Richard Lawrence walks through a process to measure the impact of your LinkedIn advertising even when you have zero clicks using the LinkedIn API and machine learning.
5. LinkedIn has already proven itself to have
value for advertisers
@richlawre
40%
of B2B
advertisers say
they get quality
leads from
LinkedIn
https://sproutsocial.com/insights/linkedin-statistics/
6. But is the full value being captured?
https://www.linkedin.com/pulse/linkedin-ad-benchmarks-aj-wilcox
0.08%
is the average
click through
rate to a website
@richlawre
7. Wouldn’t it be great if we could link your sales
with LinkedIn ads impressions?
@richlawre
8. I’m going to walk you through a process
Analysing the
impact on
sales
Analysing the
impact on
website behaviour
@richlawre
Extracting
LinkedIn ads
data
9. Starting at extracting the data
Analysing the
impact on
sales
Analysing the
impact on
website behaviour
@richlawre
Extracting
LinkedIn ads
data
10. You can get some insights about who has seen
& engaged with ads from the UI
@richlawre
11. But this is nowhere near the whole story
@richlawre
All
businesses
Ad level
insight
Daily
breakdown
15. But I have also provided a code start on Colab
@richlawre
bit.ly/apistart
16. And here’s some tips
● Just tell LinkedIn you’re using it for an
internal analytics app
@richlawre
17. And here’s some tips
● Application took around a day to confirm
@richlawre
18. And here’s some tips
● Go here to generate token once you have
access:
https://www.linkedin.com/developers/tool
s/oauth
@richlawre
19. This is the kind of thing you can get
@richlawre
20. Now onto connecting LinkedIn activity with
sales
Analysing the
impact on
sales
Analysing the
impact on
website behaviour
@richlawre
Extracting
LinkedIn ads
data
21. You need your data from your CRM
platform
@richlawre
22. A key problem is that the company names
probably don’t match up
@richlawre
Great Company Great Company
Group PLC
23. You can use a fuzzy matching algorithm to
solve this
Great Company Great Company
Group PLC
Match @richlawre
24. Here’s a fuzzy matching algorithm you can use
@richlawre
bit.ly/fuzzyalgo
25. You can then answer some important
questions
@richlawre
Do video ad impressions have a greater
impact on sales?
26. We use machine learning to answer these
questions
@richlawre
LinkedIn & Sales
Data
Machine Learning
Model
Learns from the
data how to predict
a sale
Extracts the
features it uses to
predict
Enlightened
Person!
Ah, it finds video
ad impressions to
be an important
predictor
27. But the data needs to be in the right format
@richlawre
28. You can use ChatGPT’s Code Interpreter to
guide you through training the model
@richlawre
I want to determining which LinkedIn [Campaigns/Ads/Ad formats etc] are best
associated with sales by training a machine learning model and then using
feature importance.
29. How to use ChatGPT’s Code Interpreter
@richlawre
Available in
paid version
Switch
on
Choose to
use
30. But it often runs out of memory, so run the
code in Google Colab
@richlawre
31. Here is also a useful guide to feature
importance using XGBoost (good ML algo)
@richlawre
bit.ly/xgboost-features
32. This will give you which features were most
important for predicting sales
Feature Importance for Sales Prediction
Video-based ad
impressions
Text-based ad
impressions
Image-based ad
impressions
@richlawre
33. Conducting a time-based analysis is also useful
Sale happened
Campaign Impressions vs. Sales
Impressions
@richlawre
34. You can also look connect ads data with
website behaviour - but why?
Analysing the
impact on
sales
Analysing the
impact on
website behaviour
@richlawre
Extracting
LinkedIn ads
data
35. Well - what if you don’t have many sales?
Low volume of data
will lead to a poor
prediction model
@richlawre
36. Going further up the funnel gives you more data
to use for the model
Visit
Conversion
Sale
@richlawre
37. To do this you need to figure out which
accounts are active on your website
@richlawre
YOUR WEBSITE
IP ENRICHMENT
SERVICE
Great Company PLC
Great
Company
38. Now you can match up LinkedIn ad impressions
and website activity for accounts
@richlawre
YOUR WEBSITE
Match
IP ENRICHMENT
SERVICE
Great
Company
Great
Company
Great
Company PLC
39. There are a number of services to choose from
for IP enrichment
@richlawre
40. Clearbit has a free tier for testing (but only up to
30 lookups)
@richlawre
bit.ly/clearbit-free
41. So you’re now finding out what predicts website
conversions rather than sales
Do video ad impressions have a greater
impact on sales?
Do video ad impressions have a greater
impact on website visits or conversions?
@richlawre
42. But which website conversions would work best
for measuring LinkedIn ads success?
@richlawre
YOUR WEBSITE
43. Well, you are often targeting specific job titles
from organisations on LinkedIn
@richlawre
Developers Directors
44. Website behaviour segments can help determine
the success of your job title targeting at scale
@richlawre
Behaviour Segment Potential Job Titles
Spends lots of time in your
documentation
Developers
Looks at pricing & high level
content
Directors
YOUR WEBSITE
45. To get behaviour segments you need to cluster
users by the content they visit to find patterns
@richlawre
DIRECTORS
DEVELOPERS
OTHERS
NO.
VISITS
TO
URL
1
NO. VISITS TO URL 2
46. I did a presentation about this at Measurefest in
2021
bit.ly/measurefest-21
@richlawre
47. You can also get some starter code for clustering
from my website
@richlawre
bit.ly/pycluster
50. Now you can do the same analysis, but using the
website conversions
@richlawre
51. ChatGPT‘s Code Interpreter can help like before
Visit
happened
@richlawre
I want to determining which LinkedIn [Campaigns/Ads/Ad formats etc] are best
associated with website conversions by training a machine learning model and
then using feature importance.
52. How can you take things even further?
@richlawre
53. Use the results of the analysis to inform other
channels
SALES
Can you use this
content on
other channels?
@richlawre