ADVERTISING
ANALYTICS AND REAL-
TIME TECHNOLOGY
ISMT E-111
Instructor: Justin Fortier
Teaching Assistant: Vasya Selsov
“Buy Side” “Sell Side”
AGENDA FOR 3/30/21 CLASS
• Yellowdig Participation Badge winner for week of 3/21/21 – 3/27/21
• Course calendar – quick review
• Six final student presentations (live)
• Begin Data Science textbook Chapter 4 (Conversion w/Decision Trees)
“INTRODUCTION TO PROGRAMMATIC ADVERTISING” (D.
KOSORIN)
Introduction
Ch. 1 – Basic Technologies
Ch. 2 – Programmatic Advertising Ecosystem
Ch. 3 – Programmatic Trading
Ch. 4 – Data
Ch. 5 – Mobile Programmatic
Ch. 6 – Current Issues
Ch. 7 – New Formats and Channels
Source: Dominik Kosorin, “Introduction to Programmatic
Advertising”
HOW IS DATA SCIENCE USED IN AD TECH?
• Bidding decisions
• Yes or no?
• How much ($ CPM)?
• If win, which campaign/creative to show first? Full journey?
• Audience recommendation
• Content recommendation
• Audience personas / personalization to optimize business outcomes
• Optimize player outcomes (e.g. views/clicks/shares/surveys)
• Optimize business outcomes (e.g. “conversion” pixels; client’s ROI, VG’s
ROI)
IPA CH. 5: MOBILE PROGRAMMATIC
• Introduction
• Mobile browser cookies
• Data on mobile
• Mobile inventory
• Location-based marketing
• Cross-device
Source: Dominik Kosorin, “Introduction to Programmatic
Advertising”
IPA CH5: MOBILE PROGRAMMATIC: INTRODUCTION
• Programmatic built for desktop, not mobile
• Times are changing – advertisers must follow the eyeballs
• Different experiences even on mobile (app vs. browser)
• Attribution challenges (cookies on mobile browsers, not
apps)
• Better user experience (more relevant, less disruptive)
• More behavioral data for advertisers
Source: Dominik Kosorin, “Introduction to Programmatic
Advertising”
IPA CH5: MOBILE PROGRAMMATIC: BROWSER
COOKIES
• Programmatic designed for desktop browser cookies
• Mobile browsers still allow 1st-party cookies to be set
• Some mobile browsers (Safari) do not allow 3rd-party
cookies
• Other browsers allow cookies to be cleared regularly
• Programmatic still mostly works well on mobile browsers
• Mobile apps, different story
Source: Dominik Kosorin, “Introduction to Programmatic
Advertising”
IPA CH5: MOBILE PROGRAMMATIC: MOBILE APPS
• Data not easily shared between apps, or apps and mobile browsers
• However, Device ID’s used for authentication and tracking
• Versions exist on iOS (IDFA) and Android (AAID)
• Software Development Kits (SDK’s) enable detailed behavioral data
collection within an app
• More personal, behavioral data in apps than mobile browsers (e.g. FB)
• Programmatic moving to within apps, for that reason
• New companies, e.g. Apsalar, Flurry Analytics
Source: Dominik Kosorin, “Introduction to Programmatic
Advertising”
IPA CH5: MOBILE PROGRAMMATIC: MOBILE INVENTORY
• Available on both mobile apps and mobile web:
• Banners
• 300 x 250 most common format on small smartphone screens
• Larger sizes available for tablets
• Native Ads – developed by Facebook, Twitter – blend in; don’t seem like ads
• Interstitials – full-page ads, must be minimized by user to return to
article/game
• Video – Inline playback is common
• Can be served in current environment (app; browser) or within a native
player on a device
• Less disruptive to users
• Allows for playback data collection for advertisers
Source: Dominik Kosorin, “Introduction to Programmatic
Advertising”
IPA CH5: MOBILE PROGRAMMATIC: LOCATION-BASED
MARKETING
• Location data from device GPS, apps, ad impressions (bid request form)
• City/Zip/DMA/Latitude/Longitude – precise targeting
• Some locations command higher winning bid price
• In-store location tracking – via low-energy Bluetooth Beacons, branded apps
• In-store promotions
• Path-to-purchase analysis (offline, in this case)
• Personalized recommendations
• Behavioral Profiling (e.g. locations where you work, live, play) – controversial
(why?)
Source: Dominik Kosorin, “Introduction to Programmatic
Advertising”
IPA CH5: MOBILE PROGRAMMATIC: CROSS-
DEVICE
• Goal: For marketers to track a person’s entire consumer journey
• Personalized advertising
• Device attribution (ROI)
• Data consolidation
• Cross-device frequency capping (“fatigue” in DS textbook)
Source: Dominik Kosorin, “Introduction to Programmatic
Advertising”
IPA CH5: MOBILE PROGRAMMATIC: CROSS-DEVICE,
CONT’D
• Cookies not sufficient (review: why not?)
• Deterministic Approach
• Collect a unique ID (log-in or e-mail address) from user
• More precise, but most websites /apps don’t require sign-in
• Probabilistic Approach
• Based on statistical probability of user being this person
• Big data on devices, browsers, IP addresses, etc. analyzed
• Inferences made for majority of users with unknown ID
• Publisher co-ops share (deterministic) user ID’s
Source: Dominik Kosorin, “Introduction to Programmatic
Advertising”
IPA CH. 6: CURRENT ISSUES IN PROGRAMMATIC
ADVERTISING
• Fraud
• Viewability
• Ad blocking
• Attribution
• Privacy
Source: Dominik Kosorin, “Introduction to Programmatic
Advertising”
IPA CH6: CURRENT ISSUES: FRAUD
• Key question: “Was this ad served to a human?”
• DSP’s could try to game the system to get more advertising dollars
• e.g. overstate impressions, views, clicks, conversions
• Most common: Using Non-Human-Traffic (NHT) “bots” to create false online behaviors
• Available for purchase via ad networks, exchanges
• Other examples of fraud:
• Stacking ads on top of one another
• Using tiny 1x1 pixel frames
• Manipulating domains to overstate inventory value
• Fake cookies dropped onto browsers for attribution (“cookie stuffing”)
• Fake “shell” websites built just for advertising
• Mobile fraud growing: fake app installs (clicks); in-app purchases
• Fake location coordinates (more valuable locations) added to bid requests
Source: Dominik Kosorin, “Introduction to Programmatic
Advertising”
IPA CH6: CURRENT ISSUES: FIGHTING
FRAUD
• IAB’s Trustworthy Accountability Group (TAG)
• Offer direction and insights, not policing
• Most publishers dislike fraud (it devalues legitimate impressions)
• Still, some publishers willingly participate in fraud (e.g. price arbitrage – purchase
cheap NHT, sell to advertisers for large profits)
• Advertisers have the most to lose (fraud hurts their ROI)
• Better to not just focus on clicks, impressions – more prone to fraud
• Focus on ROI (negative if fraud occurred), real conversions; viewability (e.g. MOAT; MRC)
• DoubleVerify, Integral Ad Science, Forensiq, White Ops all offer anti-fraud services
• Fraud (ad not served to a human), poor viewability (person could not see the ad) both hurt
ROI
Source: Dominik Kosorin, “Introduction to Programmatic
Advertising”
IPA CH6: CURRENT ISSUES: VIEWABILITY
• The opportunity for an ad to be seen (e.g. above fold, same tab)
• MRC (Media Rating Council) definition: >= 50% of ad pixels in viewable space of
browser window (or on an in-focus tab) for >= 1 second post ad rendering
• 30% sufficient for larger formats
• Video ads: >= 50% for >= 2 seconds
• Same requirements for mobile web and apps as for desktops
• Actual rates vary by publisher, but most 50% - 70% viewability
• Expected viewability can be used as machine learning model inputs
• ViralGains: Joint target: Delivered View with Viewability
• MRC accredited vendors:
• DoubleVerify, MOAT, AppNexus, Integral Ad Science, Google Active View
Source: Dominik Kosorin, “Introduction to Programmatic
Advertising”
IPA CH6: CURRENT ISSUES: VIEWABILITY, CONT’D
• Despite standards, viewability scores vary by vendor based on many factors:
• How each identifies, deducts for fraudulent impressions
• Moral dilemma: Better to turn a blind eye to fraud and not decrease your
viewability?
• Percentage of impressions the vendor is able to measure
• Sampling vs. measuring every available impression’s viewability
• Note: viewability estimates known in advance, but true viewability not until ad
served
• Smaller format ads have higher viewability, on average – but might not deliver
better business outcomes
• Incorrect / inconsistent measurements (discrepancies) can lead to overbidding,
underbidding
Source: Dominik Kosorin, “Introduction to Programmatic
Advertising”
CH6: CURRENT ISSUES: AD BLOCKING
• On the rise over the last five years (Ad Choice)
• Many reasons people block ads
• Not relevant, need to be closed (if you can find the X)
• Production quality (content)
• Too long (video)
• Instream – can’t watch the video I’ve selected until after video ad plays
• Retargeting – privacy concerns (plus, people @ Sbux can see my
screen)
• Rich media ads slow down page load times (especially on mobile)
• Ad blockers now easier to install (e.g. Apple content blocking on iOS)
Source: Dominik Kosorin, “Introduction to Programmatic
Advertising”
CH6: CURRENT ISSUES: ATTRIBUTION
• Ideally, align budget with ROI of each advertising channel / device
• Digital (social, search, display, video, etc.)
• Offline (TV, print, billboard (OOH), radio)
• Not always easy to measure
• Best channel overall vs. for certain customer personas / geos / devices / etc.?
• Impact of / how to measure when ROI is non-financial e.g. brand
awareness/equity/perception
• First-touch vs. last-touch (last-click or last-view), etc.
• Last-touch issues: Might have converted anyway (uplift modeling – Vasya, 4/6/21
class); maybe KPI was higher in funnel (acquisition; brand introduction), not
conversion
Source: Dominik Kosorin, “Introduction to Programmatic
Advertising”
CH6: CURRENT ISSUES: PRIVACY
• Internet browsing; app usage; social media; smart devices all leave a digital footprint
• Pareto distribution – most personal data owned by a few large companies (FB/IG, Google)
• “Forgot password/new user? Log-in with Facebook or LinkedIn” to several sites
• Trade (explicitly or implicitly) your personal data for free services
• Government regulations, e.g. Electronic Communications Privacy Act (ECPA), Children’s Online
Privacy Protection Act (COPPA)); the EU has General Data Protection Regulation (EDPR)
• Industry self-regulates (not consistently)
• GDPR – General Data Protection Regulation “is a regulation in EU law on data protection and
privacy in the European Union and the European Economic Area. It also addresses the transfer of
personal data outside the EU and EEA areas.”
• Individual countries (e.g. Germany has strict privacy laws)
• Personally Identifiable Information (PII) – name, address, SS#, credit card #’s, DOB, phone # --
none should be used for behavioral targeting
Sources: Dominik Kosorin, “Introduction to Programmatic Advertising”; Google
(GDPR)
