A Search Marketer’s Guide to Attribution Modelling

White Paper

This e-book defines attribution modeling, examines the current industry debate around various types of attribution models and describes how you can set up an attribution model within your organization to more accurately measure and value your SEM campaigns.

Get the download

Below is an excerpt of "A Search Marketer’s Guide to Attribution Modelling". To get your free download, and unlimited access to the whole of bizibl.com, simply log in or join free.


Paid search (PPC) has become an increasingly important component of the marketing toolkit for online marketers looking to generate leads and convert web visitors to customers. According to Forrester Research’s 2009 report, The State of Retailing Online, 83% of online retailers rate paid search as an effective acquisition channel versus just 51% who said the same about organic online traffic. Offline advertising was rated as effective by only 20% of the report’s respondents.

Yet many online advertisers still manage their PPC campaigns as though clicks and conversions take place within a vacuum, disconnected and unrelated to the customer engagement that occurs through other channels and at other times. PPC campaigns are actually tightly interwoven with broader marketing programs that reach customers over the entire customer lifecycle, whether on the web, through email, in store, on the phone, or any other media channel a company or brand uses to interact with its customers.

Viewing PPC transactions within the broader context of lifetime customer engagement, and attributing this information to bid rates and ROI calculations, is crucial to your company’s bottom line. In many cases, a PPC conversion is the first in a string of interactions that together comprise a long and profitable customer relationship. Sometimes PPC just happens to be the conduit through which a loyal, existing customer comes back to you for repeat business.

While this is not a difficult concept to grasp, marketers today struggle to break down the problem into actionable tactics that can be readily applied to their PPC programs. Which specific metrics should be attributed and how? What tools are currently available to help us achieve this? And which metrics can be measured with enough precision to increase real program profitability?

This e-book defines attribution modeling, examines the current industry debate around various types of attribution models and describes how you can set up an attribution model within your organization to more accurately measure and value your SEM campaigns.

Our thanks go to the following SearchEngineLand columnists, who provided valuable insight for this e-book: Chris Wine, director of product marketing, and Wister Walcott, co-founder and VP of products, Marin Software; Michelle Stern, client services director, iProspect; Andrew Goodman, founder and principal, and Mona Elesseily, director of marketing strategy, Page Zero Media; Josh Dreller, director of media technology, Fuor Digital; and Adam Goldberg, co-founder and Chief Innovation Officer, ClearSaleing, Inc.

What Is Attribution Modeling?

Every advertiser wants to find the most effective, profitable marketing mix for its products and services across all channels, including paid search. What advertisers struggle with is how to optimize paid search, in terms of accurately measuring its impact on sales and customer engagement.

Internet sales typically account for 5 - 20% of company revenue, depending on the industry, but influence 37% of sales on average, according to a Multi-Channel Research Study, co-sponsored by J.C. Williams Group, Start Sampling and the e-tailing group. As a result, measuring the bidding process in relation only to online transactions significantly underestimates SEM’s overall contribution to company revenue. In addition, paid search conversions are influenced by more than just the last click; many times a conversion happens a few hours, days or weeks after a series of clicks and searches.

Attribution modeling allows online advertisers to more accurately analyze and measure paid search’s contribution to offline conversions. The key is understanding the full sales cycle by incorporating data from advertising events that happen further up the funnel, but which still play an important role in leading to the final sale, including top-offunnel influencers such as newsletter subscriptions and store locator searches (see diagram below). By following the sales cycle all the way through to downstream, bottom-of-funnel influencers such as call center conversions and returns, advertisers can develop a more accurate picture of future revenue and conversion rates, creating smarter keyword bidding strategies.

Why Is Attribution Modeling Important?

Several factors have led to a frenzy of interest in attribution modeling, most notably the economic recession which has forced advertisers to more closely scrutinize the ROI of every marketing dollar spent. At the same time, paid search is rapidly become the most important driver of online sales. According to The 2010 State of Retailing Online report, co-produced by Forrester Research and Shop.org, retailers estimate that 35% of their sales are motivated by internet search. Online activity is having a huge impact on offline conversions. As a result, being able to correctly attribute conversions and ROI to paid search campaigns is vital.

Attribution modeling can also quantify the symbiotic relationship that currently exists between PPC and online display advertising. The Search Engine Marketing and Online Display Advertising Integration Study, copublished by Forrester Research and iProspect in May 2009, found that 27% of internet users initially respond to display ads by conducting a search on the company, product or service mentioned in the ads. That figure jumps to 49% when latency is taken into account. This data highlights the need to consider display’s value in driving consumers to search, which is exactly the type of opportunity attribution modeling can help you uncover.

Marketers can turn all of the conversion data revealed through attribution modeling into a competitive advantage, particularly during this economic downturn, and be well positioned to capitalize on the ensuing economic recovery.

By following the sales cycle all the way through to downstream, bottom-of-funnel influencers such as call center conversions and returns, advertisers can develop a more accurate picture of future revenue and conversion rates, creating smarter keyword bidding strategies.

The Pros and Cons of Last-Click vs. Multi-Touch Attribution Models

There are several types of attribution models, and each has its advantages and disadvantages. Most of the current search industry debate focuses on last-click attribution models versus multi-touch or shared distribution models. In this section, we will discuss the pros and cons of these two modeling approaches.

Last-Click Attribution Models

The Pros

Many search marketers believe that the “last-ad clicked” model is the most beneficial because search occurs far down the purchase funnel, close to the conversion. In fact, 74% of purchases can only be attributed to a single click, which by definition is the last click.

It is also very difficult to track multiple clicks or the “purchase path” that online consumers take, especially when they begin with one type of search but wind up with another. Simply put, factors that influence purchases are very complex. As a result, when marketers bucket keyword queries into distinct types, only 7% of sales can attribute significant influence to some kind of click other than the last click. Currently, Google AdWords conversion tracking as well as Google Analytics use last-click attribution.

The Cons

Last-click attribution models have their drawbacks, as well. Namely, they don’t follow consumers over extended time periods, or consider the variety of online influencers such as email, direct navigation, display ads or even organic search. As a result, a significant portion of the inaccuracies and oversimplifications now involved in crediting online media sources for conversions comes from only counting the last click.

Data from the various touch points in a sales cycle can be a veritable goldmine. Without tracking and analyzing multiple touch points, marketers can’t get the true ROI of their marketing dollars. As a result, you might not be fully optimizing your marketing initiatives.

Multi-Touch Attribution Models

The Pros

Effective SEM requires marketers to stay ahead of realtime bid markets and the evolving complexity associated with search engines. Setting up search accounts on a new search engine can be challenging and expensive while the cost of keywords continues to increase. In addition, it is difficult to measure the relationship between search marketing and other marketing programs, including organic search (SEO), offline promotions and online display without a complete view of the conversion funnel, from top to bottom.

According to Microsoft’s Atlas Institute, between 93- 95% of audience engagements with online advertising typically receive no credit at all when advertisers review campaign ROI. That means that most marketers are only looking at 5-7% of consumer media interaction when optimizing conversions. With a multi-touch attribution model you are tracking all of your user touch points and capturing more complete information about their interactions with your brands.

When all of your online media is tracked in one system, you can see all of the pathways that consumers took to finally convert. For example, Consumer #1 might have seen a display ad on Monday, clicked a sponsorship on Tuesday, received an email from you on Wednesday, and came back on Friday via a paid search ad and converted. Multi-touch attribution assigns credit to the entire conversion pathway, not just the last ad clicked.

The Cons

Multi-touch attribution models are difficult to develop and make effective primarily because there are so many variables involved. For example, you’ll need to decide what percentage of the conversion value to give to the first interaction vs. the last. Or whether display viewthroughs are worth as much as a search click. Timing and recency need to be examined as well. Did a consumer interact with your ad last week…or last month? These are all questions search marketers must consider and discuss before assigning attribute values.

Currently, most of the popular analytics solutions and paid search reporting platforms lack the capacity for tracking multiple attributions. As more analytics tools make this new functionality available to the wider marketplace, it will be easier and more affordable to track and model multiple touch points. For example, Adobe SearchCenter Plus, powered by Omniture, integrates with analytics data to provide search marketers with the ability to measure multitouch attribution on their marketing campaigns.

A Sample Comparison of LastClick vs. Multi-Touch Values

In the following example, Keyword “A” is valued using a last-click attribution model, while Keyword “B” uses a multi-touch approach. The two keywords are identical and classified as general terms, such as “toys,” “furniture,” “office supplies” or “laptop.”

Thousands of searches are done each day for each keyword, and very often more refined searches such as “outdoor toys,” “leather sofa,” “automatic stapler” or “MacBook Pro” will follow them. Like all general terms, they most often occur at the top of the conversion funnel, which we will classify as “introducers,” meaning they’re often the first step of a purchase path. We can also classify some terms as “influencers” or the middle step(s) in the path, or as “closers,” the last step in the path or conversion funnel.

On the surface, each of these keywords is very similar, in terms of search volume, click cost and where they appear in the purchase path. However, how they are valued by their attribution models varies widely, as illustrated by the following chart which tracks metrics for each keyword over the last 500 clicks.

As the chart shows, these keywords performed exactly the same:

  • They each contributed 35 times in purchase paths (introducers + influencers + closers = total contributions by that keyword).
  • When each keyword is valued solely on its ability to close, they each contributed $100 in last-click profit.

However, when valued in a multi-touch attribution approach, Keyword B is judged not only by its ability to close a sale, but its ability to introduce and influence people to buy. Therefore the model includes the number of times it was an introducer and influencer into its value calculation. This is also known as attribution management, the process of properly identifying and valuing the chain of marketing initiatives and advertisements that lead to a sale or conversion.

Both Keyword A and Keyword B must achieve a $.20 profit per click to remain viable in the search marketing plan.

Based on the chart, Keyword A is currently generating $.20 of profit per click, while keyword B is generating $.50 of profit per click.

But if each keyword is also tracked according to the maximum bid price the marketer can afford to pay per keyword (assuming its conversion rate remains constant) while still achieving the $.20 profit per click, the scenario begins to change. Keyword A is currently at the maximum bid, which is $.10 per click, whereas Keyword B could have a bid price of up to $.30 per click and still achieve its goal of $.20 profit per click (500 clicks x $.30 cost per click = $150 ad spend, producing $100 of profit, which, when divided by 500 clicks = the goal of $.20 profit per click).

If Keyword B’s ad budget is increased because of its high performance above the goal of $.20 profit per click, Keyword B will yield even more clicks and profit. Using this type of approach, Keyword B will be one of the company’s top performing keywords, while Keyword A will be stuck at a lower bid or even paused or dropped from the schedule.

This example illustrates how differently two identical keywords will be valued based upon the type of attribution model employed. Keyword A is close to being eliminated from the SEM plan, whereas Keyword B is performing far better than expected. By using last-click attribution, many valuable keywords get dropped from marketing plans.

Gauging the Impact of Multi-Touch Attribution

To further gauge the impact that a multi-touch attribution model can have on your PPC results, try the following:

  • Determine the average number of visits per conversion on your website. If the average number of visits per conversion is much greater than one, then you know your average customer is requiring more than one visit to your site to convert. If you’re using last-click attribution, you are giving too much credit to the last click and no credit to the influencers and introducers that are also integral to the conversion.
  • Bucket your keywords into three categories: introducers, influencers and closers (do this for your paused keywords as well). An introducer would be a general term that describes your business, like “toys” or “shoes.” Closers are brand terms, model numbers and exact product names. Anything that is not classified as an introducer or closer can be put in the influencer bucket.
  • Count the number of introducers, influencers and closers you currently have in your active campaigns. Are you heavily favoring your closers and ignoring many introducers and influencers?
  • Count the number of introducers, influencers and closers you currently have in your paused campaigns. Are you finding that the majority of these paused keywords fall under the introducer and influencer categories and very few are closers?
  • To establish a baseline, run a report that shows the number of times these closing keywords have converted each month over the last year.
  • Of the introducers and influencers that you have paused, select a large enough sample size to generate significant traffic and truly describes your business, and turn them back on. To measure the impact that these introducers and influencers have, count the number of conversions they currently receive under last click and more importantly, ask whether the keywords defined as closers now have more conversions, especially your branded terms, compared to the baseline report you’ve created.

While studying the numbers above, you will likely recognize two things: 1) Your overall conversion rate has decreased; and 2) Your cost per acquisition (CPA) has increased. This is to be expected. As you invest in introducers and influencers and only measure by last click, your introducers and influencers will appear to not contribute to your conversion rate. Because you are buying more advertising, you may find that your overall CPA rises but that is not always the case.

The fact is that you are in business to generate profit and overall profit can increase even while conversion rate decreases and CPA increases. The most important metric to look at is total revenue. How much of a revenue increase did you experience by adding introducers and influencers back into your campaign versus not having them at all? If you see a healthy rise in revenue you can conclude that having introducers and influencers in your campaign has been successful.

Creating an Effective Attribution Model

Attribution modeling can be a complex endeavor because there are no clear cut mathematical solutions to what is right for any particular brand or company. The first step is to begin to identify the “influence potential” of each ad click, impression and site visit. Factors to consider include the timing of the ad, decay rate, conversions, products sold and purchase amounts.

Customer lifecycle metrics focused on the value of firsttime vs. repeat customers are particularly important because they can provide actionable and long-term data about the customer’s lifetime value to your company.

New customers:

For companies where repeat purchases are relevant, the value of a newcustomer conversion is the expected value of both the upfront purchase and any subsequent purchases. This is also known as lifetime value (LTV). To quantify these metrics, you must find areas within your PPC campaign that produce a high degree of retained business after the up-front purchase or click. The key is to measure the combined LTV of all conversions for as many individual ad groups and keywords within the campaign as possible. Once enough data has been collected, these values should be attributed back to the upfront click when calculating ROI, and factored into the bid rates being set.

Repeat customers:

For accurate LTV metrics on returning customers, apply retention data to determine how much of the transaction “credit” to award various areas of a PPC campaign for the conversions they generate. Keywords and ad groups converting a higher percentage of repeat users might be awarded a discounted portion of the additional revenue because some of those sales would likely occur anyway. This discounting is particularly relevant when assigning monetary value to actions that induce future sales. If a sales-inducing action is completed by a new user that action is more likely to directly affect future behavior and that click is more valuable. In the case of a repeat user, the user’s prior experience is a powerful factor. Furthermore, because the first purchase was being over credited, the subsequent purchases need to be under credited to make the total come out to be zero, making your allocated value match your actual value.

Timing and Technology Count

Retention sales of customers driven through PPC campaigns should be analyzed over weeks and months to get necessary data. The need to conduct this analysis at the unique user (vs. visit) level also requires tools with powerful backend processing. For repeat customer, the feedback is more instantaneous, as shopping carts and analytics packages know upon close of the upfront sale or conversion whether the customer was new or repeat. While tools exist to measure this and automatically adjust bid rates accordingly, marketers can get started today with top level analysis of the percentages of new vs. repeat customers driven by their major ad groups. Once completed, this can be used to make educated decisions as to which ad groups deserve more or less credit for the conversions they produce.

Factoring the full customer lifecycle, including past and future activity when determining how much to pay for a conversion today, can greatly improve campaign scale and efficiency. And the tools needed to measure, forecast and attribute pre- and post-click activity into ongoing management are increasingly available to search marketers. Remember that a technology is only as good as the people behind it. When implementing an attribution model, make sure your staff understands how to calibrate it or that the vendor you choose offers services to assist you in building sound models.

Eight Core Elements to Consider

As you begin to develop your attribution model, consider the following eight core elements to Consider As you begin to develop your attribution model, consider the following eight core elements, particularly when tracking multiple channels.

  1. Technical resources. The availability (or lack thereof) of technical resources will help you determine how many channels to initially consider in your attribution modeling efforts. Start simply with just two channels to create a case study. Then leverage it to get the additional resources you need for modeling across more channels.
  2. Attribution plan. Once you’ve decided which channels to include in your model you should consider attribution between different products as well, for example if a consumer clicks on an ad for Product A and then later converts on Product B. In addition, if your company markets several brands, the same scenario would be applicable to multiple brands.
  3. Tracking. For each channel you’re measuring you’ll need to have the same tracking system. Keep in mind that if one of your channels is display advertising, you’ll want a tracking system that can track viewbased conversions and not just click-based conversions. Should you decide to change tracking providers, don’t forget to keep a record of the historical data.
  4. Cookie expiration. Set your cookie expiration according to what your company accepts as an appropriate length of time for the sales cycle. If you’re unsure, it’s best to err on the higher end because you can always filter the data based on the time from the initial impression or click to the conversion.
  5. Data cleaning. Set up business rules ahead of time for data that’s not appropriate to analyze. For example, if 97% of your data reveals that there are between one and 12 touch points during the life of the cookie, you’re going to have some cut off point above 12 touch points where it makes sense to scrub that data. Additionally, if you’re a global company you may want to convert any spend amounts into one currency since many engines report in local currencies.
  6. CRM data. After cleaning your data, it will be useful to marry your internal CRM data with your engine and conversion data. This will allow you to determine which purchase paths lead to the least and most desirable customers and guide your optimization strategy.
  7. Data weighting. There are several ways to allocate success across different marketing channels. The easiest method is to weight each channel equally, leaving yourself the option of taking the frequency of each channel’s exposure into account, as well as the placement of each in the purchase path (first, last or middle touch point). For example, if someone has been exposed to a display ad five times in a seven touch point path, then the credit given to display can be weighted higher. Or if display was the first touch point, you could make the case that this channel introduced someone to the brand and should be given more credit than the other channels. In addition, it would make sense to weight the data points based on whether they result in new customers or existing customers.
  8. Reporting. Several reports are critical in attribution modeling. You need to be able to see the purchase path (marketing channel, engine or site, and keyword, if applicable) by custom date range. In addition, you need to be able to see which channels introduce new customers to the brand, which channels influence, and which channels net the transaction. Viewing this data by customer type will paint a more complete picture when attempting to improve overall results.

Giving Credit Where Credit is Due

Attribution modeling is perhaps the most important tool search marketers have to prove the value of their PPC campaigns and the direct impact they have on sales – both online and offline. Here are seven key metrics to track to link PPC to offline conversions.

  1. Online/offline orders and pick-ups. Tracking how many customers searched and purchased online, then picked up their purchases in your brick-and-mortar stores provides very precise online/offline data for retailers that offer this service.
  2. Store locator page. Customers who want to visit a particular store tend to seek information on store locator pages. For these pages to be most effective and trackable, make sure you provide relevant information like “find a store near you,” store hours, directions and phone numbers, and that it’s easy to find on your site.
  3. Time spent on site. The more time someone spends on your site, the more likely they’re interested in your product or service. We know if someone searches for camera information for more than 15 minutes on a consumer electronics site that they’re interested in cameras. Online marketers should try to close the loop definitively with an online sale to these types of visitors. Don’t get “too good” at driving online searchers to physical stores; improve your conversion marketing so the customer is spending more than just his or her time online.
  4. Queries with geographic qualifiers. Some queries containing geographic qualifiers indicate offline purchase intent. For example, if someone is searching for the query “dentist in West Vancouver” they’re more than likely looking for a dental appointment in a particular geographic location. The type of insight you get will depend on your industry and may not be as relevant in other categories so use care when examining terms.
  5. Local search. Many times local PPC campaigns are designed to drive offline purchases. Using the same example as above, if the term “dentist in West Vancouver” was targeted to West Vancouver and designed to drive traffic, it is probably driving a high proportion of your phone appointments or walk-in traffic. Some Yellow Pages advertisers use dedicated phone numbers for their campaigns. Try doing the same on the search side. If not, you’re not adequately tracking the campaign’s impact.
  6. Promo codes. To track offline conversion, create a coupon or other search-specific promotion that’s only redeemable in the store (if appropriate). Customers can write down or print special codes that they can redeem in-store.
  7. Focus groups and post-purchase surveys. Run a focus group to get a better idea of how your buyers shop. Also, after people make purchases online, ask for their feedback through a post-purchase survey to get more information on buying behavior and offline impact.

In Summary

In today’s ultra-competitive marketing environment, where consumers are bombarded by advertising messages and dollars are tight, using an attribution model can be the key to better understanding PPC impact on customer engagement and lifecycle. In many cases, a PPC conversion is the first in a string of interactions that together comprise a long and profitable customer relationship.

By using a multi-touch attribution model you can track all of your customer touch points and capture more complete information about customer interactions with your brands. Many search marketers, on the other hand, use a last-click modeling approach based on the statistic that 74% of purchases can only be attributed to a single click, which by definition is the last click. However, last-click models miss the goldmine of data that becomes available when you track customers over extended time periods, or consider the variety of online influencers such as email, direct navigation, display ads or even organic search.

Whichever approach you choose, attribution modeling will help you develop a more accurate picture of future revenue and conversion rates, and create smarter keyword bidding strategies.

Want more like this?

Want more like this?

Insight delivered to your inbox

Keep up to date with our free email. Hand picked whitepapers and posts from our blog, as well as exclusive videos and webinar invitations keep our Users one step ahead.

By clicking 'SIGN UP', you agree to our Terms of Use and Privacy Policy

side image splash

By clicking 'SIGN UP', you agree to our Terms of Use and Privacy Policy