The last post covered the basics of revenue allocation in paid search, and discussed the four methods of allocation supported in ClickEquations; last click, first click, linear, and weighted.
Deciding between even these four allocation models is not simple. None of them perfectly captures the complexity of our interaction with the many different prospects visiting our site. None of them turns our paid search or online marketing efforts into a simple, accurate, clearly instructive number.
This is not true just in terms of the simple matter of how to spread revenue across successively clicked keywords. There are also other complexities in the real world that impact the accuracy and appropriateness of any allocation method.
While attribution is chiefly recognizing the fact that many visitors come to your site multiple times from multiple keywords before converting, it also matters that these visits and the final conversion event also occur over a span of time.
First, you need to define the range of time that will be considered when tracking the string of visits. This is generally referred to as the ‘cookie length’ or ‘cookie duration’ but we call it ‘Conversion Tracking Range’ in ClickEquations.
The default range in most packages is 30 days, although in ClickEquations and many others you can customize this range to be just about any length that is appropriate for your business. It should be set to a length which will cover the vast majority of full purchase cycles that occur in your business.
Since the purchase of paid search software is generally a long and considered purchase among high volume advertisers and agencies, we use a 4 month range for our own ClickEquations.com tracking, for example.
Assuming you’re using the standard 30-day length, if a person visits your site five times over two months before purchasing, only those visits within the 30 days prior to purchase will get any revenue allocation no matter which method you’re using. And since people often purchase many days after their last PPC visit, this can easily exclude the majority of the related visits if your range is incorrectly set to too short of a period.
Most web analytics packages provide ‘days to purchase’ reports. Consult these to see the history of your site and then set your cookie length appropriately.
Revenue Allocation Over Time
If visits happen over time on multiple keywords, allocation is going to spread the revenue to one or more keywords, but on what date should the revenue be recorded?
Should the keywords record the revenue on the date of the conversion event? Or should the keywords get the revenue on the date of the click?
There is a huge difference between these two options. The first results in keywords gaining revenue on days when they may not have been clicked, and avoids any matching of expenses and revenues. This is ‘cash basis’ accounting. Yet this is the most common method of revenue allocation.
The alternative is like ‘accrual basis’ accounting, where we’ll match our expenses with our revenues – if I paid for a click last week and it eventually generates revenue, that revenue is allocated to the day of the expense and I can see the net results for that day.
To look at this in greater detail, let’s revisit our example from the prior post, with a few new details. Suppose each of the three PPC clicks took place on a successive Monday morning, with the purchase on the fourth Monday. Three of these Mondays were in the current calendar month, but the first took place last month.
Assume for now we’ve chosen Linear allocation. And to make this example really clear and simple (something the real world is not) assume that these keywords received no other clicks during this time-frame. Lastly, assume each click cost us $10 – So we spent $40 to get our $100 sale.
In the common ‘cash basis’ reporting used by most web analytics and PPC management tools, a ‘month to date’ report would show that we’ve spent $30 and earned $100 related to this transaction. This reflects the fact that our first $10 click took place last month but the revenue is all allocated to the three keywords on the date of the purchase.
Our keyword report would show us that two of the keywords had costs and revenue this month, but one had revenue but no cost. The cost took place last month and therefore won’t show up in the report. Note that is also is why you’ll often see keywords with revenue and no clicks – the clicks that generated that revenue took place in a prior reporting period.
Of course, this same report (when expanded to all keywords during the timeframe) will show us costs for clicks that won’t produce revenue until some time in the future. This is why the reports don’t all look entirely whacky, but unless your business and marketing has absolutely zero seasonality or time-based variations of any kind is why the average monthly expense/revenue report for PPC – on a keyword or adgroup or any other basis – is of very questionable value.
The alternative, which incidentally is what Google AdWords does (but not Google Analytics nor just about any other package I know about), is to shift revenue back to the date of the relevant click.
The implication of this is that revenue numbers for days long gone can change. If you ran a revenue report for last week on Monday morning, but then someone who clicked a paid ad last week came to your site via a bookmark and purchased, that revenue would be credited back to the keyword last week, and your Monday mornning report is obsolete and inaccurate.
Since AdWords uses a 30 day cookie, any AdWords revenue report covering dates within the last 30 is ‘subject to change’. Adwords only supports last-click allocation so this doesn’t make the kind of mess using this method would for first or linear allocation. But very few people are aware that this is what’s happening.
Each click-chain ends in a conversion right? But what happens if they buy again?
Suppose after our $100 order, the same person returns 3 or 5 days later – before they did another search – and buys again. Should the keyword(s) that received revenue credit for the first sale get allocated money from the second sale?
In ClickEquations and AdWords and most current systems, it does, within the same time range parameters as the cookie. This raises questions and issues in it’s simplest form, and not all the forms are simple….
In The Simplest Terms
As I’ve argued before, paid search marketing has become quite complicated. But it’s still easy to spend money and even easier to aviod the complex reality and accept overly simplified views of how the marketplace you’re participating in really works.
Reality however is sinking in. The broad interst and embrace of Quality Score is a level of detail and sophistication that wouldn’t have happened two years ago. The strong growing interest in revenue allocation is another example.
Moving away from Last Click allocation to some form of Linear or Weighted allocation is, despite all the other options and complexities (including those highlighted in this post), a substantial step.
That is not to say there is really a one-size-fits-all allocation solution. There are many valid reasons why different businesses should choose different allocation models and even use/apply them differently. And we’re just talking about revenue allocation within paid search; the real solution will ultimately have to allocate revenue across all visits of all types.
Until I write a more detailed post on that, I’ll stand by the idea that last-click has gotta go. But as this post tried to point out, it’s complicated, there is more work to do, and marketers should understand all the grains of salt with which they need to read the numbers on their daily and weekly and monthly reports.
One of the new features in the May Release of ClickEquations, is support for multiple revenue allocation models. A revenue allocation model defines how the revenues resulting from your paid search campaigns are allocated to the various keywords that were clicked as part of the purchase process.
The method of allocation is critical because the success or failure of keywords is generally judged on their ROAS (return-on-ad-spend) or ROI (return-on-investment), and what looks like a highly profitable keyword when using one method could turn out to be a very unprofitable one using a different method, or visa-versa.
The paid search world – and online marketing in general – has been thus far largely based around ‘last-click’ allocation. In this method, the keyword that is clicked just before a purchase or conversion gets 100% of the revenue credit for the sale.
To look at a simple example, suppose someone visits your website three times within 30 days before making a $100 purchase.
- Visit #1: Search Query ‘Organic Dog Food’ | Keyword ‘organic dog food’ (Exact Match)
- Visit #2: Search Query ‘Natural Dog Food Coupon’ | Keyword ‘Dog Food Coupon’ (Phrase Match)
- Visit #3: Search Query ‘Dog House Pet Supplies’ | Keyword ‘Doghouse Pet Supply’ (Broad Match)
In AdWords, and most web analytic software, and most paid search reporting tools, the full $100 revenue from the purchase would be credited to the broad-match keyword ‘Doghouse Pet Supply’. The trouble with this, and the reason why multiple revenue attribution models are necessary is that this doesn’t really provide a full or accurate picture of what happened or why.
Each of the three keywords obviously played a role in this customer’s purchase process. We have no way of really knowing which was critical and which was incidental.
- If they hadn’t visited our site during the initial ‘organic’ search would they have chosen our ad when they later did the ‘coupon’ search?
- If they didn’t find us during the ‘coupon’ search would they have recalled our name when they did their final search?
- If we didn’t purchase our brand term as a paid keyword would they have just clicked our organic listing in the same search results?
These and dozens of other questions can never be answered.
The most common alternative revenue allocation method to Last-Click, is called ‘First-Click’. As the name suggests, this method gives 100% of the revenue credit for the ultimate sale to the first keyword a person clicks within the defined conversion time frame. In our example, the full $100 revenue credit would go to the exact-match keyword ‘organic dog food’.
This method is based on the view that the initial visit, the first time the person becomes aware of your site, is the valuable one. It presumes that regardless of any subsequent steps and visits prior to conversion, the initial contact was ultimately responsible.
A more democratic approach is to simply divide the revenue up equally among all the paid search keywords which the user clicked within the target date range before converting. This ‘Linear’ method divides the $100 up giving $33.33 going to each of the three words in our example.
The mathematical simplicity of first, last, or linear allocation makes them easy to understand, but for a variety of reasons many marketers don’t feel they distribute revenue in a way that fully represents the role and impact of the different keywords.
Weighted allocation attempts to correct for this by shifting the revenue across the keywords in a way that more accurately reflects their contribution. Our weighting in ClickEquations is automatic, and based on the past performance of each keyword in the converting chains.
The Right Allocation Model
We’ll take a deeper look at the pros and cons of each allocation model in a future post. There clearly isn’t a universal ‘right answer’ as it depends upon your business, the sales cycle buyers go through, and your own views on what’s important or what you’re trying to encourage.
We do however, share the agree with the growing industry consensus that last-click allocation is the worst choice (despite it being the industry standard). Our opinion is that a move to Linear allocation, while far from perfect, represents a major and simple step in the right direction.
All of the Above
One of the many limitations to better revenue allocation solutions – beyond the lack of options – has been the fact that when choices are offered, it’s been a one-way-or-the-other choice.
Switching allocations models has been, typically, a semi-permanent solution in that all revenue will be processed using the model you choose, and there will be no future way to reverse that change.
This inflexibility is one main reason that even when given the choice many users haven’t yet left last-click allocation behind.
In ClickEquations all allocation models are supported simultaneously. We calculate revenue, profit, and conversions using all four supported models every day. This means you can change allocation models at any time and all revenue numbers in all reports are immediately retroactively updated.
So you can choose ‘Linear’ allocation and spend a few minutes looking at last months’ results, then choose ‘First-Click’ and go back and review those same reports.
An even more powerful reporting option, is that you can access the revenue report for each keyword on all four allocation models together in ClickEquations Analyst, so you can see the impact of the different models side-by-side.
One warning about making allocation model changes: ClickEquations bid-rules run every night and will use the revenue numbers defined by the allocation model set at the time they run. So you can freely and quickly change models in the web application while browsing your reports, but don’t forget to set the model back to your ‘official’ model before the nightly bid calculation run.
Other Allocation Issues
There are many other issues relating to revenue allocation; the time frame considered, the dates of clicks and conversions, and the number and treatment of subsequent conversions, to name a few. We’ll take a look at these and other advanced allocation issues in the next post.
NOTE: Part II of this post is now available.