Digital Advertising and Attribution Flashcards
• Flipping the funnel
o PAST:
Aw, Cons, Pref, Action, Loyal, Advocacy
TV, DM, product test / compare, purchase, reward points, advocate
o TODAY:
WoM (advocacy) is where they hear about your brand, Online research, reviews/ratings), preference social networks (ask the twitterverse; anyone have a recommend on…?),
Purchasing: more online and mobile (but 90% is still brick and mortar), but that brick and mortar is how people learn about and decide on products and services. Online / Offline attribution is a tricky problem.
Loyalty: Friending, signing up for an email
Advocacy: reviews, ratings, likes, shares, buzz, WoM
Brand marketing vs. direct response is different
Direct Response: : have an add, click ad, buy. Hotel Ad vs. Coke ad (don’t click and buy, but get brand / awareness)
Shouldn’t stop doing TV and radio, but stop
Doing Print
GDPR & CCPA
and CA consumer privacy act Jan 1, 2020: General Data Protection Regulation EU (25May18)
1 Theme of Sinal
Differentiate correlation from causation! test, test, test!
Largest Digital Marketing Channels as of 2019?
o Search & Display are the largest
Fastest growing Digital Marketing Channels as of 2019?
- Social and mobile are the fastest growing
o No email marketing, can just use AB testing, etc.
o Display and online Video are projected to grow faster through 2021
Attribution
Connecting “what I did” to “what I saw happen” - Causal Inference
Optimize
Re-allocate marketing dollars to optimize some metric (sales growth, ROI, etc.)
Channel
Display Advertising, Search Advertising, Social Advertising, Mobile Advertising
Optimization within channel
Search keywords, make search channel perform better different ads, keywords, experimenting, etc.
Optimization across Channels
Move allocation across channels because performance changes or cost changes.
Marketing Analytics & Optimization Flow
Allocation –> Deployment –> Consumer Reponses & Bus. Outcomes
back to the beginning: Marketing Actions (Surrounded by Market Conditions & Competitor Activities) –> Attribution (modeling Causal Inferences) –> Optimization (ROI Measurement, Optimization) –> Allocation…
Integrated Digital Marketing Flow
MEDIA
Strategy -> Content Creation (copy to videos to blog posts to tweets to FB posts, etc.) ->
CHANNEL
Display Advertising, Search, Social, Mobile Advertising
OPTIMIZATION
Within Channel Optimization
Across Channel Optimization (MTA is ideal)
Cookies
Cookies: First thing that happens when a customer hits a website is they’re cookied
Piece of code that identifies you, provides info to the website about you (operating system, browser, where you are, then 3rd pary data: affinity, demos, browsing history)
This is the material that allows targeting.
Publisher
Website, mobile site, etc. with ad inventory, spaces to place ads
How are digital ads delivered?
o 1) Customer visits site
o 2) Publisher drops a cookie
o 3) NYT has direct sale of premium ads (top left and right and sometimes intersticials); you pay a premium for this
o 4) Their’s other inventory left over: provided to an ad network (e.g., admob)
Ad network has tranches of inventory
Sells some blocks of this inventory to advertisers as well
o 5) What’s left over will be given to a supply side platform like rubicon
o 6) Lowest quality goes directly from publisher to the supply side platform
o 7) Supply side platforms make all this availale to an ad exchange
o 8) On advertising side, they have creative / ads they ad to a trading desk, thr trading desk will send to demand side platform which will provide real time bidding, etc.
o 9) Left over ad inventory from advertisiers is also sent directly to the demand side platrom.
o 10) Ad exchange double click puts up the spot on the publishers
o 11) The e platform puts up the bids for the advertisers
12) Placement of an ad takes about 10 milliseconds; all that is the Real Time Bidding
Advertisers
Want to buy inventory (space) to advertise on
Ad Networks Purpose
- Aggregate ad inventory across multiple sites
* To achieve scale and reach
Ad Network Process
• Forecast inventory across publisher sites in advance (month
ahead)
• Deals with advertisers to sell inventory (in advance)
• Deliver advertisers content as ads in sold inventory
Ad Network Types
• Rep Networks: Represent select publishers, often exclusively
as an ad outsourcer for those publishers
• Vertical Ad Networks: Represent publisher sectors (“fashion”)
• Targeted Ad Networks: Sell audience segments
(demographic, behavioral)
RTB
Real time bidding network; customer arrives to site, bidding and placement happens in milliseconds
Advertiser vs. publisher pricing preference
Advertiser: CPC; performance not impressions.
Publisher pricing preference: CPM; cost per impressions, make an ad someone will click on, that’s on you.
CPM
CPM (“Cost-per-Mille” or “Thousand”)
• Cost per Thousand Impressions
• Inventory based pricing, not performance based pricing
• Premium Inventory, Premium Publishers, Premium Price
CPC
CPC (“Cost-per-Click”)
• Publisher only gets paid if user clicks on an ad
• Performance based pricing, Good for Marketers but difficult to
negotiate with premium publishers
• Lower Quality Inventory
CPA
CPA (“Cost-per-Action”)
• Payout on conversion (Affiliate Marketers operate on this model)
• Effective CPA (eCPA): Cost/Conversions (cost per conversion…
> profit on the conversion?)
eCPA
Effective CPA, eCPA takes into account cost vs. profit of conversion. VERY RARE
% Ad Revenue Performance vs. CPM pricing model
Performance (CPC, CPA) is now 65%.
Targeting
Process for determining how you bid as an advertiser
Who is this person?
Demos, interests, affluence, retargeting (have they visited my website)
Build a predictive model that predicts customers that are likely to act on an ad (click or purchase)
Conflict of advertiser and targeting agency goals
What is the advertiser goal? Get people to act
* What is targeter goal (firm I hire to target)? Find people most likely to act after being shown the ad. Targeter must find those who when shown the ad have a higher chance of buying the sweater. Conversion rate is the performance they must show you. But, you need to measure causal conversions (causal effect of showing the ad).
If I know someone has a 100% chance of acting, do we want to advertise to them? NO
Great way to tell between correlation and causation: A/B testing
Targeting Models
Targeting Models (from targeting agencies): Take into account behavior, demos, social network and location variables Precision and recall (coverage of target high likelihood customers) used to score
What is the average click through rate of a display ad?
.05%
So you MUST get good at targeting
Retargeting
Retargeting:
When an advertiser follows you after visiting their site
One item in the cookie is what site did you visit?
2% of visitors purchase on the first visit, retargeting visits the other 98% for them to purchase
What should your retargeting window be? Specific to clients.
Much more effective than display ads; 10x (0.7% vs. .7%)
But, “lift” is difficult to prove because they already have brand affinity, intent, and other activities are happening. So test it!
Affiliate Marketing
Why does Kayak point you to their competitors?
• It gets ad revenue
Promo codes
• Retailmenot 71% when you type coupon in Google
• You go there, you press the coupon, it opens the Amazon tab and it drops a cookie on the browser and tells Amazon we sent this customer, send me 4% commission.
• But did they send the customer to AMZN when someone is shopping and just goes to find a coupon
What is the difference between lift and conversion?
- Lift is the causal effect of an ad on a sale. Conversion rate is the percent of people who are shown an ad who purchase.
- It’s not the same thing!
Lift vs. Conversion Easy Explanation
I'm teaching a class • Conversion rate: 100% class students given a flyer as they walk into class • Lift: 0% - They all would’ve been there already.
Different Attribution Methods
LTA, First Touch, Linear (equal across), Time Decay (increasing to last touch), Position based (first and last most, rest linear), Algorithmic (MTA)
2012 use of Attribution Methods
30% FTA, LTA; 11% Algo; looks like it’s changed.
Attribution causal modeling approach launched 2014
High dimensional propensity score matching
Viewable Ad Problem
Viewable Ad Problem: >50% of area of the ad was viewable
• Premium sites: 53%, networks and exchanges: 31%
• FB is only priced on viewable ads
Cross Channel Marketing Problem:
Search and display are compliments; more on display, the better performing your ad dollars in search and vice versa
• Its suboptimal to do one without the other.
Problem
• How does Display affect Search and vis versa? How to manage
the interaction…
Evidence
• Nearly 1/5th of all Search Conversions have seen a Display Ad
• Experiments Show:
• Users exposed to a display ad
conduct 5-25% more campaignrelevant searches. (Papadimitriou et
al)
• Display Ads increase search
conversion and search clicks
(increasing search ad costs)
• $1 investing in Search and Display
returns $1.24 for Display and $1.75
for Search (Kireyev et al 2013)
- Ad Blockers NPR website 1hr show watch it!
Ad Blockers and the Internet of Tomorrow 5Oct15
