For years now, marketers have been able to buying display advertising, either by cost per impression (CPM), cost-per-click (CPC), or cost per action (CPA). More companies are opting for the CPA route, which on the surface seems like a much safer bet.

True to its word, CPA advertising means you only pay when you get an action or conversion, which can vary depending on the advertiser’s business model. Some common actions include a lead, an order, a download, a white paper view, or a demo.

The benefit is that the advertiser doesn’t pay for a non-action generating activity. Most marketers are savvy enough to negotiate the price they pay for actions from display networks. However, they don’t negotiate the terms as to what constitutes an action.

When buying display in a CPA model, there is typically a window of time in which a display provider can claim credit for an action. That window of time is defined as a look-back window — typically 30 days — from the time of conversion. So, in essence, any customer of yours that saw or clicked a display ad, as well as those that coincidentally was on a page with a display ad that they didn’t even see, in that window of time and converted is then claimed by the display network as an action that they deserve to be paid for.

Some display networks claim to have a reach in excess of 90 percent. By virtue then, more than 90 percent of your orders will be claimed by a display ad network. Does any marketer truly believe a display network deserves to be credited for more than 90 percent of their orders?

Unfortunately, marketers without the ability to see view through or post-impression data are forced to accept these terms at face value. Other marketers that have the ability to track an entire purchase path and truly see the impact of view-throughs and clicks on display ads, have the ability to define what truly constitutes an action worthy enough to pay a display network.

Let’s look at three examples of purchase paths that involve display, and how a marketer can use this data to negotiate a better definition of an action:

• Purchase Path 1: Display ad view through promoting diamond earrings -> Paid Search ad for diamond earrings -> Paid search brand term advertised in display ad -> Conversion

• Purchase Path 2: Paid search ad for diamond earrings -> Display retargeting diamond earrings -> Conversion

• Purchase Path 3: Display ad view through promoting diamond earrings -> (29 days pass) Paid search ad for diamond earrings -> Conversion

In Purchase Path 1, it seems clear that the use of display promoting diamond earrings was the impetus for the paid search that followed and the ultimate conversion. All marketers would agree the display was integral for making this action occur.

In Purchase Path 2, the person already had interest in looking for diamond earrings, they didn’t buy immediately, then were somewhere on the Internet when retargeting kicked in, and then they converted at a later point in time. In this scenario, it’s debatable if that display ad was necessary in getting that consumer to convert. The consumer might’ve already decided to buy but was just taking their time, when a retargeting ad was shown. In that case, the display ad was unnecessary. Or, the consumer could’ve been on the fence about buying diamond earrings, and the retargeted ad pushed them over and led to the action.

In Purchase Path 3, the sequence is similar to Purchase Path 1, except in this case 29 days elapsed between the display impression and the paid search ad. Most marketers, and a lot of advertising research, would agree that an impression’s rate of decay on a consumers mind is less than 29 days. Therefore, it’s unlikely that this impression had a bearing on the paid search that ultimately led to the sale.

A display ad network would’ve received compensation for the three actions in the three purchase paths because an impression was served within that 30-day window. But marketers can clearly see that the value of those impressions in the three examples is different. Shouldn’t we be paying display ad networks on some sort of scale for actions versus treating all actions that were exposed to display as being equal value?

Some of our clients have the purchase path data described above and have successfully negotiated the definition of an action to a display network. In some cases, even though an impression was served (Purchase Path 3), they aren’t paying one cent for that action.

In other cases, they’re paying partial credit (Purchase Path 2). Finally, when it’s 100 percent clear that the display impression was 100 percent integral to the action, they’re paying the full CPA.

Attribution is about giving credit where credit is due. Attribution is also about paying for credit when credit is due. With the ability to track the entire purchase path, marketers now have the ability to negotiate and define what “A” is in CPA when it comes to display.

…Continue reading article on All About ROI Magazine online

At the same time attribution was picking up steam on the conference circuit, and being discussed by research firms like Forrester Research and Jupiter Research (now owned by Forrester), the search engines began to pay attention to this topic, starting with Yahoo. In the Yahoo advertising interface, they began to represent not only the conversions at the keyword level, but also what they called an assist. An assist is when a keyword is used in a purchase path, but was not the last keyword clicked prior to conversion. After that, Microsoft, through the Atlas institute, coined the term ‘Engagement Mapping’, which utilized the Atlas ad server to provide attribution data across search and display media. Then, in 2010, Google entered the foray into attribution with the Google Search Funnel product, which performs like Yahoo, but has a lot more analytics around the data to dive deeper into these paths.

One question that we often get is, “I don’t have enough money to invest in a product like ClearSaleing, so what else is out there?”  Or, “My company is still skeptical that attribution would benefit us.  Is there a way I can prove that our customers do navigate paths and attribution would be beneficial?”  One product worth taking a look at that is free is the Google Search Funnel product. Though this product is far from perfect, and leaves a lot open to interpretation, any product that shows beyond the last click can help you to improve the performance of your overall campaigns.

From there, we usually get several follow up questions:

• What would be the first thing you focus on when using the Google Search Funnel product?
  • When we look at Purchase Path reporting from ClearSaleing’s technology and focus solely on paths that involve AdWords ads, one thing jumps out at us across our entire client base: when there are 2 or more AdWords ads used in a Purchase Path, the last ad clicked is more often than not one of our clients branded terms. A branded term is a company’s name or misspelling/typo of it. When we look at the terms that precede the branded term, they are mainly general, product specific, need specific, or a model number.  Usually they are non-branded terms, as seen in the graphic below.

(Click to enlarge)

• Therefore, under a last click world, branded terms end up stealing a lot of credit from the non-branded terms that preceded them. So, if I was going to look at one thing in the Search Funnels report, I would look at paths that end in brand terms to see how often non-brand terms come before them. Then I would look at how those non-brand terms are valued under last click and determine if they should be given more credit. We have found that when consumers use brand terms at the end of a Purchase Path, they are doing so to navigate back to the site they’ve already decided to buy from, therefore, it makes sense to credit the ads that were NOT used simply for navigation purposes.
• When I look at the path length report in Google, it shows me the number of conversions that took one click, two clicks, three clicks, four clicks and so on, it shows that most of my conversions occurred with just one click. Does this mean that attribution is not something I need to worry about?
  • There are a few things needed to keep in mind when looking at these reports:
    • This data only pertains to Google AdWords, so if a client went from a Yahoo ad to a Google ad, it would be represented in Google as a one click path, when in reality, it was a 2 click path.
    • On a similar note, paid search isn’t the only advertising source out there. So, if you’re using anything outside of paid search – display, affiliates, shopping engines, etc. – these are not being represented in the paths.
    • At ClearSaleing, we use three simple categories to place ads in: Introducers – the first ad a person clicks or sees en route to conversion; Closers – the last ad a person clicks/sees prior to conversion; Influencers – the ads in between Introducers and Closers. When we acquire a new customer that has been using a last click attribution method prior to coming to us, they cannot justify spending money on Introducers and Influencers; they can only justify spending on Closers. You are likely in the same boat. Therefore, their data in the beginning looks as if attribution does not occur.  One thing that we get our customers to do that you should also test is to activate some of the more general terms in your account, and with the use of Google’s Search Funnels, see if these types of terms show their value by being an Introducer or Influencer in other paths. Over time, our clients generally discover their customers walk down more paths than when they started with us because they have the data to support investing on ads and ad sources that introduce and influence.
    • Using Google Search Funnels, I found a collection of keywords that provide a lot of assists, but barely close. What should I bid for these terms?
      • Unfortunately, this is a really difficult question to answer. If you were a company that only sold one product, or every product you sold produced the same amount of profit, you could figure out what these assists are worth. If you sell many different products with many different margins, it’s impossible to know the value of these assists. The ideal method for evaluating these would be to know how much profit was earned on a conversion the keyword assisted, so you could assign it some profit credit and then you could come up with a bid to meet your business goals. Google will most likely never be able to produce profit figures because that would require companies to share margin data with Google, which is highly unlikely. Google could, however, take the revenue earned on that conversion (assuming you are an etailer) and attribute a portion of that to the assists, so that you could make a more accurate bid decision.

Though the Google Search Funnel product is not perfect, it does provide a lot of valuable reports that if one takes the time to use them and analyze the data, one can certainly improve the performance of their campaigns. If you have experience using the Search Funnel product, we welcome your comments, or if you have questions about attribution, as always, feel free to contact us.

All social monitoring technologies pull data from popular social sites, such as Facebook, Twitter, Blogger, MySpace, etc. Since they all pull data from the same sites, they more or less show the same data. One area of differentiation some companies have is the ability to also pull data from unstructured data sets. For example, a company like Social Seen could also pull in information from your corporate systems, like your CRM, Call Center, website and suggestion boxes.

These Social Media monitoring tools classify each “conversation” in a positive, neutral, or negative fashion using standard language protocols. If the word “good” is used, they assume it’s positive, and if the word “bad” is used, they assume it’s negative. If it doesn’t contain a positive or negative lean, they classify it as neutral. This challenge of accurately classifying conversations is a common problem in the social monitoring space. Recently, I learned that the Hebrew University developed a technology that has the ability to analyze conversations for positive or negative sentiment.  When tested against human reviewers, their algorithm agreed with humans nearly 80% of the time. So, there is hope that this common problem will be solved.

Actions That Are Clear

It’s obvious to me that if a company is using a social media monitoring tool, one worthwhile action to take is to reach out to those who have had negative experiences with your brand. Not only do you have a chance to turn this customer’s negative experience into a positive one and to hopefully retain them as a customer, but you increase the likelihood of that customer posting a positive message next time.

Another use of social media monitoring is to use the negative comments about your product or brand and address those at a more macro level. By this I mean that you can incorporate negative feedback into future product development or to create whole new business lines to address overarching complaints in the marketplace.

Many social media monitoring technologies allow you to monitor more than just your brand. Therefore, one smart action to take is to monitor your competition.  Understanding your own competition from their customers’ point of view can create opportunities for yourself to either expose your competitions weaknesses or capitalize on these weaknesses and turn their customers onto your company product offerings.

Actions That Are Not So Clear

Do positive comments equal profit? By this I mean does increasing the number of positive sentiment on the web increase profitability, and if so, is it quantifiable and by how much? If you could clearly state that for every positive comment, I get X return in profit, then one can easily determine how much time and how many resources should be dedicated to increasing positive sentiment on the web. But I’m not aware of social monitoring technologies being able to quantify this, so it is impossible to determine how much money and man power should be invested on generating positive sentiment.

If you could determine the ROI of positive sentiment, how do you generate more positive sentiment? Ideally, positive sentiment is driven by you providing a great product and service that people want to tell others about. The only way outside of creating great products and services to generate positive sentiment without using unscrupulous methods, like hiring cheap labor overseas to post positive things about you, is to ask your customers to post things on your behalf. “If you like our new X, please become a fan of our Facebook page.” Twitter’s new advertising model is another method that could potentially help you drive positive sentiment by paying for it.

There’s no way to track the influence of positive sentiment at the customer level. For example, if a person was in the market for product X and went to Amazon to read comments about the product, and while reading the comments learned about product Y, then went to a search engine looking for product Y, clicked on an ad and bought product Y, all a tracking technology would know about this customer is they clicked on an ad for product Y and converted. There is no tracking today that would also incorporate that the user read a positive review beforehand. Without this type of tracking, social media will never be able to be accurately valued, and therefore, it is impossible to know how much time or resources to dedicate to these types of endeavors.

I am by no means an expert in social media monitoring or in executing social media strategies. I am an expert in advertising analytics and quantifying the value of trackable steps in a purchase path. I am sure there are other actions that one can take from the data that social media technologies provide today, and I would love to hear what those are. Please feel free to share those below and hopefully change my opinion on social media monitoring for the better.

While companies jockeyed to accurately measure the success of internet advertising, Columbus-based ClearSaleing Inc. has become a leader in ensuring its clients are getting the most bang for their online advertising buck… continue article reading on hiVelocity Blog

In previous posts, I’ve tried to explain the reasons behind the frustrating fact that some report data just won’t ever match.  This week’s topic is a no-brainer.  To best describe why reports created by pulling data from different data sources won’t match other reports, let’s see it visually.

So, why is this a problem?  Say you haven’t configured your web analytics to track your PPC, or maybe an agency is managing your PPC, but you have not given them access to your web analytics, so you must rely on the data the search engines provide.  Inevitably, someone in your organization is going to want a roll-up or an executive summary of all the engines.  The result is then the creation of a report from disparate data sets.

The problem – each vendor report is not aware of the other vendors.  So, in the example below, each vendor report will claim credit for the entire purchase and claim all of the revenue.

As you can see, this greatly overvalues the conversion and creates an unrealistic view of the performance.
If you do try to compare the compiled report to your web analytics, there will be a problem as to how web analytics will credit that sale, as you can see below.

The real need is to conceptually ‘divide up’ the order and revenue and give everything credit.

By thinking about dividing up credit, you more accurately value the contribution of each advertising source.  With this accurate and comprehensive picture, you can really optimize your spending, ensuring that you focus your spend, time and attention on what truly is working. What you may then find clicks (and impressions) that occur at the very beginning of the ‘funnel’ are getting the credit they deserve, so you may be able to increase bids on your more general keywords or show true ROI on banner impressions.

• How to analyze your current attribution management status (or lack of it);
• Are there any good tools available?
• Which ads and sources should get credit?
• Is path analysis a waste of time?
• Will attribution pull dollars away from search?
• And much more…

View the webinar now. (Registration required)

All reports are not created equally nor are they calculated the same way.  In an attribution world, there are two types of reports, ones that factor in attribution and ones that don’t.

The two types of reports have two different purposes:

• Operational Reports
• Performance Reports

Operational Reports

Operational reports don’t factor in attribution.  Operational reports are reports used to measure how your business is doing.  Operational reports are used to ensure there is nothing ‘broken.’  Operational reports include reports run from your ecommerce system, most traditional web analytics reports and reports that are compiled from several different sources.

Performance Reports

Attribution is crucial for performance reports.  Performance reports are used to judge how well your advertising is performing.  Performance reports also take into consideration latency and latent conversions described in my last post.

Both reports are extremely useful.  The problem arises when you try to compare an operational report to a performance report.  One common example of this  that I see is when a new ClearSaleing customer attempts to compare a report from their web database, an order report or a booking report, to the ‘All Sources’ screen of the Performance tab or any other report that factors in attribution.

These two reports will never ‘match,’ nor should they.  Operations reports give 100% of credit to the last click, 100% credit to the date and source of conversion (no attribution).  Performance reports divide up the credit and attributes it to each date and source that led up to the conversion (attribution).

So before you try to compare two reports, take a step back.  Think about the report that you are trying to compare to a performance (attribution) report.  Does the report factor in attribution?  If the report is coming from your ecommerce database or booking engine, or even your traditional web analytics, it probably does not use attribution.

On Wednesday, May 12 at 1pm EST, ClearSaleing will be presenting a webinar on the most common myths and misconceptions by attribution’s supporters and detractors. Here are the 12 most common misconceptions that will be addressed during this webinar:

1. I don’t have an attribution problem
2. The last click is the chosen one
3. There are no good methods for assigning attribution credit
4. There are no good tools for attribution
5. Attribution can be done with web analytics
6. Attribution can be done in a silo
7. Attribution is about buying the right mix of media
8. Attribution pulls dollars away from search
9. Path analysis is a waste of time
10. A/B testing is effective for attribution
11. Attribution Management takes too much time to be worth it
12. Attribution Management is a silver bullet

If you are interested in learning more about these topics, please join us for this free webcast to hear ClearSaleing’s point of view on these items.

There are many, many, many reasons why, when comparing two reports, even from the same system, the numbers will not match.

The reasons are (and each week, I will cover one reason in detail):

1. Attribution Reporting

Attribution reports’ calculations update daily and divide credit (orders, revenue and profit) over all of the contributing ads, impressions and clicks that led to a sale or other conversion.  Attribution is a newer concept for many marketers.   The fact that there is now the idea of a fraction of an order is definitely a different way of thinking.

2. Non-Attribution

Most marketers are very comfortable with non-attribution reporting.  Common examples of non-attribution reports are reports that are used for operational decisions, for example a Daily Order Report from a shopping cart.  Problems arise when trying to compare an attribution report to a non-attribution report (comparing apples to oranges).

3. Multiple Data Source – Over Counting/Over Crediting

If web analytics have not been configured to track all of the different sources of traffic to a conversion level, marketers are left to rely on reports provided by the source itself (e.g. Google AdWords Reports, Yahoo Search Marketing Reports, as well as other vendors like Display, Affiliates and Email).  Because none of these sources are ‘aware’ of the other sources, conversions and revenue can be over counted, and therefore, over credited.

4. Under Counting – Under Crediting

This is a by-product of the limitations of last click.  If you are using traditional web analytics, this is likely a problem.  Because the referrer only pays attention to the last place where a visitor came from, you wind up under crediting certain sources.

5. Accuracy – Data Quality of Web Reports

Here is the big elephant in the room that no one wants to talk about.  It is possible that the data that you are relying to make important spending decisions is wrong.  The number one cause, in my experience, is a problem with the implementation.  Examples I have seen of implementation problems are code not being on every page, conversion code not being on the page or not being correct, and profiles or other settings that have not been configured properly.  It is definitely worth auditing your implementation, even if you are using a free solution.

6. Bonus:  One web vendor report to another

Even armed with the knowledge that both implementations are correct and complete, it is most often an exercise in futility comparing one vendor report to another.  The reason for this:  each company has a different idea of how metrics should be defined and calculated.  And each company has settings both internal and external that define how metrics are calculated (e.g if someone is on your site for 60 minutes are they one visit or two).

This week – Attribution

Attribution is a difficult concept.  Take pay-per-click as an example.  In order to fairly and accurately measure an ad, you have to think about latency.  On the day an ad runs, some of the conversions will take place.  But some people will come to your Web site from the ad and not be quite ready to purchase.  They may decide to return to your site a few days later and purchase.

To be ‘fair’ to your ads, you have to be able to count these conversions and credit them to the original ad (or multiple ads if several were seen along the way).

For example, say someone clicks on a Google ad on April 27^th.  They get to your site but are not convinced to purchase.  Instead they continue their research.  The next day, April 28^th, they go to Yahoo, click on an ad and purchase.

If you are attributing credit, here is what your reports look like:

April 28, 2010 – Yesterday’s Report

Here’s the tricky thing – the numbers change.  In order to give credit for latent conversions, the historic performance will then change.

April 29, 2010 – Custom Report Range 4/27/2010

There is no additional ad spend, but the .5 order is now credited to the correct day.

And this impacts all of the calculated metrics.

April 28, 2010 – Yesterday’s Report

Here’s the tricky thing – the numbers change.  In order to give credit for latent conversions, the historic performance will then change.

April 29, 2010 – Custom Report Range 4/27/2010

And even a step further, this impacts the ROI for each day.

The best thing about solutions that provide attribution reporting is that they supply this information already pre-calculated, so you don’t have to think about these types of calculations or even consider the fact that there are now fractions of orders.