Bulletin, December/January 2006


Repeat Search Behavior: Implications for Advertisers

by Nico Brooks

Nico Brooks is director of research & development, Search at Atlas. He can be reached by email at nico.brooks at atlassolutions.com.

Research by comScore Networks in December 2004 found that people who search online before making a purchase tend to search many times before purchasing (www.comscore.com/press/release.asp?press=526). For purchasers of keyword ads, this finding raises the question: What impact does repeat search behavior have on how search marketing performance is evaluated? To gain insight into this issue, we used advertiser campaign tracking data and some of the metrics commonly used in managing online advertising campaigns. The data analyzed included all tracked search clicks leading to conversion for 10 different advertisers. One month of click-to-order conversion data was analyzed. Unique users were identified by means of a tracking cookie. Orders were tracked by a tracking pixel placed on an order confirmation page. The advertisers included in the analysis were all consumer-focused, but they encompassed a broad range of products and services.

The first question we investigated: Of purchasers driven by search, what percentage clicked on more than one keyword ad for a given advertiser before purchasing?

The findings are summarized in Table 1, which indicates nearly a third of visitors who completed an order clicked on more than one keyword ad. This percentage certainly has a material impact on how search advertising campaign performance is evaluated.

The next question we investigated: What impact does repeat search behavior have on conversion attribution?

In this context, conversion attribution refers to the attribution of an online purchase to a keyword ad click - the user who clicks on a search ad converts from a visitor to purchaser. Typically, a click-to-order conversion is attributed to the last click before purchase. However, if a searcher clicks on more than one keyword ad before purchasing, then presumably each keyword ad that is clicked influences the sale and should be considered when calculating performance metrics. We looked at the impact of attributing conversions to the first click before purchase versus the last click before purchase on the two metrics often used in the management of online advertising campaigns: time-to-convert and cost-per-order (CPO).

Time-to-convert is calculated as the period from when a visitor clicks on an ad to when the visitor completes a purchase. For example, if a visitor clicks on a keyword ad at 1:00 p.m. and returns the next day and completes a purchase at 11:00 a.m., the time-to-convert is 22 hours. Time-to-convert is a useful gauge in understanding the consideration cycle for products and is relevant to performance analysis by time-of-day or day-of-week as a recent study by the Atlas Institute documents (www.atlassolutions.com/pdf/DaypartAnalysisDMI.pdf).

In Table 2, average time-to-convert metrics are compared for the advertisers in our sample. The "First" column shows the average time-to-convert from the first search ad click and the "Last" column shows the average time-to-convert from the last search ad click. For most of the advertisers, there is a dramatic difference between the two figures. On average, the time from first click to conversion is 2.7 times as long as the time from last click to conversion.

The observed increase in time-to-convert has an important implication for advertisers: if an advertiser is only looking at the last click before purchase, part of the consideration cycle may be missed. For example, the last click before purchase may largely represent shoppers who have already made a decision regarding which product to buy and are now deciding on where to buy.

The second metric we looked at was CPO, which is calculated as advertising cost divided by the total number of tracked orders. If considered at the campaign level, CPO does not change when we account for repeat searches, since the total cost and orders do not change. When calculated at the keyword ad level, however, how we look at ad performance may change dramatically. For example, consider the following with keywords X, Y and Z:

Scenario 1:     100 clicks on keyword ad X resulting in 5 orders at a cost per click of $1.

Scenario 2:     100 clicks on keyword ad X preceded by 25 clicks each on keyword ads Y and Z resulting in 5 orders. Keyword ads X, Y and Z are priced at a cost per click of $1.

In Scenario 1, the CPO for keyword X is calculated as (100 x $1)/5, or $20. In Scenario 2, we donít have enough information to calculate CPO. We need to know which keywords specifically were clicked before purchase. For the purposes of the example, we will presume that keyword ad X was the only ad clicked before 3 of the 5 orders and was preceded by ad Y for one order and by ad Z for one order. The ads clicked before each conversion have been summarized in Table 3.

With this information, we still have to make a decision regarding how we will attribute click-to-order conversions to each ad clicked. Should the first or last keyword ad clicked receive all credit for the conversion? Should credit for the conversion be shared among all ads clicked? This topic most certainly merits further study, but for the purposes of this analysis we will choose the latter approach and share credit equally. While this choice may be arbitrary, it is no more arbitrary than the last click attribution method proscribed by most ad tracking technology available today.

Sharing credit equally, we apply the data in Table 3 to Scenario 2 as follows to calculate the CPO for keyword ad X:

1.       100% of orders 1, 2 and 3 are attributed to X, totaling 3 orders

2.       50% of order 4 is attributed to X, totaling 0.5 orders

3.       50% of order 5 is attributed to X, totaling 0.5 orders

4.       The sum of orders attributed to X is therefore 4

5.       The total cost for keyword ad X is $100 (100 x $1)

6.       The CPO for keyword ad X is $25 ($100/4)

Comparing Scenarios 1 and 2, the CPO has increased by 25% when we account for previous keyword ads clicked.

To better understand the impact of shared click conversion attribution versus last click conversion attribution when calculating CPO, we compared CPO using the two methods of conversion attribution for the top-performing keyword ad for each advertiser in the sample shown in Table 1. The top-performing ad in this case was determined by which ad drove the highest number of conversions. The results are summarized in Table 4.

We were very surprised by these results, which generally showed very little difference between the two calculations. Given the fact that 32% of purchasers overall clicked on more than one keyword ad before purchasing, we had expected that there would be a more significant shift in CPO valuation if we included clicks preceding the last click in our calculations of CPO.

This unexpected result prompted us to dig further into the distribution of keyword ads in the sequence of ads clicked before purchase. What we found was that high volume keyword ads tend to appear frequently among all ad clicks leading to conversion, whether first, middle or last. In fact, we found that searchers would often click on the same ad multiple times before converting. Analyzing repeat search behavior for the advertisers sampled, we found that 82% of visitors who clicked on more than one keyword ad before completing an order clicked on the same ad multiple times. The visitor may have also clicked on more than one unique ad, but his/her clicks included clicks on the same keyword ad at least twice.

Summing up: People who click on search ads often click more than once before purchasing. There is usually a long delay between the first and last click before purchase, and they are very likely to conduct the same search over and over.

While there are quantitative implications in these results for search ad performance, perhaps the greatest implication is in the nature of search behavior itself. The fact that visitors are conducting the same search multiple times to get to the same place implies that their behavior shifts from being inquisitive in nature to being navigational in nature as the search sequence progresses. While this knowledge appears to have minor implications for how keyword ads are valued, it has major implications for the state of mind of the searcher and therefore has major implications for how the advertiser can best serve the visitor when he or she arrives.


Table 1: Percentage of visitors who clicked on more than one keyword ad before completing a purchase.

Advertiser

Percentage

a

13%

b

42%

c

33%

d

30%

e

41%

f

18%

g

48%

h

36%

i

47%

j

11%

Average

32%

 

 

Table 2: Difference in time-to-convert when comparing first-click-to-conversion with last-click-to-conversion.

Advertiser

First
(hh:mm:ss)

Last
(hh:mm:ss)

a

4:04:57

1:48:22

b

33:56:40

23:25:52

c

24:26:26

11:45:17

d

45:28:34

4:30:20

e

27:13:13

9:58:34

f

23:49:35

20:17:05

g

214:11:43

111:26:17

h

16:37:34

16:22:18

I

69:02:53

29:56:00

j

0:52:17

0:26:06

 

 

Table 3: Illustration of click-to-order conversion attribution.

 

Keyword Ads Clicked

 

X

Y

Z

Order 1

yes

no

no

Order 2

yes

no

no

Order 3

yes

no

no

Order 4

yes

yes

no

Order 5

yes

no

yes

 

 

Table 4: Difference in CPO for the top performing keyword ads when comparing last-click-to-conversion attribution with shared-click-to-conversion attribution. A difference with a negative value means that CPO decreased when calculated using shared click attribution.

Advertiser

Difference

a

0.17%

b

1.67%

c

-0.21%

d

-0.11%

e

0.35%

f

-0.59%

g

0.68%

h

1.67%

i

-0.94%

j

1.01%