Getting the Full Picture
They say the human eye can only see a fraction the total light spectrum – and that’s just those sources identifiable with current technology. We live among these unseen radio and television waves, x-rays, solar rays, and various others, missing a huge portion of the true nature of everything around us.
I have to say that I’ve felt the same way as I pour through the data collected my clients’ websites, looking for insights and trends to optimize their search engine marketing (SEM) campaigns. I sometimes feel that I just don’t have the full picture I need to do the best job possible.
For the most part, spikes and dips in SEM performance can be attributed to influences such as a large television campaign kicking in, holiday seasonality, or a news story highlighting a particular keyword causing its search volume to rise. However, the truth is that I never know the complete story of why Person A went online and purchased a specific product or service or signed up for my site’s newsletter. I may see that he or she clicked Keyword X immediately before converting, but was there interaction with any media before finally making that decision? Did the individual hear a radio spot or read about the company in a magazine? Did someone from work recommend a product at this site? And SEM professionals will wonder what influenced the buyer to search for one of their ad-triggering keywords?
For a data guy like me, not having access to this data is a bitter pill to swallow. Luckily, the picture is becoming clearer by the day, as more companies realize the advantages gained from an understanding of the processes that convert a prospect into a customer.
Conversion Attribution and Search
Conversion Attribution (CA) is the methodology used to accredit all of the media interactions (events) in a converting action (sale, subscription, etc.) to better understand what actions helped influence the conversion and the extent to which each action played a part. Research conducted by Microsoft revealed that in many cases, “between 93-95% of audience engagements with online advertising receive no credit at all when advertisers review campaign ROI.” (http://www.atlassolutions.com/institute_engagementmapping.aspx) Attributing 100% of every conversion to the “last ad click” in a campaign with millions of ad impressions literally leaves a mountain of data on the table filled with details of various actions taken prior to that last click. This data is crucial to understanding and optimizing online campaigns – especially search.
Intuitively, marketers have long known there is a synergy with search in the digital media mix, as keyword search volume is almost guaranteed to spike upon the launch of a sizeable banner campaign. SEM professionals prepare for this surge and adjust bids and budgets accordingly. However, the first solid industry data I saw on Conversion Attribution (before it even had that name) was a Yahoo! study attempting to measure the impact that running display along with search had on both advertising vehicles. Yahoo! found that “when combined, search and display advertising deliver profoundly better results than when used independently [with] … a significant lift in onsite engagement and an increase in online and offline purchasing by consumers who are exposed to integrated campaigns that employ both types of online advertising.” For offline sales, the study concluded that when both search and display were exposed to users, they combined to lift sales 89%, versus 42% with search alone and a meager 9% via display only. (Yahoo, Inc. Press Release, Yahoo! Research Helps Close the Loop Between Search and Display Advertising, December 4, 2006)
Search marketers may also remember that with the launch of the Panama platform, Yahoo! introduced a data field called “Assists,” which allowed advertisers to see which other clicked keywords by the same user generated conversions tracked by Yahoo! Search Marketing. Prior to this, visitor reports in analytics tools may have tracked user-level site visits and showed the referring source, but this was the first model in a search engine platform. With even the most basic analysis, you could suddenly see the layers of possibilities. Normally, it is difficult to discern the value between two keywords with similar costs, volumes, positions, and conversion rates. However, what if one of the keywords consistently “assisted,” while the other almost never seemed to support the others? That is a fairly easy optimization decision if you had to choose which keyword’s budget to increase, right?
How Does It Work?
Conversion attribution (CA) works by tracking users via cookies each time they are exposed to your media or visit your site. Currently, search engines don’t allow tracking at the impression (Search Engine Results Page – SERP) level, but you can record view-through exposures with trackable media such as online display and email. Once a user converts, you can see their engagement path on the way to a conversion. A marketer may refer to any of these media interactions as “events,” “interactions,” touchpoints,” or “exposures,” but they are all talking about the same thing.
Of course, one limitation of CA is that it can only be applied to ad placements that can be tracked via cookie interaction. In the tool I currently use (Mediaplex’s Path to Conversion reporting), I can track all user interactions for display (banners, sponsorships, rich media, etc.), email, paid search clicks, natural search visits, etc. I can even see other conversion tags in user conversion paths – helpful, for example, if a user converts on a lead generation form or software download and then returns and converts on a sale. This way I even know if other converting actions are influencing a specific conversion funnel.
Here’s an example of a single conversion pathway via a Conversion Attribution report:
|Last Ad Click Credit|
|Event 1||06/12/08 – 08:40:32 AM||Ad Email ID #SPORT7||
|Event 2||06/14/08 – 07:22:14 PM||ESPN.com 728×90 banner||
|Event 3||06/14/08 – 7:22:44 PM||ESPN.com 728×90 banner||
|Event 4||06/14/08 – 07:38:29 PM||Google Ad – KW = “sneakers”||
|Event 5||06/17/08 – 11:55:03 AM||CNNS! 300×250 banner||
|Event 6||06/20/08 – 04:16:52 PM||Google Ad – KW = “rebook running shoe”||
|Event 7||06/21/08 – 9:04:47 AM||Yahoo Organic – KW = “rebook trail wave lite”||
Click thru & CONVERSION
As a first step to attribution, you need to create a model that assigns a percentage of the conversion to each interaction based on its value to that conversion. In the above example, the initial Google ad triggered by the keyword “sneakers” (Event 4) actually may be more important than the recent (in the path) ad triggered by “rebook running shoe” (Event 6), as the user probably had already made up their mind to buy that brand and type of product. You probably also want to give some credit to the email (Event 1), as it kicked off the entire user experience. The three view-through banners (Events 2, 3, and 5) may not merit much of the credit, but the CNNSI banner may receive more CA weight as it was viewed only a few days before the purchase, as opposed to the ESPN banners viewed days earlier. As well, you still can’t deny the clear value of the last ad clicked, because it was the final interaction by the user. Viewing the conversion weighted this way, it is very possible for it to look a lot different from the standard “last ad clicked” model. The following chart adds the CA model you have chosen into the scenario.
How Does Search Fit In?
Some initial theories suggested that search would end up with the short end of the stick with regards to CA. Many email and display professionals argued that it was their media activity that drove users to search and, therefore, many “so-called search conversions” should actually be attributed to them. However, search marketing requires user queries to generate the SERPs that display ads. Basic logic tells us, therefore, that users would have had to been exposed to knowledge of a term as complex as something like “Audio-Technica Folding Portable Speakers” before being able to search for it.
The good news for search marketers is that the early data tends to support the idea that search is not going to lose ground to other media, but rather will become even more important as it solidifies its strength as the final touchpoint on the way to conversions. In fact, it may shift the focus of campaign strategy from something as general as “How can I get users to convert?” to a more data-driven goal of “What can I do to get more users to search on my high-converting terms?” In fact, the latter question may well become a more attainable strategy than the former once CA becomes more readily used throughout the industry.
What You Need To Get Started
At the heart of any CA system is a tech solution – a central hub to track all of the media interactions. Third-party ad-serving platforms (such as Mediaplex, DoubleClick, and Microsoft, among others) have developed or are in the process of developing these tools. Their ability to tag ads at their source to apply tracking variables is what makes this possible.
The goal is to get as many of your media placements as possible running through this central hub. This can be quite difficult to accomplish, especially if there are multiple agencies managing a single advertiser’s display ads, email, search, etc. Each agency will already have competency and trust in the tool they prefer, and getting them to switch their operations to reliance on a central hub they are unfamiliar with may be extremely difficult. However, the more media you can track via this central hub, the more interactions you can track, which leads to more accurate CA reporting, so the effort is worthwhile.
The second thing needed is a CA model that will weight interactions in a way that makes sense. In the previous example of a sneakers sale, each event was given credit based on its value to the final conversion. However, it would be almost impossible for a company to evaluate each conversion and assign weighting at the individual granular level, so an algorithm is devised to assign weighting value to such variables as the type of ad, whether it was a clickthrough or a view-through, its timing in the conversion path, etc.
The problem is how to come up with this weighting. How much credit should you give to an unclicked banner seen more than a month before the conversion? What about a dynamic rich media ad that wasn’t clicked on, yet the user spent over five minutes engaged in the unit looking at video clips? These are difficult questions to answer, but coming up with a proper CA model is key to attributing conversions accurately.
Just some of the variables and issues to consider include:
- Type of Ad – a search ad click requires a user to query a search engine on their relevant topic, so shouldn’t that get more credit than a standard display unit?
- Clicked vs. Viewed – fairly clear that one should weigh a clickthrough more than a view-through.
- Ad Size – shouldn’t a full-page Yahoo! homepage takeover get more credit in a conversion pathway than a 120×120 pixel ad at the bottom of a page?
- Messaging - a direct response ad for “Click Now to Save 25%” may seem more valuable in a conversion than a general branded ad.
- Time Lag - what’s the value of an email sent out two months before the conversion versus one sent the week before?
- Event Path – in a conversion pathway with five events, how do you weigh the last ad clicked versus the first four?
Generating a precise CA model is the art of accurate conversion attribution. I recommend applying various models to campaign data and judge which one(s) seem to be the most accurate. When I first started doing CA, I began with testing three simple algorithms:
- the final event gets 100% of the credit
- an equal sharing of the conversion value among all events
- the final event gets 50% of the credit and the remaining events evenly share the other 50%
Each of these models had their own data pitfalls, but they were also very helpful in understanding CA. The only way to truly know if your model is accurate is to take the insights the model provides and optimize your campaigns accordingly. If the results are positive, then you are probably on the right track.
Here’s the sneakers conversion example with the three basic CA models identified above:
Coming Up With Your CA Model
My main recommendation to any analyst beginning conversion attribution is to consider each campaign individually and apply a custom model that makes sense. An international branding campaign in March is going to be different from a U.S.-targeted, retail, direct-response campaign in Q4. Take into account variables such as the scale of the campaign, the previous/current media run by the advertiser, and especially the unique aspects of the conversion goal.
A very handy tool is Microsoft-Atlas Engagement Mapping, which lets you quickly change the weighting for each variable. As well, it provides an easy-to-use visual timeline of all events leading up to a conversion, color-coded by media vehicle and presented as circles sized in proportion to their weight in your model.
(SMS – this pic can be found within the slides at http://www.atlassolutions.com/uploadedFiles/Atlas/Atlas_Institute/Engagement_Mapping/20080415Slides-EMap-Advertisers.pdf )
However, the question of how to weight each variable is still an issue. Here is an example of a CA model that assigns a point value for each variable of an event as a means of indicated weight.
Here is how this point value model would look with our sample sneakers conversion path.
Once all of the points are awarded and combined, the final step is to add them all up and divide by the total to figure out each event’s percentage of attribution. You multiply that by the conversion value to get the final conversion attribution dollar value of each event. In this example, even though the email occurred 17 days before the conversion took place, it still receives credit for 12.6% (or $11.19) of the final $89.14 sale.
Keep in mind that there is no way to know if your model will always be correct. You must use the tools available to you and your industry knowledge to construct your weighting system. At times, you may be spot-on and at other times, you may be completely wrong. However, once you tweak enough, a fairly accurate model should appear obvious to you to guide your future media buys and campaign strategies.
Sample Conversion Attribution Reports
To help illustrate the types of CA reports that can be put together from tracking data feeds, the following are six sample reports, put together from the Path to Conversion reporting system offered by Mediaplex/MOJO, where reports typically are either pulled straight from system or derived from the conversion/event tracking data feeds. When combined, the insights from these reports such as these can provide a clearer picture of how your media is interacting, which sources are influencing the most conversions, and – most importantly – help you generate an accurate CA model.
In the following examples, clickthrough events were tracked for 122 days; view- through events for 7 days.
1. Total Events Leading Up To a Conversion
Note: This chart has been truncated for size. Normally it would include columns back to the 10th event prior to conversion.
This report includes a row for each site influencing sales conversions for a “sales” conversion tag. The number of events that each site delivered as the last event prior to conversion is reported in the “Last” column. These events are then divided into the number of clicks, views, and the combined total of clicks and views. The remaining columns (“2nd,” “3rd,” “…10th”) likewise display click, view, and total events by their position in line leading up to the conversion. The far right column (“All Events”) shows the aggregate contribution of clicks, views, and combined events leading to conversions for each site.
For example, looking at the data for Google, there were 111 clicks on ads on Google that were the third event away from conversion. In total, there were 2,601 events leading to conversions on Google, and in this instance, all were composed of clicks on ads.
2. Percentage of Total Events Leading Up To a Conversion
Note: This chart has been truncated for size. Normally it would include columns back to the 10th event prior to conversion.
This report presents the same data as the first one, but relays the contribution of each site’s clicks and views in terms of percentages rather than whole numbers.
In this case, 79% of Google’s clicks that eventually led to a Google conversion were from the last ad clicked, while 13% of the clicks were the second-to-last event in all Google conversions.
3. Frequency of Interactions Prior To Conversion
This chart illustrates how conversions are distributed over the events leading up to them.
As you can see, most users converted on their second and third interaction with the media in the campaign. In fact, there were almost as many conversions with four-event interactions as with single-event conversions.
After seeing this chart, one account manager asked: “Could we make the first time a user sees an ad be their second or third time?” So, on the following campaign, we strategically ran on low-cost, high-volume (reach) ad networks before launching with our full media campaign. Search conversion rates greatly benefited from that strategy.
4. Event Versus Cost Breakdown By Category
To put total events triggered by a particular media channel into perspective, it is helpful to understand how much those events cost. This is a very easy way to weigh a media buy.
In this example, it is easy to see that Content sites may not have been a very good value – they were represented in 22.01% of all events in the conversion pathways, but comprised a higher percentage (32.78%) of the cost. Conversely, Shopping sites represented 26.86% of all conversion events, but the media value was high, as they only took up 11.28% of the budget.
5. Influence on Conversions
This advanced report shows how a campaign’s display buys influenced paid search (last ad click) conversions. As you can see, users who were exposed to a display ad prior to converting via a search ad click, represented 67% of all search conversions.
6. Winning Conversions and Total Events To Conversion By Site
The final report example again lists the sites involved in conversions. The ‘Conversions’ section displays the number of traditional, last-event conversions attributed to each site, broken down by clickthrough, view-through, and the combined total. The ‘Events Influencing a Conversion’ section details the site’s total number of events that occurred for users who ultimately converted.
For example, Valueclick Media generated 1,739 clicks that appeared in the conversion pathways for this campaign. Of that total, 1,545 were the “last ad clicked” and counted as Valueclick Media conversions. This means that 194 of the clicks generated by Valueclick Media influenced other conversions, even though they weren’t given “last ad clicked” conversion attribution.
As the examples above illustrate, there is an abundance of data to track relating to online conversions, as well as the actions leading up to and influencing those conversions. What are the correct choices in terms of which data to track and the analysis and interpretation of the avalanche of information recorded by analytics programs? What is the best model to help you decide where to place the emphasis – not to mention your dollars – in your advertising campaigns? There is no “one size fits all,” but the judicious use of conversion attribution theory and practice can help you decide on choices which make sense for your purpose. Experimentation with different possible solutions will lead you to the conversion attribution model that will help you put your efforts and budget where it can do the most good.