74) to the original goal of the underlying intervention.
The probability of obtaining this effect by chance is very small (Bayesian
one-sided tail-area probability p = 0.002). This means the causal effect can
be considered statistically significant.
ENTERPRISE DIGITAL MARKETING & ANALYTICS | 20
3. DATA
SUPPLY FOR
ML MODELS
DIGITAL MARKETING & ANALYTICS |
ENTERPRISE DIGITAL MARKETING & ANALYTICS |
PROBLEM: CAN WE
IDENTIFY THE PAGES
WHICH WILL
RESPOND BEST TO
RECEIVING MORE
INTERNAL LINKS?
21
DIGITAL MARKETING & ANALYTICS |
ENTERPRISE DIGITAL MARKETING & ANALYTICS |
OUR APPROACH:
TRAIN A MACHINE
LEARNING MODEL TO
PREDICT IF A PAGE
WILL RANK IN THE TOP
10 SEARCH RESULTS
FOR A GIVEN
KEYWORD
22
DIGITAL MARKETING & ANALYTICS |
ENTERPRISE DIGITAL MARKETING & ANALYTICS |
THEN: TEST IF THE
PREDICTION CHANGES
IF WE TEST THE PAGE
WITH DIFFERENT
PAGERANKS AND
INTERNAL LINK
COUNTS
23
Is Page X predicted
to rank?
No!
Is Page X predicted
to rank if we
increase link count
by 500? Yes!
DIGITAL MARKETING & ANALYTICS |
ENTERPRISE DIGITAL MARKETING & ANALYTICS |
DOING THIS WITH
ANY LEVEL OF
ACCURACY? THAT’S A
LOT OF DATA
24
DIGITAL MARKETING & ANALYTICS |
ENTERPRISE DIGITAL MARKETING & ANALYTICS |
BOTIFY WERE ABLE
TO PROVIDE US
HUGE DUMPS OF
DATA - BOTH FOR
ON PAGE AND
INTERNAL LINKS
25
● Full URL
● No. of Internal Inlinks (From Distinct
URLs)
● Inlinks Percentile
● No. of Incoming Redirects
● Internal Pagerank
● Internal Pagerank Position
● LCP: Agg
● FID: Agg
● CLS: Agg
● No. of Similar Indexable Pages (Score ≥
75%)
● No. of Words (Total)
● …And 240 more!
DIGITAL MARKETING & ANALYTICS |
ENTERPRISE DIGITAL MARKETING & ANALYTICS |
WE THEN
COMBINED THIS
WITH DATA FROM
OTHER SOURCES
26
Delicious Data Salad
On-site details CTR
Bounce rate External backlinks
DIGITAL MARKETING & ANALYTICS |
ENTERPRISE DIGITAL MARKETING & ANALYTICS |
THE MODEL WAS
THEN TRAINED
WITH THIS DATA
AND WAS QUITE
SUCCESSFUL IN
PREDICTING PAGE
PERFORMANCE!
27
DIGITAL MARKETING & ANALYTICS |
ENTERPRISE DIGITAL MARKETING & ANALYTICS |
AND WE USED THAT
TO PLAN LINK
DISTRIBUTION
28
Is Page X
predicted to
rank?
No!
Is Page X
predicted to
rank if we
increase link
count by 500?
Yes!
Is Page Y
predicted to
rank?
Is Page Y still
predicted to
rank if we
decrease link
count by 500?
Yes! Yes!
Reassign 500 links from
Page Y to Page X (with a
bunch of logic to ensure
this is not done
destructively and those
links are relevant)
DIGITAL MARKETING & ANALYTICS |
ENTERPRISE DIGITAL MARKETING & ANALYTICS |
LIVE TESTS ON
THIS ARE A
WORK IN
PROGRESS
29
THANKS
FOR LISTENING.
JAMES BARDSLEY
james@inmarketingwetrust.com.au
www.inmarketingwetrust.com.au
30

Data Driven Internal Linking With Botify

  • 1.
  • 2.
    ENTERPRISE DIGITAL MARKETING& ANALYTICS | JAMES JAMES BARDSLEY HEAD OF DATA & ENGINEERING HI MY NAME IS JAMES ● LEADER OF THE IMWT NERD SQUAD ● BASED IN MELBOURNE ● BEER, RUNNING, PETTING CATS ● WAS UNABLE TO GROW A BEARD UNTIL THE AGE OF 36 CLICK TO CONTINUE 2
  • 3.
    DIGITAL MARKETING &ANALYTICS | ENTERPRISE DIGITAL MARKETING & ANALYTICS | I’M ALSO A DEVELOPER 3
  • 4.
    DIGITAL MARKETING &ANALYTICS | ENTERPRISE DIGITAL MARKETING & ANALYTICS | HOW BOTIFY TOOK MY JOB HOW BOTIFY TOOK MY JOB MADE MY JOB MORE INTERESTING 4
  • 5.
    DIGITAL MARKETING &ANALYTICS | ENTERPRISE DIGITAL MARKETING & ANALYTICS | DATA DRIVEN INTERNAL LINKING OPTIMISATION… WITH BOTIFY AS THE DATA PROVIDER 5 I like to make your life easier so I brought you this data! Wow, thanks Botify! I love you!
  • 6.
    DIGITAL MARKETING &ANALYTICS | ENTERPRISE DIGITAL MARKETING & ANALYTICS | 3 DIFFERENT USE CASES… 6 1.Identify quick wins and tidyup work. 2.Post hoc selection of test and baseline groups after launching experiments. 3.Data supply for machine learning models.
  • 7.
    ENTERPRISE DIGITAL MARKETING& ANALYTICS | 7 1. FINDING QUICK WINS
  • 8.
    DIGITAL MARKETING &ANALYTICS | ENTERPRISE DIGITAL MARKETING & ANALYTICS | HAVING WELL TARGETED INTERNAL LINKS IS POINTLESS IF THE SITE IS BROKEN 8
  • 9.
    DIGITAL MARKETING &ANALYTICS | ENTERPRISE DIGITAL MARKETING & ANALYTICS | FINDING QUICK WINS 9 The Botify Action Board tells us where internal linking needs to be tightened up before we begin to work on targeting links better.
  • 10.
    DIGITAL MARKETING &ANALYTICS | ENTERPRISE DIGITAL MARKETING & ANALYTICS | EXAMPLE: CUTTING DOWN HUGE SITES 10 Discoveries: ● 280M pages known by Google (that’s 11 pages per every human in Australia). ● With only 100M crawled, and less than 1% of the pages crawled producing traffic in a given period. Actions: ● We dropped 60% of the 'crap' pages for Google. ● Increased 'the good URLs' by 5M by redirecting crawl resources. Results: ● 47% increase in traffic. ● ADDED 1.6M NON-BRANDED VISITS PER MONTH.
  • 11.
    DIGITAL MARKETING &ANALYTICS | ENTERPRISE DIGITAL MARKETING & ANALYTICS | IDENTIFYING AND FIXING THESE ISSUES GIVES AS A SOLID FOUNDATION TO BUILD FUN STUFF ON… 11
  • 12.
    ENTERPRISE DIGITAL MARKETING& ANALYTICS | 12 2. POST-HOC SELECTION TESTS AND BASELINES
  • 13.
    DIGITAL MARKETING &ANALYTICS | ENTERPRISE DIGITAL MARKETING & ANALYTICS | YOU SHOULD ALWAYS DESIGN YOUR EXPERIMENTS WELL BEFORE YOU LAUNCH THEM… 13
  • 14.
    DIGITAL MARKETING &ANALYTICS | ENTERPRISE DIGITAL MARKETING & ANALYTICS | BUT SOMETIMES THINGS GO AWRY… 14
  • 15.
    DIGITAL MARKETING &ANALYTICS | ENTERPRISE DIGITAL MARKETING & ANALYTICS | OR SOMETIMES YOU FIND YOURSELF IN THE MIDST OF AN ACCIDENTAL TEST 15 “One sometimes finds, what one is not looking for. When I woke up just after dawn on September 28, 1928, I certainly didn't plan to revolutionize all medicine by discovering the world's first antibiotic, or bacteria killer. But I suppose that was exactly what I did.” — Alexander Fleming
  • 16.
    DIGITAL MARKETING &ANALYTICS | ENTERPRISE DIGITAL MARKETING & ANALYTICS | WE’VE EXPERIENCED BOTH OF THESE SCENARIOS. BOTIFY TO THE RESCUE! 16
  • 17.
    DIGITAL MARKETING &ANALYTICS | ENTERPRISE DIGITAL MARKETING & ANALYTICS | SPLITTING TEST/BASELINE GROUPS FOR INTERNAL LINKING 17 Filter ourselves a test group Then filter ourselves a control group
  • 18.
    DIGITAL MARKETING &ANALYTICS | ENTERPRISE DIGITAL MARKETING & ANALYTICS | WE CAN THEN RUN CAUSAL IMPACT ANALYSES ON THESE RESULTS 19 …the positive effect observed during the intervention period is statistically significant and unlikely to be due to random fluctuations. It should be noted, however, that the question of whether this increase also bears substantive significance can only be answered by comparing the absolute effect (2.74) to the original goal of the underlying intervention. The probability of obtaining this effect by chance is very small (Bayesian one-sided tail-area probability p = 0.002). This means the causal effect can be considered statistically significant.
  • 19.
    ENTERPRISE DIGITAL MARKETING& ANALYTICS | 20 3. DATA SUPPLY FOR ML MODELS
  • 20.
    DIGITAL MARKETING &ANALYTICS | ENTERPRISE DIGITAL MARKETING & ANALYTICS | PROBLEM: CAN WE IDENTIFY THE PAGES WHICH WILL RESPOND BEST TO RECEIVING MORE INTERNAL LINKS? 21
  • 21.
    DIGITAL MARKETING &ANALYTICS | ENTERPRISE DIGITAL MARKETING & ANALYTICS | OUR APPROACH: TRAIN A MACHINE LEARNING MODEL TO PREDICT IF A PAGE WILL RANK IN THE TOP 10 SEARCH RESULTS FOR A GIVEN KEYWORD 22
  • 22.
    DIGITAL MARKETING &ANALYTICS | ENTERPRISE DIGITAL MARKETING & ANALYTICS | THEN: TEST IF THE PREDICTION CHANGES IF WE TEST THE PAGE WITH DIFFERENT PAGERANKS AND INTERNAL LINK COUNTS 23 Is Page X predicted to rank? No! Is Page X predicted to rank if we increase link count by 500? Yes!
  • 23.
    DIGITAL MARKETING &ANALYTICS | ENTERPRISE DIGITAL MARKETING & ANALYTICS | DOING THIS WITH ANY LEVEL OF ACCURACY? THAT’S A LOT OF DATA 24
  • 24.
    DIGITAL MARKETING &ANALYTICS | ENTERPRISE DIGITAL MARKETING & ANALYTICS | BOTIFY WERE ABLE TO PROVIDE US HUGE DUMPS OF DATA - BOTH FOR ON PAGE AND INTERNAL LINKS 25 ● Full URL ● No. of Internal Inlinks (From Distinct URLs) ● Inlinks Percentile ● No. of Incoming Redirects ● Internal Pagerank ● Internal Pagerank Position ● LCP: Agg ● FID: Agg ● CLS: Agg ● No. of Similar Indexable Pages (Score ≥ 75%) ● No. of Words (Total) ● …And 240 more!
  • 25.
    DIGITAL MARKETING &ANALYTICS | ENTERPRISE DIGITAL MARKETING & ANALYTICS | WE THEN COMBINED THIS WITH DATA FROM OTHER SOURCES 26 Delicious Data Salad On-site details CTR Bounce rate External backlinks
  • 26.
    DIGITAL MARKETING &ANALYTICS | ENTERPRISE DIGITAL MARKETING & ANALYTICS | THE MODEL WAS THEN TRAINED WITH THIS DATA AND WAS QUITE SUCCESSFUL IN PREDICTING PAGE PERFORMANCE! 27
  • 27.
    DIGITAL MARKETING &ANALYTICS | ENTERPRISE DIGITAL MARKETING & ANALYTICS | AND WE USED THAT TO PLAN LINK DISTRIBUTION 28 Is Page X predicted to rank? No! Is Page X predicted to rank if we increase link count by 500? Yes! Is Page Y predicted to rank? Is Page Y still predicted to rank if we decrease link count by 500? Yes! Yes! Reassign 500 links from Page Y to Page X (with a bunch of logic to ensure this is not done destructively and those links are relevant)
  • 28.
    DIGITAL MARKETING &ANALYTICS | ENTERPRISE DIGITAL MARKETING & ANALYTICS | LIVE TESTS ON THIS ARE A WORK IN PROGRESS 29
  • 29.