Will Robots Take Over PPC?When
About Me@SiliconVallaeys
Agenda: Trends That Will Change PPC
Automation
The On-Demand Economy
Artificial Intelligence
Trend #1 Automation
Making a Case For Automation...
The Case Against Automation...
5Levels of automation for PPC
Level 1 - Monitor and Alert
Alert a human when something may need attention.Try this AW Script: bit.ly/GranularAnomalyDetector
Level 2 - Manage Individual Components
Run automated tasks for each entity, keywords, bids, ads, etc.→ Automated Rules
Level 3 - Manage Multiple Systems To Achieve a Simple Goal
Set a CPA target and automation finds keywords, manages bids and budgets.Try custom AdWords Scripts
The current state of
Bid Management
CPC Bid
A Tale of Two Kinds of Bids
Demographics / RLSA
Geographic
Dayparting
Device
Modifiers
Matrix Bid Management
Locations
Devices
Dayparts
Audiences
1 keyword may have tens of thousands of permutations1000 locations * 3 devices * 6 dayparts * 2 genders * 5 age ranges = 180.000 bids
CPC Bid
How to Do Matrix Bidding
Requires a very fine grained account with lots of duplication
Single Keyword Campaigns (Skamps)
AdWords has a 10.000 campaign limit per account
You can’t do this at scale without automation
Alternatively use Enhanced CPC to get the benefit of real-time auction data
Average Data
A Common Bid Management Mistake Since Enhanced Campaigns Started
Mobile Data
Desktop Data
Tablet Data
vs
How to Manage Bids Correctly With Multiple Devices
Desktop Bid (Your CPC bid)
Mobile Bid Modifier
Anchored Bids Introduce Management Complexity and Dependencies
Desktop Bid
When the bid you are
anchored to changes,
modifiers must also be
adjusted.
Anchored Bids Introduce Management Complexity and Dependencies
Desktop Bid
Mobile Bid Modifiers
Generic Static Bid OR a bid based on all device types
Mobile Bid Modifier
Tablet Bid Modifier
Desktop Bid Modifier
Bidding By Weather
Discovering a Relevant Signal
Predicting Conversion Rates For Shopping Queries
PREDICTIONS
1. Prediction for data: ["big and tall undershirt",0,43,1,"Style Exact","custom0=style exact and product_type_l1=t-shirts and product_type_l2=short sleeve and brand=gildan"] is 0.025902
2. Prediction for data: ["hanes sleevles tees",0,1,1,"Style Color","custom0=style color and product_type_l1=t-shirts"] is -0.0003863. Prediction for data: ["bulk baseball sleeve shirts",0,1,1,"Style","custom0=style and product_type_l1=t-shirts and brand=anvil"] is -
0.0230494. Prediction for data: ["big & tall t shirt green",0,1,1,"Style Exact","custom0=style exact and product_type_l1=t-shirts and
product_type_l2=short sleeve and brand=gildan"] is 0.0666165. Prediction for data: ["navy and gray raglan",0,4,1,"Style Color","custom0=style color and product_type_l1=t-shirts"] is -0.0150156. Prediction for data: ["cuffed long sleeve shirt",0,1,1,"Style Color","custom0=style color and product_type_l1=t-shirts"] is -
0.0492877. Prediction for data: ["big and tall mens shirts sale",1,41,2,"Style Exact","custom0=style exact and product_type_l1=t-shirts and
product_type_l2=short sleeve and brand=gildan"] is 0.2428568. Prediction for data: ["plain tees",0,19,1,"Style Exact","custom0=style exact and product_type_l1=t-shirts and
product_type_l2=short sleeve and brand=gildan"] is 0.0451519. Prediction for data: ["mens fleece quarter zip pullover",0,5,1,"Style Exact","custom0=style exact and product_type_l1=jackets"] is
0.099721http://bit.ly/PredictShoppingQueries
Level 4 - Intelligent Systems Manage Entire Well Defined Accounts
We set parameters for several things and the system manages the account.→ locations, products (i.e. business parameters)→ goals (e.g. ROAS)→ restrictions (e.g. only run on Google Search)
Level 5 - Fully Automated
Connect to your business financials, write a blank check to Google and let their robots handle it.
Some Automation is Too Conservative
Humans Can Learn the Robots’ Techniques
Trend #2The On-Demand Economy
More Freedom and Fewer Responsibilities
Applying Principles From the On-Demand Economy to PPC
Break complex PPC management into small tasks that can be done by a freelancer
30
Keyword Suggestions Example
Store a set of keyword suggestions in Google Sheets where they can be reviewed by a freelancer.
Get Freelance Specialists to Process the Suggested Keywords
Expanded Text Ads How to achieve the results
Google’s case studies show
Ads That Are Optimized Outperform ETAs
“When ETAs do better, they significantly outperform legacy ads. The problem is that advertisers need time to test and tweak their new ads before they will achieve these results.”
- A Google Product Manager
Google pushed back the sunset date for legacy ad creation to January 2017
Analysis #1 and #2
Boxplots and Quartiles
Analysis #1 and #2
Benefits of Using DKI: In Title of Any Type of Text Ad
Does the use of dynamic keyword insertion boost an ad’s performance?
Benefits of Using DKI:In Description of Any Type of Text Ad
Does the use of dynamic keyword insertion boost an ad’s performance?
Benefits of Using DKI:In Expanded Text Ads
Using Path1 and Path2
Does using the URL paths boost performance of ads?
Benefits of Repeating Path Text in Text of ETA
How often are Path1 and Path2 used in Headline1 and Headline2
Do ads perform better when the text from the path also appears elsewhere in the ad?
Does Ad Text Length Matter
AdWords doesn’t always show all the text we submitted in our ads. How does the length of our ad impact performance?
VS.
Impact of Length of Headline 1 vs Headline 2
Mobile Preferred Expanded Ads
Are you upset that mobile preferred ads are not available for Expanded Text Ads?
95%Of mobile preferred ads were different from the standard ad by only a
single word
Run Different Mobile Expanded Ads With Ad Customizers
1. Use the Ad Customizer attribute for “Device preference”:
2. Set up your ad using this new data:
Run Different Mobile Ads withValue Track
Coming Soon! ValueTrack
{ifmobile:say this} {ifnotmobile:say something else}
Already available for use with the final URL
Coming soon to visible parts of the ads
Trend #3Artificial Intelligence
Quality Score - Finding Correlations in CTR
Obvious correlations... Less obvious ones...
In our analysis, 71% of keywords with a change in QS had no change to the First Page Bid Estimate on the same day.
If you use a Google Bid Strategy that depends on ‘first page bid’ or ‘top of page bid’ estimates, be careful!
The Correlation between QS and FPB
• Going from QS 4 to 8 does NOT mean your ad rank doubles
• Hence much of the published data on monetary impact is incorrect…
The Quality Score You See Is Not Linear
If Quality Score Was Linear
Quality Score Is Not Linear
Decrease in CPC for 1 Point QS Gain
Going UP from QS 5 to 6 reduces First Page Bid estimate by 37%
AI Assistants“It’s clear to me that we are moving from a mobile-first to an AI-first world.”
–Sundar Pichai, Google
Alexa and Optmyzr
Thanks@SiliconVallaeys