Dan Kuster held a workshop at General Assembly Boston on how machine learning is changing -- and improving -- the way digital marketers do their jobs.
Overview:
"Machine learning allows a marketer to target people based on an actual understanding of their interests, habits, and personality, rather than typical demographic data. To get more concrete here, machine learning lets you say: I want to target people that have posted a picture of a guitar in the last three months, or: I want to target people with the INTP personality type that posted something angry about Bernie Sanders recently.
It also allows marketers to look strategically at the content they use to engage their audience and reflect on what works and what doesn't work in a scientific way. If you make 30 posts with very different engagement rates, you can use your own intuition, but then also scientifically vet the wording of your message to get a sense ahead of time about how engaging it may be."
7. many channels under the digital marketing umbrella
Image credit: eperales via flickr.com; modified by cropping
Video(YouTube)
Social m
edia
(Twitter, Fb, Insta)
Search ads (Google)
Email(Mailchimp)
Content(news,blogs)
e-com
m
erce
(Am
azon)Mobile apps (iOS, Android)
SMS/text messaging (telecoms)
Chat/instant messaging (Slack)
8. OK, what ISN’T
digital marketing?
Broadcast media
(TV, radio)
Most media-on-disc
Word of mouth
Books
Printed newspaper
Signs & Billboards
15. Two kinds of data here
1. Demographics about you. Users often don’t know what data exist on server
(definitely not “permission marketing”). Need scalable ways to store and
access data about each user:
• Metadata
• Friends, Likes, browsing history, stuff in cart, …
• Social graph
2. What you’re saying. Users post their own content. We know what we post,
and we do it on purpose. We want to engage, productively, with others. To do
this at scale, need a gazillion people evaluating content, or machine learning.
• Images
• Text
17. Pro
• Facebook takes care of
administration
• Facebook controls everything
• Facebook already has tons of
data, you don’t need to gather
• Can be effective way to reach
audience and build brand
• Can be cost-efficient compared
to other channels
• Can give you more data
about your campaigns
• You aren’t in control, Facebook is
• Your reach might be limited by your
own network
• “Who you know” is an after-effect of
interactions in the real world;
Correlation != intention
• Can be creepy to users?
• Usually not a leading indicator
• Interests change faster than your social
graph. (Shopping for a house -> buy a
house, not buying another. Might need
a plumber though…)
Con
18. machine learning in a nutshell
Content = images + text
deep neural networks
DATA
FEATURES
REP
MODEL
PREDICTIONS
informative pieces of things
feature vectors
search / similarity, sentiment, emotion,
19. 1. Machine learning
2. Digital marketing
3. How it works: using ML to automate digital marketing
4. Examples: real-world use cases
38. Example:
user-generated content
Image credit: Brigitewear International; www.shop-brigite.com, modified with crop and pixelation
You have a brand
Your brand has an identity
(Disney vs. Calvin Klein)
Your audience might
have different sensibilities
than you do, about
what is appropriate
for your brand
Use ML to filter out the
inappropriate content
39. What digital channels require
content filtering?
• Social media (Instagram, Facebook, Twitter, …)
• e-Commerce (Ebay, Amazon, …)
• Web content (News, Forums, blogs)
• Mobile (Apple, Google)
Anywhere users upload
content that everyone
can see.
40. …also brands using
social media to engage
directly with the public?
Doh! Don’t let this
happen to your brand.
41. …also brands using
social media to engage
directly with the public?
Doh! Don’t let this
happen to your brand.