Werner von Heisenberg courtesy of Wikipedia

In 1927, Werner von Heisenberg documented what he referred to as an “Uncertainty Principle” governing quantum mechanics.

The Uncertainty Principle holds that an observation cannot precisely reveal both the position of a particle at a point in time and its momentum. The more the observation reveals about one, the less the observer can know about the other. A similar principle governs paid search.

Physics sidebar: feel free to skip! If memory serves, the notion is that the act of observation impacts the object. To observe anything — by sound echo-location, sight, touch — we have to bounce something off of the object we observe.

For big things, this is irrelevant. Shining a flashlight on a tree doesn’t impact the tree. However, in the world of subatomic particles, bouncing photons or anything else off a tiny particle has a big impact.

If you want to know where a particle is at an instant in time, you have to “hit” it hard to get the answer quickly, but in doing so you impart a huge and unpredictable change in momentum to what you’re trying to observe.

The Paid Search Uncertainty Principle

This reads: “The amount of variance tolerated in Advertising Efficiency (“E”) times the amount of variance tolerated in the volume of advertising spend (“V”) is greater than some constant, K.” As the variance in one approaches zero, the variance in the other approaches infinity.

Okay, okay, enough of the physics metaphor, the idea is this: you can’t control both spend levels and efficiency metrics. The more you predetermine one, the less control you have over the other.

This fundamental law of paid search may be its least understood, as many companies determine an efficiency target and then simultaneously fix a rigid budget for how much they will spend in media. Those companies are often frustrated with their results.

Fixing a budget for ad spend translates to the following:

“I’m going to spend this amount of money regardless of market opportunities. I will flush money down the toilet if I must in order to spend my entire budget. If, on the flip side, the market conditions are such that I can make $20 in profit for every $10 I invest, I will nevertheless stop spending money when I reach my budget.”

Fixing an efficiency target translates to the following:

“I am going to spend my advertising dollars at an efficiency that makes sense for my business. I will spend an unlimited amount of money if I can do so efficiently. I will spend no money if market conditions dictate that I can’t spend efficiently.”

Stated as such, the Uncertainty Principle clearly follows.

Yet enterprise-level advertisers try to violate this principle all the time:

“We’re going to spend $1 Million this month, and we need to see an ROI of 5 to 1 (or a CPL of $30).”

It’s nice to have goals, and it’s smart to try to forecast what you’re likely to see, but at the end of the day market conditions may dictate that you pick one or the other target to hit, or you’ll likely end up missing both.

Important caveat: this assumes that your program is well managed. We’ve taken over many horribly mismanaged programs and both grown the spend and greatly improved the ROI.

But all things being equal, for any well managed program there is a trade off between efficiency and volume that has to be understood.

Paid search managers have control over many pieces of the game:

• The user searches that will and will not fire an ad
• The text of the ad fired by a particular search
• Where ads are served (both domains to a greater or lesser degree, and geographies)
• On what devices they are served
• The maximum amount the advertiser pays for a click on a given ad under almost any circumstance

As I’ve argued relentlessly in the past, we’d like more and better controls over many of the above, but we certainly do have a large degree of control.

But it is also important to recognize those factors paid search managers do not control:

• The volume of user search in a category. If the users don’t search, we can’t serve ads.
• How much other advertisers are willing to pay for clicks. If other advertisers behave irrationally, or change their bidding practices significantly, it will have a material impact on the marketplace opportunities.
• The advertiser’s selection, pricing, content quality, and promotions are often outside of the control of the paid search manager. These impact the value that can be extracted from traffic and thus the amount the paid search manager can spend for that traffic.
• Competitors’ promotions, selection, content quality, pricing, etc. Significant changes in the business practices of competitors impacts CTR, QS, and conversion rates for their ads and thereby shifts the Volume v Efficiency landscape in ways we can’t control.

The point of this isn’t to make excuses. Indeed, if the goal of enterprise search advertising  is to spend a large budget, there is no excuse for spending significantly more or less than the budget. If the goal is to hit an efficiency metric (CPL, ROI, Margin to cost ratio, whatever) for non-brand search, there is no excuse for missing that target.

However, introducing two targets often creates an unsolvable problem for paid search managers of any size.

Hitting both an efficiency metric and a volume goal requires good execution, yes, but it also requires favorable market conditions over which the paid search manager has no control.

Paid search managers must be held accountable for those elements that are within their domain, but holding folks accountable for conditions outside of their control is both unfair and unwise.

Advertising on your brand allows you to:

• Control the message. The text of the message can be kept fresh, highlighting promotions, shipping cut-off dates, whatever makes sense for the brand.
• Direct traffic. Site-links provide an opportunity for users to navigate to the next page deeper rather than just the home page, thereby aiding conversions.
• Occupy real estate. Taking up more space on the SERP means less leakage to affiliates and competitors.
• Capture incremental traffic. In some cases, our research shows that paid search ads on brand names do indeed bring in some incremental traffic. Your mileage will vary, so test this yourself, but in some instances the advertising more than pays for itself without any other considerations.

However, the fact that it is generally wise to advertise on one’s trademark doesn’t change another crucial principal of paid search marketing:

Never, never, never mix the results of brand advertising with competitive non-brand advertising.

All averages lie. Blending the results of brand and non-brand search doesn’t just lie, it will kick you in the teeth and take your lunch money, too.

Mixing the results of brand and non-brand paid search results leads to a false sense of scale and efficiency for the program which in turn creates chronic problems down the line.

The greater the fraction of overall “paid search” sales brand ads represent the greater these problems become.

Problem 1: Lack of Control

The core problem is that the paid search manager has very little control over the volume of traffic, the conversion rates, or even the cost of brand advertising. The paid search manager can and should test and adjust copy for maximum positive effect.

He or she can and should make sure site links and seller ratings are used to full advantage. S/he can and should test landing page versions and messaging to wring the most successful visits out of brand search traffic.

Assuming the good paid search manager has done these basics, they’ve exercised about all the control they have over brand search performance.

Other than guarding against some crazy CPC penalties Sid Shah and we have seen, the ads will serve at the top of the page for most folks at low cpcs and there won’t be any leverage to get more.

Traffic and conversion volume on brand search will rise and fall largely, if not entirely, as a result of offline marketing, brand awareness, friend referrals, and loyal customers navigating to your doorstep; the paid search manager controls none of those factors.

We’ve seen brand sales account for anywhere between almost nothing for internet pure-plays in start up mode, to over 90% of overall sales for very well established brands with huge offline advertising budgets.

Evaluating the performance of a paid search program by looking at overall numbers, when 70 – 90% of the conversions are on your brand, is ludicrous. You end up being praised or scolded based on factors entirely outside of your control.

Competent people want to be evaluated based on what they control, and in search that’s the performance of competitive non-brand keywords.

Problem 2: Disconnect Between Average Efficiency & Incremental efficiency

We’ve long preached the importance of understanding more than just the average ROI of your paid search efforts, but also the incremental efficiency of the last dollar spent.

Predicated on the notion of Diminishing Marginal Returns, smart marketers will always get more ROI from the first dollar of advertising than the last dollar of advertising, because s/he always picks the lowest hanging fruit available.

Understanding the incremental ROI will let you predict what the next chunk of ad spend is likely to generate.

Because of Diminishing Marginal Returns the incremental ROI is always worse than the average ROI — if it isn’t you’re doing something wrong. Our studies of Bid Simulator data find that within competitive non-brand paid search the incremental ROI is usually on the order of 60% – 80% of the average ROI.

Your average non-brand ROI may be 5 to 1, but additional spend over that same period is likely to be at 4 to 1 or 3 to 1, which may be under water for your business.

However, that disconnect pales in comparison to the disconnect between the two when the average includes brand sales.

We see instances where an advertiser sees an “average ROI” (including brand sales) of 8 to 1, where the competitive non-brand average ROI is 1 to 1, and the incremental ROI is significantly worse than that.

The brand may say: “We’re comfortable letting the overall ROI drop to 7 to 1 for now” but oftentimes they don’t seem to realize that they’re feeding money into a shredder at that point.

Let’s think about a 1 to 1 ROI. You spend a dollar in advertising to drive a dollar in sales. It would be just as efficient to have your paid search manager place orders on your website using the marketing budget to pay for it…and then throw the merchandise in the trash can when it arrives!!!

The 1 to 1 ROI may make sense for some folks, and that’s cool. The point isn’t that you shouldn’t spend for branding purposes, just that incorporating brand data hides the likely ROI on the next dollar spent.

Problem 3: Complete Misread Of Online To Offline Spillover

Hugely misleading studies, like this one, suggesting that for every dollar generated online by search ads there are $6, $7, I’ve even seen a claim of $10 in sales generated offline, make people believe that they’re spending money cost effectively when they aren’t. Not one of these studies has made the brand non-brand distinction.

Obviously, plenty of people want to find a store locator, the nearest State Farm Agent, business hours, in-store availability of what they’re looking for, etc. before they hop in their car and drive to the store or business.

They search for the brand they’re about to drive to, and you can bet your last dollar that the ad that shows up after they search for you by name isn’t capturing much incremental traffic.

Problem 4: Pulling The Wrong Levers

Because the misunderstanding of advertising efficiency is SO profound companies compound their problems by taking money out of other offline channels to pour more money into search because it seems so much more efficient.

“We’re getting an 8 to 1 ROI from paid search and only 2 to 1 from catalog prospecting; let’s dump the catalog and put the money into search.”

I’ve seen this happen more than once with budgets for catalogs, infomercials, circulars, TV ads, etc slashed to put more into search.

Sometimes that’s the right call, but often it’s based on the fundamental misunderstanding we’re describing and it leads to a death spiral.

The advertiser presents us with more money to spend in search and says: “we need sales to grow by 50% to cover the loss of the other marketing efforts”, and frequently paid search sales go nowhere or even decline because the advertising that was driving the brand search was cut off.

Problem 5: Inevitable Disappointment

As a result of the first four problems, the advertiser will invariably end up disappointed with their paid search manager or agency at some point.

Frustration comes because incremental advertising spend in search is hugely inefficient and this comes as a surprise for some reason.

Or, overall sales goals aren’t met because brand sales sag unexpectedly and the paid search manager has no way to compensate. Or the reckless spending encouraged by the combined view is a symptom of deeper business problems resulting in a trip through Chapter 11.

Or someone else in the organization — a CFO, a CEO, a new search manager — comes in, pulls apart the numbers and is horrified at the wasteful spending in competitive non-brand search that was previously hidden from view.

We’ve seen all of these scenarios play out, despite the fact that we have always encouraged our clients to evaluate performance strictly by the non-brand numbers.

With few exceptions, when goals are expressed in terms of aggregate brand and non-brand performance frustration is inevitable.

Advertisers end up frustrated with the paid search manager for not controlling that which s/he doesn’t control, and the paid search manager with their bosses for not understanding this fundamental disconnect.

The Charlatans Don’t Help

Unfortunately, many agencies, engines and others who feed on online advertising have a vested interest in blurring the brand, non-brand distinction.

These folks exaggerate the complexity of search funnels, buying cycles, and now attribution across channels and encourage advertisers to look at data more ‘holistically’ and with less granularity. That’s garbage!

Advertisers should look at data more intelligently, not throw up their hands and call it holistic marketing.

Intelligent use of data includes attribution of credit within paid search and across channels — as we do — and may include complex media mix studies to gauge optimal advertising spending levels for each program.

There are many valuable ways to look at aggregated data. Performance by category, subcategory, geography, keyword specificity, landing pages, number of ‘tokens’, and other useful classifications can reveal actionable insights.

However, folding in brand results never aids clarity and only causes confusion. While it may, in the short run, make paid search look better than it is, it will invariably make the paid search manager look bad in the long run.

Overall, the discussion was great, but chopping up the issue into facets missed the broader implications of PSE. So, I thought I’d close out my blogging for 2011 with a prescription of my own and how such changes could impact paid advertising.

Let’s start from first principles and address the question: “What do users want from a search engine?”

The most concise answer might be: we want the engine to provide results that match our intent.

When I search for “pictures of Abraham Lincoln” I want the results to be images of Abraham Lincoln, not websites that have those images. If I search for “Newton’s gravitational constant” I’d like PSE to give me the number, not websites where I might find that information. If I search for Walmart, why not take me to their website directly, or perhaps to a map if I’m searching on a mobile device?

But herein lies the rub: sometimes the user’s intent is obvious, sometimes, as with that last example, it’s somewhat unclear, and other times it is utterly ambiguous. Google and Bing try to guess based on the behavior of other users who conducted similar searches, based on the browser’s past activity, based on geography, and a host of other factors.

The engines have done an amazing job of “organizing the world’s information” as the Google folks describe it, and the intent matching continues to improve, but we’re still pretty far from understanding exactly what Susy wants this time when she searches for “Golf”.

The perfect search engine will not be able to read user’s minds either – at least, not in my lifetime – but, until we get to clairvoyance, the next best notion might be: how quickly can PSE return results that match the user’s real intent? In many circumstances, the fastest way to get comprehensive results is to ask follow up questions.

The perfect search engine should recognize degrees of ambiguity and respond with:

• Exactly what I ask for when the intent is clear,
• A range of potential options (universal search) when it’s less clear, and
• Appropriate follow up questions when the answers will get the user what they want quicker than they will get it from an array of widely disparate choices.

Many people search for “furniture” but very few actually want something that vague.

Quick refinement options, targeted appropriately might be the best response. For example: “Indoor or outdoor furniture?” “What type of furniture?”

Most good websites do this already, through navigation drop downs and shopping widgets.

The results of these refined searches would get users much closer to their end goal and save them lots of time looking through websites that don’t carry what they’re actually after.

How The Perfect Search Engine Would Affect Advertisers

Advertisers would benefit tremendously from this as well. Instead of competing for traffic they may not actually want (because the user really wanted a mission-style dining room set, and they sell inexpensive living room and bedroom furniture) they can compete for traffic after the user has clarified their intent. This would improve conversion rates, and save users time as well.

Let’s think about “flights to Cancun”. Spitting out websites that will then ask questions about dates and accoutrements is okay, but the future might see the engines take all the information needed from the user and then show the choices from the airlines directly bypassing the OTAs entirely.

In eCommerce, perhaps the PSE would provide the same type of navigation options to get user’s to the product level before shipping them off to the advertiser’s site.

PSE has to figure out how to present the right questions and options, and exercise judgment in when to seek follow up and how many refinement questions to offer.

Clearly, we don’t want PSE to turn into Mr. Clippy.

Effectively PSE could evolve into the uber-website, taking most of the weight off of internal site navigation because the users come into the advertiser’s site at the right level of depth in the first place.

Perhaps the perfect SERP would have a handful of “best guesses” at the top — paid or organic or some of each — with a prominent set of pull down refinements targeted to the category.

User: “Attorney” PSE: “Here are some general listings for “Attorney”, but to expedite your search, please select the type of attorney do you need from the following list.”

Perhaps today’s engines are heading down this path already. Quite possible that they’ve been down this path and found it to be less user friendly than I imagine (not PSERP but LSERP where the “L” is for lousy).

Perhaps this isn’t as different from “instant” as I think it could be.

What Does The Perfect Search Engine Look Like For Advertisers?

The other question the panel mentioned but didn’t have time to flesh out was: is there a difference between what users would find to be PSE and what the advertisers would find to be PSE?

I think the answer is “no”. Satisfying the user’s intent on the first non-engine page will save users time and frustration, encouraging more search. Getting users what they want on the first click will greatly improve conversion rates for advertisers by better qualifying the traffic.

Increasing the value of the traffic will increase the price advertisers are willing to pay which will more than compensate PSE for the reduction in fruitless clicks. Win-Win-Win.

A New Year’s present for all of us?

There are plenty of good reasons to do so regardless of the ROI, but those keenly interested in ROI and vexed by the concept of paying for traffic on their trademarks still ask the question.

We’ve done some research on this topic that we’ll share below. We focused on trying to answer a couple of specific questions:

1. Do brand ads cannibalize traffic from organic listings? and
2. If the answer to the first question is: “it depends”, on what does it seem to depend?

First let’s define our terms. What do we mean by ads cannibalizing traffic?

The impact on traffic and eventual revenue of combining ads with organic listings can take many forms.

A concept that might be useful to discuss is: The incremental traffic ratio (“ITR”) of ads. That is the fraction of the reported clicks on an ad that are in fact incremental.

Let’s say that absent an ad, we expect 1,000 clicks on an organic link on a given day. Ads could have 3 general effects on traffic.

• Destructive Cannibalization: we flip on the ad and find that the overall traffic declines.

For example:

The ITR is the ratio of incremental traffic to measured ad traffic, which in this case is -200/400 = -50%

This might also be referred to as “1 + 1 = 0.5″

• Constructive Cannibalization: presence of the ad increases the overall traffic volume, but some of the traffic on the ad is ‘stolen’ from the organic listing, so the ad can’t take credit for all of the traffic it sees, only the incremental traffic driven.

For example:

Pretty clearly, Constructive Cannibalization can range from one edge case we’ll call No Cannibalization whereby 100% of the traffic on the ad is incremental, to the other edge case of Perfect Cannibalization where the presence of the ad doesn’t destroy traffic, but doesn’t increase it either.

In the language of ITR, Constructive Cannibalization exists when the ITR is between 0% and 100%.

Clearly, somewhere in the Constructive Cannibalization ITR range from 0% to 100% the advertising spend on brand ads likely shifts from a negative ROI to a positive ROI.

• Amplication: The traffic through the sponsored link is 100% incremental AND the ad is responsible for increasing the traffic on the organic link as well.

For example:The best of all scenarios, the ad creates more love all around.

When ITR is greater than 100%, an ad is having more positive impact than what is readily apparent.

Many folks with an agenda conflate Constructive Cannibalization with Amplification, using the fact of incremental value greater than 0 to imply that 1 + 1 = 3. Amplification may exist in some instances, but the existence of incremental overall traffic doesn’t imply amplification.

7 Tips For Designing A Good Test

First, we must define what constitutes a brand ad and think about whether there are different classes of brand searches. There may be some search phrases that clearly demonstrate navigational intent: they’re looking for you!

There may be others where the navigational intent is less clear: some are looking for you, some are looking for something else with a similar name. Classify these ads by type so that we may see whether the incremental value of the ads is perhaps higher when the navigational intent is less clear. Also, classify the corresponding organic listings by position on the page.

Hopefully, your organic brand links are at the top of the page, but it isn’t always the case if your brand is also a competitive search term for others (eg “online poker” is both a brand and a search term for the category).

Second, establish the measurement mechanics. You need to be able to see traffic by keyword by source (which engine). It’s best to use the same platform (attribution system or analytics package) for consistency.

Third, establish the measurement metrics. It’s best to measure conversions within X hours of a click as a benchmark. However, conversions are a much thinner data set than click traffic. Reaching statistically significant data can take much more time if one insists on using conversions rather than traffic, so traffic with an understanding of average conversion rates may get the most valid results withing the shortest period of time.

Fourth, establish the measurement methodology. Unless you actually serve the ads (that is, unless you’re Google or Bing) the only way to do this is by turning off ads and watching performance carefully. It’s best to cycle ads on and off several times to try to separate anomalous performance (I feel compelled to use the term “exogenous factors”) from cause and effect. One hour on, one hour off, one hour on, one hour off; or week on, week off, week on, week off, etc.

Fifth, make sure off is “off”. It’s best to not just pause the brand ads but to add brand phrases as account negatives to be sure that no ad is showing.

Sixth, decide how you’ll analyze the data. How will you account for expected hourly, daily, weekly, seasonal variations in search volume. A couple of good ideas: benchmark against Bing.

In other words, only conduct the test on Google, and use the Bing brand search (paid and natural) as a control. Overall site traffic can be a control as well. This is not trivial, but with sufficient data you should be able to get a very good sense of the interaction without having to hire a statistician to perform mathematical wizardry. Deciding how you will look at the data before you collect it helps keep you honest, and prevents the analyst from cooking the books to produce the desired outcome, whatever that might be.

Seventh, run the test.

We performed two such tests for different clients recently, and the results were very interesting, and not what we expected.

We expected to find that brand ads cannibalized all or almost all of their traffic from brand organic listings. We didn’t expect Destructive Cannibalization but we did think we’d see ITRs of less than 10%.

In the first test, we were stunned to find that overall 67% of traffic on brand ads was incremental. Only about a third of the traffic appeared to be cannibalized from the organic link. Interestingly, the phrases we categorized as clearly navigational did have a markedly lower ITR (55%) than those that were more ambiguous (ITR = 90%).

However, 55% ITR for navigational searches was still much, much higher than we expected to see.

The second test turned out quite differently. The data was somewhat thinner, and we’d like to have more to work with for this one, but from every angle the ITR of brand ads was essentially 0%. In this case the advertiser had never advertised on its brand, and we ran ads for a period only to find the traffic to be pretty much completely cannibalized from the organic listing.

The hypothesis we want to test is this: The more navigational the search the less likely the traffic is to be incremental. However, if outside factors make it difficult to navigate to your site, brand ads can absolutely bring in incremental traffic. In the second test, the advertisers’ brand name was very unique and unlikely to be confused for anything else. There are no other ads on the page, there are no close cousin websites that might create confusion, hence the ad provides navigational help that no one really needs.

In the first test, the advertiser’s name is closely connected with a very well known, related entity. There are many other ads on the page for brand search and many cousin links that could draw in users easily. Here the navigational assistance provided by an ad is much more necessary to make sure you get the traffic that is trying to find you.

We’d expect that advertisers with wide distribution networks like product manufacturers, entertainment, and travel will find more incremental value in brand advertisements in that there will be many other companies competing well for that traffic. “United flights to Las Vegas” would prompt ads from many OTAs with attractive offerings, not just United Airline’s ad.

Conclusion

Since N=2, here, we really can’t say we’ve done much more than establish an hypothesis for further testing. We do believe we’ve defined a repeatable methodology for studying these data in a sophisticated and useful fashion going forward and that this methodology can be extended to non-brand competitive search as well.

If 10% of sales attributed to a paid search ad are cannibalized from an organic link, doesn’t that change its ROI? If, in non-brand search, ads and organic links amplify one another, isn’t the ROI better than what it appears to be? Shouldn’t we act on that information?

One conclusion we can draw definitively is: answers lie in the data, not in marketing blather. Average values are meaningless as each case varies. If you wonder about whether you should or shouldn’t advertise on your brand there is only one way to find out.

The statistics to tease out exogenous effects ‘properly’ is non-trivial, but don’t let the perfect be the enemy of the good enough. You can get pretty close to the ‘right’ answer with smart application of basic math.

Happy Testing!

Given that my last post for SEL was on Quality Score and suggested that landing page quality was mostly a hammer used to beat up bad actors and didn’t have much meaning for legitimate businesses, this announcement was a bit embarrassing. It also made me curious as to how quality will be defined.

I got a chance to speak with Jonathan Alferness, director of product development on the Google ads quality team, to delve deeper into the purpose of this initiative and try to glean insight as to how landing page quality would be measured.

Jonathan was very helpful with respect to the former question and understandably protective when it came to the latter.

Below, I’ll make a clear distinction between what Google is willing to say publicly about landing page quality and what I believe this means. I’ll summarize what Google has said either through my conversation with Jonathan or elsewhere in print in italics. My speculation will be in normal font.

Why Did Google Do This?

Jonathan made the case to Pamela and to me that the intent is to improve user experience.

When landing pages don’t line up well with the user’s search, the user has a poor experience, and ultimately that’s bad for everyone.

It’s bad for the user because they don’t end up where they expected to end up; it’s bad for the advertiser because they clicked on an ad and ended up somewhere they didn’t want to be, winning the advertiser no great brand impression; and it’s bad for Google, because that user is more likely to seek a different search experience and less likely to use Google sponsored links in the future.

This makes perfect sense. As I argued in last month’s post, Google maximizes its immediate term revenue by making QS 100% based on anticipated Click-Through Rate (CTR). Factors related to landing page or anything else reduce Google’s revenue per SERP view by reducing the importance of CTR.

However, poor landing page experiences might reduce Google’s long-term revenue by training users not to click on sponsored listings. That could jeopardize the business.

By creating a finer gradation between “your landing page stinks” and “it doesn’t stink” — essentially where we’ve been with landing page QS — Google rewards advertisers that pay attention to landing page decisions and creates another incentive to provide a great user experience.

In the long run, better landing pages lead to higher conversion rates, which in turn means an increase in the value of traffic, therefore allowing higher bids and correspondingly more traffic at the same efficiency. It’s a beautiful virtuous circle.

It’s also at least part of the reason that Google bought Urchin and made Google Analytics free. Giving advertisers the tools to diagnose user behavior and improve website effectiveness should ultimately lead to better monetization of traffic and higher bids in the paid search auction. Smart. Very smart.

How Much Weight Will Be Given To Landing Page Quality?

More.

Google isn’t saying much on this topic, and what they’re saying is a bit of a mixed bag. Jonathan told Pamela that ads with high-quality landing pages will get a “strong boost”; he told me it would be a “slight boost”; and on the AdWords blog post it says: “We expect most campaigns will not see a significant change in overall performance.”

They began the roll-out in South America a month ago and appear to be rolling it out to smaller accounts in the US first. This suggests to me that they think it could have a material effect that they want to watch very carefully. Perhaps not.

Making even minor changes to software should be tested thoroughly before roll-out. Perhaps most campaigns won’t see a significant effect because most landing pages are roughly equivalent, so this change will mostly impact the “outliers”?

It’s hard to imagine that Google will sacrifice revenue on this, either. They’ve likely done the math to see how much gradations in landing page quality affect the propensity of people to use sponsored links in the immediate future, and have tuned this dial to be revenue positive even in the short run. It will go on to be yet another one of Google’s revenue dials, subject to periodic “tuning.”

How Will Google Determine ‘Quality’?

Jonathan said: “Google is focused on both relevance and content on the landing page in addition to the user response to such landing pages.”

I pressed on this with the question: “Will advertisers with Google conversion tracking on their sites have an advantage over other advertisers, given that conversion rates are a very clear signal as to whether users found what they wanted?”

His response was: “There is no special information derived for some customers but not others. We apply the principles of landing page quality and relevance to every ad and every landing page. The advertising system is dynamic, as an auction takes place for each query.”

The behavioral signals that Google traces must therefore be related to what they can track for all advertisers, not just those with Google conversion tracking: bounce rate, propensity to search again in short order, time on site, etc. I’d love to know how much of this behavioral data is gathered through Google toolbars as a sampling method rather than from watching browsers return behavior from Google.com, but I don’t know that.

With respect to the relevance and content on the page, Google means to determine: Does this page match the user’s intent? Behavior is one guide, but content on the page will matter as well. Pages that seem to be about the topic of the user’s search will do better here than general landing pages.

It would be surprising if something as simple as blowing the user’s search string or the keyword back at the user dynamically on every page would be meaningful in terms of content on the page unless it affects user behavior.

I suspect the advertiser will be rewarded for landing a “ball peen hammer” search on a page with ball peen hammers on it, rather than a page with only wrenches.

High quality landing page for “ball peen hammer” search:

Previously high quality landing page in Google’s eyes — ie: “not spam, and loads fast” — now considered low quality content:

Another High quality landing page for the same search: I bet they like a product page just as much

Whether the advertiser selects the ideal ball peen hammer page chosen among several options is probably beneath the level of detail they’re getting into at this point on the “content on the page” piece of the equation.

I’d be stunned if Google tried to differentiate between the first and third based on content on the page, but our experience suggests that the conversion rates will probably be higher for the first page.

One worry that I expressed was that Google would bake in some preconceived notions about good pages and bad pages.

For example, we’ve heard it expressed that Google thinks taking someone from a Google SERP to someone else’s SERP is a bad experience, therefore a website’s search results pages might be frowned upon as a landing page choice.

Would Google arbitrarily say this was a poor quality landing page for a “ball peen hammer” search?

That would fly in the face of our experience, which suggests that often a good site-search results page outperforms a well-chosen sub-category page in terms of conversion rates.

Similarly, a “view all” page often outperforms landing folks on page 1, but it’s conceivable that latency considerations due to loading “all” the images will trump what we know to be better performance from a conversion rate perspective.

Google could also conceivably decide which page on an advertiser’s site ranks highest for them in organic listings on a given search and reward folks who choose that page for paid search ads as well. That would rankle those of us who believe smart humans who pay attention pick better pages than Google does. More on that in a later post.

Jonathan responded that Google wasn’t baking in preconceived notions, that relevant content and user behavior would define the landing page quality.

Conclusions

Within e-commerce, this could serve as one more penalty for those who aren’t mass merchants. If you sell office chairs of the $500 variety, not the $40 variety, you’ll likely find your conversion rate on the term “Office chairs” to be worse than a “competitor” who sells the $40 version even with qualifying ad copy to wit “Starting at $499.”

The conversion rate is lower, the bounce rate is higher, so behavioral signals after the click incur a landing page quality penalty on top of the CTR penalty incurred by the qualifying ad copy. This double whammy may make it even tougher to compete.

Maybe that’s the “right” result for users, but maybe it takes away diversity of choices for those who actually read the copy.

At the end of the day, it appears likely that this change will primarily reward advertisers for doing what they should have done all along. Pick the best page among the available choices for your PPC landing pages, and design those pages for maximum effectiveness in monetizing traffic. If you’ve been doing that since day one, expect a bit of a boost in traffic in the coming weeks.

(Stock image via Shutterstock.com. Used with permission.)

The exact recipes for Quality Score on Google and Bing are unknown, and it is that uncertainty that creates so much room for mischief and blather. Uncertainty causes some smart people to rack their brains trying to figure out what’s in the black box.

Other folks in the scoundrel category seize the opportunity created to present themselves as the QS Master who knows the secrets that can lead to riches. At least one such pretender claimed that instead of managing bids he preferred to manage through QS manipulation. Aye Carumba!

The name itself allows some misguided folks the opportunity to create garbage ‘diagnostic’ tools that identify ‘opportunities’ to ‘optimize’ one’s account by identifying ‘poor quality’ ads.

Let’s clarify a few things.

According to Google, Quality Score is Calculated from:

• The historical clickthrough rate (CTR) of the keyword and the matched ad on the Google domain; note that CTR on non-Google sites (such as AOL.com) only ever impacts Quality Score on our search partners – not on Google
• Your account history, which is measured by the CTR of all the ads and keywords in your account
• The historical CTR of the display URLs in the ad group
• The quality of your landing page
• The relevance of the keyword to the ads in its ad group
• The relevance of the keyword and the matched ad to the search query
• Your account’s performance in the geographical region where the ad will be shown
• Other relevance factors

CTR

We can essentially combine the first three bullets with the ‘relevance’ and geography bullets to get: Predicted Click-Through Rate. This is the most important piece of the puzzle.

When you look at the math, Google maximizes its revenue per SERP impression when it considers only bid and CTR as factors. Therefore: Any component of QS that diminishes the role of CTR in this equation loses them money, at least in the short run.

The engines aren’t going to swing too far away from CTR as the ultimate measure of quality, because doing so costs them money.

Landing Page

If the landing page provides a lousy experience the user will have a bad impression of not just that site, but the experience of clicking on ads as well. Short term, this doesn’t impact the engines, but in the long run it certainly could.

What landing page provides a lousy user experience? One that is slow, one that is unrelated to the search term, one that is chock full of ads.

If you run a legitimate business and the ads link to the right pages on your site, and the pages load reasonably fast, you’re home free. Moving pixels around on the page isn’t going to lead you to QS Nirvana.

That’s all there is to it.

So if it is really all about anticipated CTR, why might my CTR be low?

It’s important to create tight Adgroups and write good copy. Well worth the time to do this right.

But here’s the thing: when people see a low QS they often mistakenly assume the reason is number 1 or 2, when in fact it’s often 3, 4, or 5.

People who don’t understand this waste countless hours trying to find the magic grouping and magic wording to get them 10′s. In many instances, there isn’t anything you can or should do to “fix” the “problem”.

Qualifying Ad Copy

Let’s say your business sells high-end jewelry. The mass market jewelers will tout discounts and “Ruby earrings starting at $15…” Your earrings start at $300. You’ll have a higher quality score if you tout discount percentages and offers, but doing so will torpedo your conversion rate.

It’s likely better to qualify the traffic with “Starting at $300″ to weed out the discount hunters, but doing so will hurt CTR and therefore QS. Testing will reveal which makes the most sense. As I’ve argued previously, this game favors the mass marketers over the high-end merchants.

User Search Is Ambiguous

Someone searches for “Yamaha”. You sell Yamaha keyboards, but you don’t sell motorcycles, or boats, or stereo equipment. In all likelihood, unless you’re Yamaha corporation, your QS is going to be awful, because some fraction of people typing that query are looking for each of those  and the QS for merchants of any of the single categories will have poor CTRs.

Low Commercial Intent

Someone searches for “Central Park”. You own a hotel near Central Park, but your QS on that keyword stinks. In all likelihood, this is a function of the fact that most of the people typing that phrase aren’t looking for hotels, trips, restaurants or anything else related to commerce.

Maybe they’re looking for a map, or directions, or history, but the CTR will be lousy because too few people typing that search have an interest in any specific ad.

The fact that the quality score is low doesn’t mean the ad text is sub-optimal and it doesn’t mean you can’t profitably advertise on that keyword!

People worry far too much that having low QS keywords in the account or adgroup will hurt the QS of other keywords. The engines have a huge  interest in preventing that from happening.

Anticipating CTR accurately is a fundamentally important activity for them tied directly to their bottom lines. They’re going to get this right because it matters to them, and they have a ton of really smart people working on how to get it right.

Again, poor wording and poor grouping absolutely can be problems and should be addressed, but too much time is spent applying medicine to keywords that aren’t sick.

3 Request For The Engines That Would Help Us All

1.  The Reset Button

We’ve inherited accounts that have been poorly managed historically and have to fight an uphill battle initially to overcome poor QS history. Setting up a new account is somewhat of a hassle and doesn’t address the piece of the problem tied to the advertiser’s display url.

It would be great if there was an “erase history” option that could give the new management team an opportunity to right the ship quicker. Historical data is often a good predictor of the future, but in cases where new management has taken over, the history can be a lousy predictor. Getting the CTR predictions right benefits the engines and the high quality managers and advertisers.

The challenge is that bad actors will want to hit that button every day to start over, so I propose that each advertiser can only hit that button once a year. Maybe only the engines can hit that button based on a verified and reasonable request.

2.  End the promotional event penalty

Some advertisers have promotional events that they want to tout in the ad copy, but we find that changing copy frequently incurs a penalty because the engines have no history to go on with respect to the new Keyword-Copy combination so they erase the good history and start you back at square one.

By the time the short-term promotion is over, the engines may not have figured out yet that the new copy actually boosts CTR. This is a tougher engineering nut to crack. We can’t reasonably ask the engines to always assume a change is for the better because in a transition from control copy -> promotional copy -> control copy, the second transition likely hurts CTR.

Perhaps copy blocks can be deemed promotional or control. The engines could then track the average CTR impact of transitions from one to the other and make better assumptions about the CTR of a new change based on that history. Those who try to game the system will have data trends that indicate that there is no historic difference between the two, so no help.

3.  Give us a scatter plot

In a big program, it’s tough to tell whether CTR and QS are low because of the copy (#1, 2, or 3 above), or because of the user intent. One good way to know would be to reveal a scatter plot of the QS for all the ads in the top 10 positions.

Without identifying which QS went with which advertiser, we could still get a sense of whether we’ve done a poor job of matching keyword to copy, or whether no company scores well on this particular Keyword.

For example, the plot below would tell us that different copy could help QS. It may not be wise for us to ‘fix’ this copy if it qualifies traffic, but at least we’d know it’s an issue.

This plot would tell us that no advertisers have a great score on this keyword and we’re likely wasting time trying to ‘fix’ a problem that doesn’t exist.

Some of us geeks would prefer a numerical value like standard deviation rather than graphical representation, but perhaps both would be ideal.

We all benefit from working towards better performing programs. These changes will help advertisers and engines alike, by getting better predictions of CTR and helping advertisers identify where to put their finite management resources.

We get really excited anytime the engines give us new information to analyze or tools with which we can fine-tune our approach.

Google has just given us a treasure trove of new information: click level data on an ad’s position. In the past, we’ve had to rely on observations and average position day to day to get a sense of where an ad is appearing on the page. But we know that averages lie.

By giving us information identifying the position on the page for each click we get a whole new window on auction dynamics.

How It Works

Similar to the Google Click id, gclid, by flipping a switch in your adwords account, Google will append an ad position parameter called ‘adpos’ to your destination url. We can then parse this information as it’s passed to our clients’ sites and analyze from there.

The values passed in the adpos parameter reveal the page of the results off of which the ad was clicked, whether the ad appeared on top of the organic listings or to the side, and its overall position rank on that page. All of this is passed in one text string.

• 1s3 means: page 1, on the side of the page, third position among ads.
• 2t1 means: page 2, on the top of the organic listings, first position.

What It Reveals

Matt Mierzejewski, our VP of Paid Search, flipped that switch immediately for one of our clients to start gathering data, and I couldn’t resist digging in, even though we only had a couple of days worth of data.

To wrap my arms around this, I narrowed my study to 8 ads representing 4 different keywords to reveal among other things what impact match type might have on ad serving.

Pages

Of the 471 clicks in my sample: 97% happened when the ad was on page 1; 2% on page 2 and less than 1% on page 3 or more.

The academic in me wonders: if an ad is in position 2 on page 2, how does that figure in the average position calculation at the end of each day? Does that count as position 2, or position 12 or whatever it might be given that the page 1 ads were ‘above it’? I suspect the former, but in truth given that 97% of the time we’re looking at page 1, it probably doesn’t matter.

Position

Fascinating, fascinating, fascinating!

Keyword 1. Adwords reports that Keyword 1 on broad match had an average position of 3.7

But take a look at the positions from the perspective of click traffic:

This is just jaw-dropping, IMHO!

The Average Position reported is 3.7, but the weighted average of the clicks by position is 2.6. Bear in mind, the reported Average Position is based on impressions served, not ads clicked. The click-wise view will always be a lower ordinal number reflecting higher position on the page. This is a product of the higher click-through rates of higher positions.

That said, for a reported average position of 3.7, 67% of the clicks came when the ad was in position 1 or 2!

Boy-oh-boy, do averages not tell the whole story!!!

Let’s take a look at Keyword 2, also on broad match. Average reported position for the days covered: 4.9

Even more dramatic! The average position of ads clicked is under 2.1!

These results are so startling to me that I wonder if I’m comparing apples and oranges. The average position data comes in day long chunks, and the click data I’m looking at starts midday and ends midday.

I also wonder if Google appends the parameter for network partner clicks…I think the answer is yes, but…Also worth noting that our own day-parting activities could be driving this as much as auction dynamics.

This view of data will have us all scrambling to revisit what we thought were settled questions about paid search. One obvious example: we long ago concluded that conversion rates don’t appear to vary much by position. I’d be surprised if we found variance given the new view.

Hal Varian’s study, with the crystal clear view of the auction that only available to Google, suggested that conversion rate is relatively invariant with position, and that’s actually somewhat contrary to Google’s best interest — they’d benefit from saying higher positions on the page convert better. Still, the new visibility means we can and should take another look.

Placement

For ads that are primarily in position 1, what fraction of the clicks come from impressions when the ad is on top of the organic listings versus on the side?

Keyword 3 on broad match had 114 of its 140 clicks placed when the ad was on top of the organic listings. The same keyword on exact match saw all 112 of its clicks on top of the organic listings.

This seems to be the trend. There is somewhat less variance in both position and placement when the ad is on exact match than when it’s on broad match. That makes sense, as the competitive landscape is likely different for every broad match variant.

It’s worth pointing out that we don’t get a perfect view of the page layout from this.

We don’t know whether 1s3 means the ad was third from the top on the right, or at the top of the right column with 2 ads served over the organic listings, or whether it was below Product Listing Ads. Nevertheless it gives us much more information at the click level than we’ve had before.

So What?

Well, that’s a hard question in this case.

What are we going to do with this data? I don’t know…yet. Perhaps it will shed light on how national ads compete in various local markets?

We may be able to see using geographic overlay data that clicks from certain locales are coming from significantly lower on the page than the national average, indicating stiff competition from local brick and mortar stores. That might suggest creating a geo-targeted campaign for that region with separate bid logic and copy to provide a competitive edge.

That’s just one idea. The point is that every time Google has given us greater visibility and greater flexibility the leaders in the industry have been able to figure out ways to raise the performance bar.

I expect this to be another opportunity for differentiation.

Today, I want to pontificate a bit on the subject and then dive into some brass tacks, practical methodology.

Pontification

1. Cookies are an imperfect tracking device, but remain the best thing to happen to online marketing.
2. As cross-device tracking becomes increasingly important with people surfing on mobile devices but putting off form completion for desktops and notepads, cookie data will need to be de-duped when possible. Site login/user recognition will allow advertisers to connect user behavior across devices, and advertising managers will need to incorporate that data, like call-center sales, in the near future.
3. First-party cookies will be critically important with the growth of mobile and the market share possessed by Apple’s mobile devices. Safari browsers by default reject 3rd party cookies. Re-directs are the cleanest mechanism for establishing first party cookies but that data must be supplemented with impression data to create the true picture.
4. The notion of a cookie window is awkward, but convenient. To say that an ad, paid search or other, should get some credit for a conversion event that takes place X days after the interaction with the ad, but no credit after X + 1 days is inelegant at best. Indeed ‘first touch’ attribution, and any system weighting ‘first touches’ heavier than others needs to think about the fact that with windows in place ‘first’ is a challenging concept. Some notion of decaying credit makes sense but poses material technical challenges and needs to be managed as part of an attribution methodology. Statisticians use the highly technical term ‘icky’ to describe methods of handling time lag effects.
5. If hard windows are necessary, it’s best to handle them in reporting rather than through cookie expiration. By setting a permanent cookie and applying a fixed number of days of credit in reporting, you can go back to the database to conduct what if analysis on longer windows. If cookies expire, you lose that visibility.
6. The duration of a hard cookie window should be influenced by the conversion cycle of the business. The longer the average consideration cycle, the longer the window should likely be.
7. The business model should impact the window as well. Interested parties are scrambling to prove the connection between online advertising and offline conversions, but it’s important to recognize that that connection works both ways. For businesses that spend the majority of the advertising budget offline, the offline advertising can also impact your view of online advertising’s effectiveness. If your company or client spends 90% of it’s marketing budget on TV ads, or Newspaper circulars or catalogs, and hits a wide chunk of the population every week, a shorter credit window for online advertising might be more accurate.
8. You may be able to identify the right cookie window in the data. We’ll talk about methods of doing so below.

3 Methods For Establishing Cookie Windows

One beauty of online marketing is that the data usually makes intuitive sense: brand search traffic converts at a higher rate than competitive search traffic; specific search keywords convert better than more general search terms; etc.

Letting our marketing intuition guide us, there are certain patterns in the data that can help steer us to the right cookie window.

I fleshed out all of these in more detail a few years back, but it’s worth touching on the concepts here.

1.  The Click To Conversion Curve

Here’s the notion: An ad is likely to impact your behavior more in the immediate future than it is in the distant future. Producers of infomercials can talk about this subject in great depth.

Therefore, we would expect that the greatest number of orders following an ad interaction would happen on the same day, the second greatest number would occur the next day and that that trend would continue for as long as we extend the window.

However, that is not what we usually see. What we usually see is the number of orders declines up to some point and then levels off, with roughly same number of orders following everyday thereafter.

That suggests to us that the behavior of that group can no longer be explained by latency from exposure to the ad, and instead their behavior must be driven by something else. That’s the point at which credit should no longer be attributed to the ad.

Graphically, it looks like this:

2.  Average Conversion Value

It makes logical sense that one reason people spend more time between the initial visit to your site and conversion is that they’re making a larger financial commitment and want more time to consider the decision and shop around.

If that was the case then in e-commerce, we’d expect to see the AOV of orders placed increase over time with the people ordering 100 days after the initial visit having a higher AOV than those who purchased 50 days later, etc.

That pattern also breaks down over time. For most e-commerce advertisers, the AOV grows up to a peak value then drops back down to an average. This again suggests that those making purchase after that peak are likely on a different shopping mission, no longer related to the initial visit.

For e-commerce sites, those who see method #1 and method #2 point to the same window duration can be fairly confident with that decision. Obviously, the graphs above are highly idealized, but the trends are pretty clear in the real data for most advertisers.

3.  Keyword – Purchased Item Agreement Rates

This is extraordinarily difficult to study comprehensively, but sampling can give a reasonable sense. We’d expect most people to buy something generally related to the search term they used to find us. We’ve done some research on this ourselves, and the variance advertiser to advertiser is material, but generally something on the order of 80% of orders have items clearly related to the search phrase.

However, as the time between click and conversion increases, the match rate tends to maintain for a time and then drop off sharply. This indicates people are likely on a different shopping mission driven by something other than that aging ad experience.

Conclusions

Tracking cookies are an imperfect but still tremendously valuable piece of the online marketing equation. It is worth spending time analyzing how best to use them to meet your marketing objectives.

Image used under license from RustyBrick on Flickr.

All marketing/advertising impressions are not equally valuable. A thirty-second TV spot, a quality visit to your website, a walk through your brick and mortar business are significantly more valuable ‘impressions’ than exposure to a print ad, a display ad or a text ad.

A link (paid or organic) on a SERP for a competitive non-brand search is much more likely drive incremental business than traffic from someone searching for “YourTradeMark Coupons” and coming through an affiliate. These touches shouldn’t all be treated the same, and good attribution systems need to ‘understand’ and/or ‘sniff out’ those distinctions.

Let’s take a look at an example from basketball:

The Bulls led the series 3-2. Game 6 came down to the wire.

Before the action in the video started, Pippen in-bounded the ball to Kerr.

Click here to view the embedded video.

So the path to conversion looked like this:

Pippen => Kerr => Pippen => Jordan => Kerr => Conversion

Let’s take a look at how 5 different approaches to attribution would handle that conversion:

1. No Attribution System: The silo view would spread credit for the conversion as follows:
   • Kerr: 100%
   • Jordan: 100%
   • Pippen: 100%

   Grade: F

2. Last Touch Attribution:
   • Kerr: 100%
   • Jordan: 0%
   • Pippen: 0%

   Grade: B-

3. First Touch Attribution:
   • Kerr: 0%
   • Jordan: 0%
   • Pippen: 100%

   Grade: D-

4. Proportional Attribution: Crediting each touch equally we’d split the credit this way
   • Kerr: 40%
   • Jordan: 20%
   • Pippen: 40%

   Grade: D+

5. Assist Tracking Attribution: Crediting the order to the last touch and assists to each preceding touch we’d view this transaction as follows:
   • Kerr: 100% + 1 Assist
   • Jordan: 1 Assist
   • Pippen: 2 Assists

   Grade: C+

Pretty clearly, none of these attribution models provides a very good understanding of that conversion, but any of the attribution models is preferable to none.

‘Assist’ counting can be particularly misleading over time. Consider our basketball metaphor, extended. Pippen always in-bounded the ball. Blind assist counting will lead one to conclude that he was the greatest play-maker in history, averaging 50, 60, maybe 70 assists a game since he often touched the ball on the offensive end of the court as well!

Consider another case. Suppose someone develops the ultimate blanket advertisement. An ad for Acme pops up on every computer, every mobile device, every TV screen in the country on boot-up. Was every conversion on Acme’s site that day impacted by those ads? Would the site have had no conversions absent the ads? Of course not.

What we’re really interested in learning is not what ads consumers were exposed to, but what lift can be credited to those ads.

A better metaphor might be the plus/minus ratio in hockey. When this player was on the ice, did our team perform better or worse and by how much?

Pure A/B split tests for email, display ads, and direct mail provide the cleanest answers to those critical questions.

Unfortunately, pure testing isn’t possible in paid search, natural search and other, less track-able, forms of offline marketing.

It is possible to hack at the incremental value of paid search through testing, but those tests can be challenging to design and execute, and incur material opportunity costs.

Attribution systems help defray the cost of ongoing testing. We believe firmly that periodic A/B testing remains crucial to calibrate coefficients for attribution systems particularly for display advertising, but attribution allows advertisers to reduce the need for ongoing tests significantly.

As we pointed out last year, mathematicians without guidance from marketing experience will build the wrong types of models. Bayesian models tend to over-credit affiliates, email and brand ads because visits to these just before a purchase strongly correlate with conversion success.

As marketers, we recognize that the cause of this correlation relates to the unique manner in which consumers use coupon sites, email offers, and navigational search.

The Need For A Better Statistical Model

Building a smarter statistical model is a heck of a challenge. We want a model that more closely matches our intuition as marketers without unfairly biasing the outcome towards one channel or another.

We want a model that recognizes cannibalistic patterns, recognizes the difference between display impressions and display click-throughs,* and handles behaviors associated with some channels more than others.

As an example of the latter, we see consumers sometimes bang through 5 or 10 affiliate ads in the space of a few minutes looking for the best offer, and would argue that channels demonstrating that type of sequence shouldn’t get 5 or 10 bites out of the apple. This is not simple, but can improve our perceptions of how well our marketing efforts work for us and lead to better resource allocation.

*Note: This is not to suggest that display impressions are meaningless. Far from it. Well-targeted, thoughtful display advertising with in-market data and smart Real Time Bidding can be a terrific means of driving business cost effectively. It is also imperative to understand that the lift in traffic from display advertising is mostly produced by impressions, not direct click-throughs. Those impressions really do have value. That value just needs to be measured carefully.

At the end of the day, attribution’s real value lies in providing better insights to drive by. Managing paid search based on attributed conversions rather than silo-view matters. Paying commissions to CSE vendors and affiliates based on attributed conversions can save a ton on unearned commission.

Teasing insights out of conversion path data can prove useful as well. One client of ours believed that 40% of orders ‘driven’ by affiliates were from new-to-file customers. Studying the data more carefully revealed that most of the new-to-file customers credited to affiliates had come to the site previously through search. Affiliate-only conversions were existing customers 92% of the time!

The interrelationships between marketing activities creates new and exciting challenges for marketers from every channel. The more we understand and embrace those challenges, the more we work together instead of in silos, the more successful our businesses will be going forward.

To do this at the highest levels, it is often necessary to pull in data beyond what is available from the engines and typical conversion tracking systems.

Let’s take a look at six types of data that when fed into your paid search bid management will take your program to another level.

1.  Product Feed Data

Whether or not a product is in stock will have a tremendous impact on the value of traffic to your business. The historical data will be misleading if temporary stock outages and product discontinuations aren’t known to the system.

 
• Identify out-of-stock and dropped products. Note: sometimes it is advantageous to advertise on discontinued products if those products have been replaced by a newer model and the site handles those connections well. That said, the value of that traffic will certainly be different than it was prior to discontinuation.
• Identify new products. Use that same product feed to carefully build out new keywords with well thought out landing pages, and copy as new products and categories become available.
• Recognize inventory shortages. Layer in an understanding of the inventory position of each product. This is not simply a matter of counting size variations: being out of shoes sized 7, 8, 9, and 10 is a significantly bigger problem than being out of sizes 5,6,11 and 12.
  • Other examples: Which hotels are filling up in the likely shopping window? Which airline routes are booking up?

2.  Offline & Cross-Device Conversion Data

People often shop on a website but then call a call center to ask questions, or visit a store for touch, feel or get in-person counseling. Making sure search is credited properly requires going the extra mile.

 
• Phone tracking. Embed unique ad identifiers on the bottom of web pages. Call center employees ask web callers (unique web phone number) to scroll to the bottom of the page and read off a code.
  • Banks of 800 numbers are both more expensive and less granular in their ability to match conversions to ads.
• POS tracking. Unique coupon codes can help catch some of this spillover. Offline to online offers and loyalty programs (“Fill out this survey on our website and get…”) can identify those who had cookies from visits prior to the offline transaction.
• Cross-Device Tracking. Website login data can help tie sessions from different devices together and give a more complete picture of the impact of marketing.

3.  Conversion Value Data

The true value of an order may not be apparent at the time of check-out. Back feeds can be used to:

 
• Knock out frauds and canceled orders.
• Fold in margin data when not available on the site.
• Provide lead valuation data (credit scores, application quality, etc)
• Tie in after-the-fact lead conversion data. Applications that follow leads, accounts created following applications, account funding after creation. All leads are not created equally. Which search ads drive the higher versus the lower value leads?

4.  Customer Type Data

Some types of customers are more valuable than others. Knowing what keywords drive what types of customers can be critically important, but you may not have that data available at checkout.

• New customers versus existing customers. Many firms place more value on attracting new customers than generating sales from existing customers. Also, knowing which existing customers use search ads (particularly competitive non-brand search ads) can set up useful tests of offline marketing treatments; maybe those folks don’t need expensive mail pieces to drive those orders?
• Direct-mail match-back data. Who got the last book?
• Small Business vs Medium Business vs Large Business vs Consumer vs Government buyers. The type of customer effects the lifetime value expectation, and different types of buyers often search differently.

5.  Customer Value Data

Obviously, customers of the same “type” have different values as well.

 
• Do people who buy after a search for “discount/cheap foobar” or “foobar coupon deals” buy again from you at an average rate? More? Less? Would knowing the answer effect how much you’re willing to spend to drive those orders?
• Do higher AOVs associated with some keywords also mean the customers generated are more valuable in the long-term as well?
• Do the click-paths followed, inform customer value? Do customers who venture from search to comparison shopping engines or affiliates exhibit the same lifetime value as others? Do those price-shoppers search differently than others?

6.  Attribution Credit

If search shouldn’t get all the credit for an order, what fraction should it get? Good attribution systems can feed fractional order information to other systems, and good bid management systems can accept fractional order information from other systems.

Even if attribution systems aren’t sophisticated, passing the channel credited to the search platform can provide improved marketing allocation and provide some interesting insights.

“Have you noticed that half of the orders your last-touch attribution system credits to affiliates have come through competitive paid search ads within 30 minutes of the order?”

Many of these feeds can be set up to run in the background and the results fed dynamically into the bid management algorithms. For some, periodic human analysis and target adjustments based on those analyses may be more practical.

In either case, it’s important to recognize that the simplest view of marketing efficiency may be misleading, and correcting for those assumptions in bid management can make you more money.