multi-channel attribution - where do leads come from?
DESCRIPTION
Gary Angel, President of Semphonic presented a keynote discussion to the 2010 AIM Conference on the topic of attributing value to leads when the advertiser knows that multiple lead sources contribute to the buying decision. aimconf.comTRANSCRIPT
AIM 2010
Where to Leads Come FromMulti-Channel Attribution
Gary Angel
What is Attribution
• Attribution is a process for assigning/crediting a lead or a conversion to a specific set of marketing activities and touchpoints.
Easy or Hard?
• Sometimes it’s easy:
• Sometimes it’s not:
We got 1,500 calls.
Saw TV Visited Web Searched Visited Web Called
It’s Getting Harder
• The number and diversity of marketing channels is growing making attribution even harder.
Why it Matters
• Fights over attribution are sometimes perceived as being largely political:
My TV ads drove 5 million in business
How come we only had 6 million in sales?
My PPC campaign drove 2.5 million in business
My Display Campaign drove 1 million in business
Why it Matters
• But you can’t optimize your overall Media Mix or your individual channel strategy unless you understand how each marketing channel works and how the overall program works together.
Why it Matters
• For Instance:– If Natural Search attracts early-stage buyers, it
may seem to perform worse than Paid Search even though it drives more new customers.
– If Paid Search is heavily dependent on branded traffic, it may only perform well in conjunction with Mass Media buys.
– If Display creates awareness but doesn’t drive direct traffic it may only work in conjunction with heavy search buys.
Common Methods of Attribution
• Last (Most Recent): The last known campaign touchpoint gets the credit for a conversion.
• First (Original): The earliest known campaign touchpoint gets the credit.
• Equal: Credit is divided equally between all known touchpoints.
• All: Full credit is given to every touchpoint.• Spoken: Credit is given to the touchpoint
identified by the customer.
Last (Most Recent)
• The more your lead/sales environment is multi-touch – the less well this method works.
Over-Counts operational mechanisms like Google
Search
Easy to Implement
In this example, 35% of the visitors who clicked on a paid brand term
had already clicked on another paid term in the SAME session!
First (Original)
• Especially problematic with longer sales-cycles.
Cookie Issues make
accuracy questionable
Emphasizes “true”
acquisition
For this logged-in site, 33% of the people who registered used a different cookie IN THE SAME
MONTH. More than 50% used a different cookie within 3 months.
Jun-09 Jul-09 Aug-09 Sep-09 Oct-09Visitors 33% 45% 53% 59% 64%
Logged In With Different CookieVisitors Who Registered in June, 2009
Equal (Shares Credit)
• Doesn’t really help with Marketing Mix calculations.
Punts on media mix
issues
Doesn’t overstate total and
shares credit
All (Every Touchpoint full Credit)
• Doesn’t really help with Marketing Mix calculations.
Overstates marketing
efficiency and punts on
media mix
?
• If you’ve siloed reporting, there’s a good chance this is your de facto strategy!
Spoken (Customer id’s Touchpoint)
• Customer testimony is heavily influenced by collection point and method and, even at best, can be quite unreliable.
Understates some channels –
overstates last or easiest
Does Media Mix and doesn’t
overstate results. Politically
reasonable.
Survey Respondents Rating ToolRating % Visitors % Who Actually Used Tool1 (Very Dissatisifed) 14 64%2 11 67%3 6 69%4 10 73%5 15 58%6 13 75%7 15 78%8 8 79%9 5 74%10 (Very Satisfied) 3 71%
Only 64% of the most dissatisfied visitors in an online survey had
actually used the tool they were rating.
Proper Attribution
Aggressive Collection
Careful Analysis
Necessary Research
Aggressive CollectionUse of Multiple 800#
Services
Use of Online Promo-Coding
Careful tracking across online
Reward the Ask
• Numbers by Source and Channel
• Provide incentives to identify
• Universal Campaign Coding• Campaign Stacking• Universal Sourcing (Treating social as campaigns)• Time-Coding• Global Roll-ups• Shared Cookies
• Sweeps for completed forms• Separate Sourcing Questions• Off-Premise (online) fulfillment
Web Analytics Techniques
• Universal Campaign Coding• Code all URLs with Params• Don’t neglect URL Shorteners• Standardize campaign coding and enforce.
Web Analytics Techniques
• Separation of types of campaigns:• Internal• External
• This method will allow separate attribution studies to be made of internal vs. external without one type screening off the others.
Web Analytics Techniques
• Campaign-Stacking• Use a cookie to store all campaign values
recorded for a visitor• Pass to a variable in your web analytics
solution• Omniture has a plug-in for this
Web Analytics Techniques
• Use multiple variables to track different attribution periods at least until confident of your sales-cycle.
• We suggest the following attribution strategies:• Campaign = 45 days• Internal Campaign = visit• Redundant Campaign 1= Visit• Redundant Campaign 2= 7 Days
Web Analytics Techniques
• Capture of Visit Number• Use a cookie to track the visit number of
each visitor to your web properties• Record the Visit Number in a web analytics
variable• Omniture has a plug-in for this
Web Analytics Techniques
• Key Event Time-stamping• Time-stamp key events in custom variables• The timestamp should include at least the
full date (YYYYMMDD form)• Key events to timestamp include external
campaign arrival, lead and conversion• Timestamps are used in segmentation to
study attribution and before/after key event behavior.
Web Analytics Techniques
• Transaction Ids• Record the Lead Id in your web analytics
solution • Where possible (Omniture, Unica, Core),
update this value with actual lead-status• This allows you to track actual lead
effectiveness by campaign / behavior
Careful Analysis
Always track over-time and across web session
Always segment by sales-stage and known attributes
Use type of search and arrival method to help identify acquisition method
Necessary Research
Identify what you don’t know
Use traditional research methods (especially survey) to answer
Use Dark-Period Testing to answer questions about media mix
Recommended Conceptual Model
• Display Ad to Site to Lead
• TV to Call but identified Web Site
• PPC on Brand Name to Site to Lead
• SEO to Site no Lead in that Visit
• Site Visit no Lead
• TV to Call• SEO long-tail to
Site to Lead
True Acquisition
Contributor
Acquisition Trigger
Operational Trigger
Recommended Attribution Model
Operational Trigger
Any Previous Touchpoint and Conversion method is user-push (PPC Branded Term to Site to Lead)
Acquisition Trigger
Any Previous Touchpoint and Conversion but method of Conversion isn’t user-push (Display or TV Call-in)
Contributor – Sourced by didn’t convert
Any Previous Touchpoint within a defined Sales Cycle and no Conversion
True Acquisition – new customer
Single Known Touchpoint (e.g. 1 web visit or direct to call)
It’s not just campaigns
0% 10% 20% 30% 40% 50%
AT-InvestorPlace-NXPAT-Nasdaq-NUN
AT-ON-BarChart-NPCAT-ON-BIGCHARTS-KJS
AT-ON-BRIEFING-KJXAT-ON-CBSMW-KKH
AT-ON-CNBC-NHZAT-ON-IndBrains-NFV
AT-ON-INVESTOPEDI-KHLAT-ON-INVESTORSCO-KHY
AT-ON-MSN-NFQAT-ON-Quote-NJA
AT-ON-SCHAEFFERS-NHNAT-ON-THESTREET-KHH
AT-ON-TRADINGMRKTS-KKEAT-ON-WSJ-KHX
AT-ON-YAHOO-KJKAT-SC-Public_Site-MNV
AT-SeekAlpha-NQEAT-WS-CYBER-BDI-NHB
% Decline in Response
• This analysis looked at attribution based on previous visits without campaign sourcing.
For two campaigns to this site, almost 50% of their click-
throughs had already used the site!
Showing Multi-Attribution
• This Report shows how often each campaign interacts with others and what the total conversion rate is.
Showing Multi-Attribution
• Here’s the same report with the biggest incremental lifts highlighted.
Dark-Period Testing
• Testing media mix by going dark and tracking to baseline is a consistently useful method.
In this case, going dark
showed that the client had to purchase 6 clicks from these
terms to get a single incremental visit.
Correlation Analysis
• Using correlation models, you can match marketing efforts to outcomes (if you have enough clean data):
For this home products company we correlated weekly TV, Radio and Print TRPs, Banner Impressions, Search plus
econometric data on housing starts and prices by region with web site activity.
Media effects only correlated with a 1-2 month lag – TV was by far the most
impactful.
Comprehensive Reporting
• This client chose to compare campaigns by pipeline stage:
Comprehensive Reporting
• Here’s an example using our 4 Stage terminology:
Comparison by Type Best Companion Best Overall & Trend
Key Take-Aways
• Aggressively Code Sourcing– Multiple #800s and Campaign-Specific #800s– Multiple Asks– Reward Asks– Code Campaigns Everywhere
Key Take-Aways
• Over-Time Tracking– Campaign Stacking– Universal Sourcing– Time Encoding– Global Rollups– Shared Cookies
Key Take-Aways
• Analyze Carefully– Segment by Attribution and Life-stage type– Look at attribution beyond just marketing
campaigns (existing customers/visitors)– Use Dark-Periods to Baseline and Tune Media Mix– Use Correlation Models to track marketing
impacts vs. econometric variables
I’m Out!
• Gary Angel• President, Semphonic• 415 884-2511• [email protected]• http://semphonic.blogs.com/semangel/• http://www.semphonic.com• @garyangel on Twitter