Multi channel attribution moving beyond last click
discovering high value ppc ad sequences with multi-touch attribution
1.
2. 1
CONTENTS:
DISCOVERING HIGH VALUE AD SEQUENCES WITH MULTI-TOUCH ATTRIBUTION:
INTRODUCTION....................................................................................................................................2
METHODOLOGY...................................................................................................................................2
FINDINGS .............................................................................................................................................3
HOW DO PPC AD PURCHASE FUNNELS CHANGE Q3 TO Q4? DO THEY GET LONGER? DO THEY
GET MORE VALUABLE? ........................................................................................................................3
ARE MULTI-TOUCH PURCHASE FUNNELS MORE VALUABLE? .............................................................4
PREVALENCE OF NON-TRADEMARK CAMPAIGNS LEADING TO FUTURE TRADEMARK TERMS? ......4
DO HIGH PERFORMING SEQUENCES OF CAMPAIGN TYPES HAVE MESSAGE THEMES? .................7
HOLIDAY MESSAGING IMPACT IS NOT GREAT ................................................................................10
DIFFERENCES BETWEEN MESSAGES AT THE BEGINNING VS THE END OF THE SEQUENCE ..............11
STRATEGIC IMPLICATIONS ................................................................................................................12
GOOD SELLING!.................................................................................................................................12
THE NETELIXIR UNIVERSITY TEAM.......................................................................................................12
FOR MORE INFORMATION ................................................................................................................12
APPENDIX...........................................................................................................................................13
GLOSSARY OF HEADLINE CODING SYSTEM.....................................................................................13
EXAMPLE OF AD BEFORE AND AFTER CODIFYING ..........................................................................14
3. 2
INTRODUCTION
Our latest research was inspired by the findings of our Mastering Attribution Through Conversion Path
Analysis paper and our ensuing conversations with clients. Whereas the Mastering Attribution study was a
very “left brain”, logical quantitative analysis, we set out with this study to explore the “right brain” aspect
of marketing as it is manifested in the messages contained in ppc advertising. Specifically, with the
understanding that ~60% of
purchase funnels involve more
than 1 customer touchpoint,
we set out to determine if there
were specific ppc ad copy
sequences that stand out, as
well as whether or not those
sequences change during the
peak holiday selling season.
Reading our latest search
engine advertising research
will help you:
Learn how high-
performing, top-of-the-
funnel ads change
during the holiday
season.
Learn how to use your conversion path reports to identify your most valuable ad copy sequences
With suggestions for ad copy testing that you can implement immediately
METHODOLOGY
Most ppc advertising effectiveness studies focus on the click-through-rate (CTR) of the ads as the key
performance indicator. For this study, we chose to focus on the end result, the ones that matter most,
which of course are orders and revenue. We pulled conversion funnel data for last year’s Q3 and Q4
selling seasons from a subset of the Google Analytics accounts that were included in our attribution
paper. We selected only those paths that had more than 1 ppc ad customer touchpoint, specifically
Google AdWords touchpoints (source/medium= google / cpc for the GA geeks out there), so that we could
easily access ad copy content and performance data together. We then “normalized” ~10,000 individual
ad copy messages into their building block components so that we could aggregate statistics. For
example, headlines with language like “Official Store”, “Official Site”, “Official Website”, etc. were coded as
“Official”. All trademark terms were coded as “TM”, 3rd party brand names became “3rd brand”, and so
on.
The industry sectors included in this study are the same sectors we examined in the attribution study,
which are:
Apparel
Home Furnishings
Health & Beauty
B2B
Fashion
Hobby & Leisure
Food
Gifts
4. 3
Some of the questions we sought to answer were:
How do conversion funnels change from Q3 to Q4? Do they get longer? Do they get more
valuable?
Are multi-touch conversion funnels more or less valuable?
How prevalent is the common perception that non-trademark paid search campaigns result in
subsequent return visits using trademark terms?
FINDINGS
HOW DO PPC AD PURCHASE FUNNELS CHANGE Q3 TO Q4? DO THEY GET
LONGER? DO THEY GET MORE VALUABLE?
Table 1 Paid Search Conversion Path Length Summary
CPC PATH LENGTH Q3 AVG # STEPS Q4 AVG # STEPS
At least one CPC touch
1.54 1.06
>1 CPC touchpoints
5.68 4.02
Maximum path length
60 40
We found that paid search related conversion funnels do change in the 4th quarter: they shorten. All
paths that contained at least a single paid search touchpoint dropped from an average of 1.54 steps to
1.06 steps in Q4. These numbers include all of the touchpoints in the path, regardless of whether they
were paid search induced or not.
When you filter the data for paths that contained multiple paid search touchpoints, the trend was still
evident with paths containing multiple paid search driven visits dropping from an average of 5.68 steps in
Q3 to 4.02 in Q4. Again, these numbers include all steps in the conversion path.
One reason we see the average path length of conversions with at least a single CPC touchpoint drop in
the 4th
quarter is because of a jump in those conversion funnelsthat are just one step: a CPC visit. They
jump from 46% of conversions in Q3 to 53% in Q4. Another reason for this drop is because of the
maximum path lengths we saw in the data. In Q3, we encountered path lengths that reached 60
touchpoints, while in the 4th
quarter that maximum dropped by a third to just 40 touches.
5. 4
Table 2 Paid Search Conversion Path Contribution Summary
Q3 Q4
Single CPC conversions share of total
conversions
46% 53%
Single CPC conversion value share of total
conversion value
42% 51%
Single CPC conversion path AOV
$110 $106
AOV for Multi-touch conversion paths with at
least 1 CPC touch
$126 $115
Correlation score between path length and AOV
(.25) .42
ARE MULTI-TOUCH PURCHASE FUNNELS MORE VALUABLE?
As you can see in Table 2, at the transaction level, multi-touch CPC conversion funnels are more
valuable. In Q3, they were 15% more valuable, while in Q4 that delta was 8%. We suspect this
phenomena might be related to the level of purchase consideration. Generally speaking, larger purchases
involve more logical decision-making than impulse purchases, so the correlation between path length and
AOV could simply be a result of consumers doing their homework before hitting that “checkout” button.
What is surprising in this dataset is the correlation score between path length and AOV for each quarter.
For Q3, it’s actually negative, as we see a drop in AOV at the high end of the number of conversion path
steps.
PREVALENCE OF NON-TRADEMARK CAMPAIGNS LEADING TO FUTURE
TRADEMARK TERMS?
One of the more surprising facts in multi-touch attribution that we have seen in both our own research and
in other published research is the presence of trademark paid search campaigns closer to the top of the
funnel than the bottom. This fact surprises us because, like many retailers, we were believers in the
theory that non-trademark, or what some call generic keywords help advertisers get in front of prospects
at the beginning of their shopping journey, and then when those shoppers decide to go back to the
advertiser’s web site, they enter the site name or some variation of it into the search engine and click on
the advertiser’s trademark campaign ad. We bought into this theory because it is, as Spock, (may he rest
in peace), would have said “entirely logical, Jim”.
So in this study, we wanted to size the non-trademark to trademark ad sequencing and were surprised to
find that it’s not as prevalent as many perceptions would believe. Table 3 shows the share of conversions
for the 2 quarters we studied by type of ad sequence for all sequences in our study’s data. For example,
if a conversion path had 3 sequential steps with a paid search ad, we identified 2 sequences: the first to
the 2nd
ad and the 2nd
to the 3rd
ad. TM indicates a trademark ad, NTM indicates a non-trademark ad.
You can see that the non-trademark to trademark sequence, (which means the user clicked on a non-
trademark ad, then clicked on a trademark ad), contributed 18% of conversions in Q3 and 17% in Q4.
Interestingly, a non-trademark to non-trademark ad sequence was larger in both quarters with a sequence
of two trademark ads being the largest driver of conversions. Note that there is a big jump in the
trademark to trademark sequence in Q4. In our Mastering Attribution research paper, we showed the top
of funnel impact trademark campaigns have. Here is yet another reason to think hard about a decision to
cut back trademark campaign budgets during the holidays.
6. 5
Table 3 Paid Search Ad Sequences – All Path Sequences
We wanted to hone in on the top of the purchase funnel, so we quantified the ad sequences for the first 2
ads in the conversion funnels we studied. If the conversion path had more than 2 sequential ad steps, we
only looked at the first 2. Those results can be seen in Table 4. The shape of the results are very similar.
One notable exception is the non-trademark to trademark sequence drops in contribution share from 17%
to 12% in Q4. Here again we see non-trademark to a subsequent non-trademark ad outweighing the non-
trademark to trademark ad sequence, and the trademark to trademark sequence dominating.
Table 4 Paid Search Ad Sequences, First 2 Ads only
When you focus in on just the first step in the paid search portion of the conversion path, you see an even
more dramatic shift from Q3 to Q4. Table 5 shows these results where you can see the % of conversions
that had a non-trademark ad appear first drop from 42% to 29%.
7. 6
Table 5 First Ad Step Summary
One other finding of interest that seems almost counterintuitive is the average conversion path lengths by
ad sequence. Table 6 shows this data, and the average # of steps shown is for the entire conversion
path, not just the paid search version. You can see that when trademark ads come first in the sequence,
the path length is longer. For each quarter, the non-trademark initiated sequences are almost a full step
shorter than those that start with a trademark ad. We call this almost counterintuitive because another
wide held belief is that non-trademark ads attract a higher percentage of visitors who are new to the site,
and therefore it is likely that those new users’ conversion funnels would have more steps or visits in them.
For whatever reason, here we see that it is visitors coming from trademark ads that have a longer
conversion path.
We also looked at the AOV for each sequence and there was no clear trend. For example, the AOVs for
the first ad in the sequence analysis were within 1% of each other for both quarters.
Table 6 Avg. Path Length by Ad Sequence
8. 7
DO HIGH PERFORMING SEQUENCES OF CAMPAIGN TYPES HAVE MESSAGE
THEMES?
We wanted to identify whether or not different types of ad copy messages performed differently in Q4. So
we took the ad content for every ad in Q3 and Q4 and “normalized” the contents so that we could
aggregate across the different accounts. It’s also important to note that the clients that are included in this
study include, to borrow a retail term, a mix of new and comparable clients. That is, NetElixir was not
managing some of these paid search efforts last Q3 and Q4, meaning we had no input into the ad copies
that ran then.
In all, we codified over 15,000 different ads across the 2 quarters. For example, any trademark ad that
included some kind of “official” message, such as “Official Store”, “Official Web Site”, “Official Site”, etc.,
was simply coded as “Official”. Terms that represent a type of product or a product category were coded
“product keyword”.
Tables 7 & 8 show the differences in headline content for the first ad position in those conversion funnels
containing more than one paid search contact point.
For the non-trademark ads, one notable change from Q3 to Q4 is that headlines featuring some aspect of
the product’s unique selling proposition fall out of the top performing headlines. For trademark
campaigns, one common theme to point out that applies to both quarters, is that all of the headlines
include some variation of the trademark name. A notable difference between the 2 quarters in trademark
copy was the jump that action words in the headlines took, going from leading to 4% of conversions in Q3
to almost 10% in Q4. Finally, take a look at how the conversions are concentrated by first ad type. The
trend moves in opposite directions from Q3 to Q4 based on type. Non-trademark headlines become more
diversified with the top 4 spots sliding from ~14% to ~11% of conversions, while on the trademark side,
the reverse is true as the conversion share of the top trademark headlines gets more concentrated,
moving from over 33% to over 45%.
Table 7 NTM Headline Formulas for First Ad Position
Q3 HEADLINE % OF CONV Q4 HEADLINE % OF CONV
NTM DKI 5.3% DKI 4.5%
Product USP 3.3% Product Keyword 3.5%
Product Keyword 3.0% Product Keyword & Sale 1.7%
Product Keyword & Sale 2.4% Product Keyword &
emotional adjective & sale
1.5%
9. 8
Table 8 TM Headline Formulas in First Ad Position
Q3 HEADLINE % OF CONV Q4 HEADLINE % OF CONV
TM Official 10.6% Official 18.2%
TM 8.0% Action TM 9.7%
Emotional adjective
& sale & TM
5.6% TM 9.3%
TM product USP 5.3% DKI 4.9%
Action TM 4.0% Emotional adjective & sale & TM 3.7%
Table 9 shows the top headline sequences for the first 2 ads in the conversion path.
They are dominated by trademark to trademark sequences. The DKI to DKI sequence does include some
non-trademark ads. In Q3, 56% of those conversions had non-trademark ads kicking them off while in Q4
the non-trademark share of that sequence dropped to 41%.
Table 9 Top Headline Sequences
Q3 1ST HL Q3 2ND HL % OF CONV Q4 1ST HL Q4 2ND HL % OF CONV
Official Official 10.1% Official Official 13.8%
TM TM 7.1% TM TM 7.5%
DKI DKI 4.5% Action TM Action TM 6.1%
Home of TM Home of TM 3.9% DKI DKI 4.1%
Emotional adjective
& sale & TM
Emotional adjective
& sale & TM
3.9% Home of
TM
Home of TM 2.6%
As you might surmise from looking at the formulas for the top headlines in Table 9, the sequences of 2
consecutive paid search touch points were dominated by sequential trademark ad touches, which are
denoted by TM > TM. In Q3, those sequences accounted for 73% of the subset of conversion funnels that
had sequential paid search touches, and that share grew to 80% in Q4.
The top ad message headline sequences by campaign type sequence are shown in Table 10. For our
study, we looked at 4 different sequences:
1. Trademark to trademark (TM > TM)
2. Non-trademark to trademark (NTM > TM)
3. Non-trademark to non-trademark (NTM > NTM)
4. Trademark to non-trademark (TM > NTM)
As many retailers might suspect, the trademark to non-trademark sequence had little data and was almost
negligible, contributing less than 2% of the total conversions in our dataset.
10. 9
Table 10 Top Ad Messaging Sequences by Campaign Type
Q4 Q3
HL 1 HL 2 HL 1 HL 2
TM > TM TM > TM
Official Official 33.1% Official Official 21%
TM TM 17.6% TM TM 17%
Action TM Action TM 13.0%
TM emotional adj.
Sale
TM emotional adj.
Sale 11%
Home of TM Home of TM 6.3% TM product USP TM product USP 8%
TM emotional adj.
Sale
TM emotional adj.
Sale 6.0% Home of TM Home of TM 8%
TM product USP TM product USP 3.7% Action TM Action TM 8%
NTM > TM NTM > TM
DKI Official 2.6% Product USP TM 3%
Product KW Official 2.6% Action 3rd brand Official 1%
Action 3rd brand Official 1.5% Product KW Official 1%
3rd brand Official 1.4% Product KW sale
TM emotional adj.
Sale 1%
Action TM Action TM 1.2% 3rd brand Official 1%
NTM > NTM NTM > NTM
DKI DKI 3.6% Product USP Product USP 5%
% off product KW % off product KW 1.8% DKI DKI 5%
Product KW DKI 1.4% Product KW Product KW 3%
DKI Product KW 1.3% Action 3rd brand Action 3rd brand 2%
Emotional adj. sale
product KW
Emotional adj. sale
product KW 1.3% % off product KW % off product KW 2%
TM > NTM TM > NTM
Action TM Action product KW 0.8% TM Product USP 1%
Official Product KW 0.6% Official DKI 0%
Official DKI 0.6% Action TM Action TM 0%
11. 10
If you look closely, you’ll notice some interesting contrast between the TM > TM and the NTM > NTM
data, with respect to changes from Q3 to Q4.
1. The top 2 headline patterns for the TM > TM sequences remain the same in Q3 and Q4. That
said, note the significant jump in the Official > Official share of the conversions in Q4, which is
almost 50% larger. Our guess is that this phenomena might be driven by gift buyers who received
suggestions on where to shop who want to be assured they are going to the right place.
2. DKI > DKI is moving up – which is more of a result from ad copy change by 1 client, moving away
from the Product USP format.
The top 3 headline patterns in the TM > TM group control a significantly larger share of the conversions,
going from 49% in Q3 to almost 64% in Q4, while the top 3 NTM > NTM headline sequences’ share of the
total shrinks in Q4 from 13% to almost 7%. The takeaway here would be to examine your own results for
the 2nd
half of last year to see if your trademark campaigns behaved in a similar fashion. If so, you might
want to ensure that you have ads with those Q4 messages ready to go for this year.
Table 11 shows the ad headline formats that appear most frequently in our analysis by campaign type
sequence (as in, how many times that format appeared in Q3 and Q4 top sequences). As indicated
above, using that “Official” language when referring to your business is crucial for your trademark
campaigns and Dynamic Keyword Insertion is critical for your non-trademark campaigns.
Table 11 Top Headline Formats from Ad Sequencing Study
HEADLINE FORMAT # OF OCCURRENCES
Official 13
DKI 9
Action TM 9
Product KW 7
TM 6
TM emotional adj. Sale 5
HOLIDAY MESSAGING IMPACT IS NOT GREAT
About 2/3 of the businesses in this study had holiday themed ad headlines including phrases like “Black
Friday”, “Cyber Monday”, “Christmas”, “Holiday”, etc. Those ads did not figure significantly in conversion
paths, as they were present in the paths that contributed just 6% of conversions. That share includes
both non-trademark and trademark ads.
12. 11
DIFFERENCES BETWEEN MESSAGES AT THE BEGINNING VS THE END OF THE
SEQUENCE
We were curious to see if there was a notable difference between the types of messaging that appear
early in the conversion path vs those at the end of the path. Table 12 lists the top headlines at the two
ends of the conversion path funnel. 4 of the top 5 headlines that appeared as the last paid search ad
touchpoint in Q3 were in the top 5 in Q4. Of the top initial headline in Q3, 3 of them are in the top 5 in Q4.
When you compare the two positions within the same quarter, you see some interesting results. In Q3,
just 1 of the first headline formats appeared in the last headline list, while in the 4th
quarter, not only do 4
of the first headline formats appear in the last headlines, they also appear in the same order. One
possible contributor to this phenomena is the fact that in the 4th
quarter, trademark ads in the first position
contribute a larger share (82%) of conversions than they did in Q3 (75%).
Table 12 Comparing First vs Last Headline in Path
Q3 LAST HEADLINE Q4 LAST HEADLINE
Official Official
TM Action TM
TM emotional adj. Sale TM
Home of TM DKI
Action TM TM emotional adj. Sale
Q3 FIRST HEADLINE Q4 FIRST HEADLINE
DKI Official
Product KW Action TM
Official TM
Product KW sale DKI
TM Online Product KW
14. 13
APPENDIX
GLOSSARY OF HEADLINE CODING SYSTEM
Here Is a glossary of the headline coding used in this study that is referenced in this paper’s content.
CODED MEANING
Official Official language such as official site, official store, etc.
TM Some variation of advertiser’s trademark terms
Action TM An action oriented phrase plus a variation of advertiser’s
trademark terms,e .g. “Shop at Toys R Us”
Home of TM Some variation of the phrase “Home of” plus a trademark
term, e.g. “Home of the Whopper”
TM emotional adj. sale Trademark variation plus an emotion-invoking adjective like
huge, monstrous plus sale oriented terms, e.g. “Best Buy’s
Monster Blowout”, “Wal-mart’s Craziest Deals Ever”
TM product USP Trademark variation plus a component of a product’s
unique selling proposition (USP), such as “New Balance –
In Wide Sizes Too”
DKI Included dynamic keyword Insertion. In AdWords, you can
code your ad copy so that the user’s actual query gets
inserted in your ad copy at points you designate in the
headline, description lines and display urls. If insertion
would cause character length violations, a default setting
that the advertiser chooses is used.
Product KW Product keywords, could describe sku level or category
level products. Does not include any brand reference.
3rd brand 3rd
party brand reference. This refers to a brand that an
advertiser sells which is not the advertiser’s own brand.
Can include a product keyword coupled with the brand or
may not.
Action 3rd brand An action oriented phrase plus a 3rd
party brand reference
as described above. “Find Reebok Sneakers Here”.
Product KW sale Product keywords, could describe sku level or category
level products, combined with terms describing sale events
such as “Area Rug Sale” or “Sale on Area Rugs”
% off product KW A % off discount offer for a particular product, such as “33%
Off Window Covers”
Co USP The advertiser’s unique selling proposition. Phrases like
“Free Returns”, “Same Day Shipping”, “No Sales Tax”.
15. 14
EXAMPLE OF AD BEFORE AND AFTER CODIFYING
Here is an example of an ad as written and as codified for our research.
AD BEFORE AD AFTER CODIFYING
Nielsen Metal Frames
Save Over 70% On USA Made Frames.
Frame Like a Pro w/ American® Frame
americanframe.com/Nielsen_Frames
3rd
brand
Action % off prod kw
Co USP
/3rd
brand