Part 6 in a Series on Web Analytics and Search Engine Marketing Programs
My co-founder Joel Hadary loves to quote "If you don't care where you are going any road will get you there." And it does pretty much sum up why finding and choosing the right optimization goal is the single most important part of doing any Search Analytics. In Part 5 of this series, I went over five different conversion optimization strategies for SEM when you HAVE traditional eCommerce like conversions. And that's the simple case. Because when you don't have easy, mappable conversions (or you don't have enough of them to measure well) things are quite a bit more complex.
It's a mistake to think that the world cleaves into two simple buckets: eCommerce and non-eCommerce sites. The different optimization strategies for eCommerce sites reflect basic differences in type of site (by sales cycle, product mix, margin mix and long-term customer value). For sites that aren't doing online sales, the mix is larger and has an even more dramatic effect on optimization strategy.
Because of this variation, I'm only going to work my way through some common cases and hope that the basic ideas for choosing optimization points come through.
Let's start with a "transitional" case - sites that have conversion but where the numbers are too low to optimize all or most of a SEM programs' variables. In this situation, you're ultimate goal is still an endpoint conversion. And what you're looking for on the site are actions that we call "conversion proxies." The idea behind a conversion proxy is simple. You correlate various actions on the site (downloads, video views, interactions, specific page views, number of visits, time on site, etc.) with conversion. You then use those highly-correlated points as proxies for conversion when optimizing. Keep in mind, you're looking for upstream actions that have significantly higher levels of occurrence than conversions. Otherwise, you aren't solving your problem.
With a conversion proxy, you're less concerned with navigational effects and self-selection than you might be in most correlation analysis. For instance, suppose you have an order process that takes 4 steps and has an 80% drop-out rate. It's quite likely that the first or second step of the order process could function as a conversion proxy. While it's obviously meaningless to say that people who reach step 1 are more likely to order than people who don't, for a proxy that isn't a big issue.
That doesn't mean that choosing a conversion proxy is without risk. You need to make sure your proxy is independent of the SEM programs you're analyzing. For instance, suppose you found that Page X on your site was correlated with conversion. You choose it as a conversion proxy. But, if some of your PPC programs land on that page or on another page that is immediately linked, these programs are going to "win" versus everyone else because they are navigationally close to the proxy.
This problem can even occur with examples like the first one - where the proxy is the first step of the order process. If one landing pages is nothing but a single giant drive to the order process and another drives to different places, then the first is going to perform better when page 1 of the order process is the chosen proxy. There is no 100% guaranteed way to screen off these effects. It's your job to think about the possible biases your proxy might introduce. So when you build a proxy, look very carefully at the first batch of results. If you see some SEM programs with extraordinarily high rates of "conversion" then there's a pretty good chance your proxy is biased.
Lead Generation sites form another, completely distinct tier. In a sense, lead generation sites often feel more like traditional eCommerce sites since the lead often has a specific site endpoint. But there are a few common twists you need to be aware of. One of the biggest mistakes I see with lead generation sites is that they often have multiple lead types on the site but don't weight them differently as conversion goals. This can be disastrous. It's precisely equivalent to an eCommerce site with vast disparities in average cart size choosing to optimize on sales not revenue or margin. But while eCommerce sites rarely make this mistake, lead gen sites do it all the time. If you do have multiple lead types on your site, and you don't weight them, then it's almost inevitable that my basic rule about SEM (bad traffic is cheaper than good traffic) will cause your program to emphasize crappy, low-value leads at the expense of high-value leads. Why? Because your competitors may have taken the trouble to actually do this - and they will bid up the words that send good leads and leave you with apparently less expensive bad leads.
Every bit as common a problem is what to do about phone leads - and it's much less tractable. Most lead gen sites rely heavily on phone lead generation. Typically, that means every page and every call to action has an 800 number. Ideally, that ought to be a set of unique 800 numbers. But even when it is, there's lots of fog in the analysis. If you take lots of phone leads, you have to make some tough decisions. First, you may decide that your online lead gen is a good proxy for your total lead gen. If that's true, then your online leads should be considered by your sales staff to be about as good as phone leads. If that's not the case, you've got a problem. If you don't put phone numbers everywhere on your site (and you probably should), you may be able to identify good conversion proxies based on tracked, unique phone volumes. More likely, you'll have to build some model of engagement on your site and test it as best you can with tools like survey research.
Media and Ad-Based sites form the next tier worth considering. For sites like this, there actually is a clear path to success. It just isn't a success with a traditional single-page endpoint. For most ad-based sites, success is measured by a combination of page views generated and value per impression. There's often a pretty dramatic difference in the impression value for ads placed on different pages. That complicates the optimization model in web analytics - which already does a pretty poor job of letting an analyst use measures like "Total Lifetime Pages" or "Pages in Period X After Event" when doing either optimization or analysis.
We generally use two strategies for ad-based optimization. The first is to optimize outside of real-time by analyzing chunks of data. We typically use campaigns or time-based variables to create a population of visitors who entered from the SEM program during a specific period. Then we analyze their total behavior from that period to date. This allows us to calculate value per visitor and appropriately optimize the SEM program.
If a program needs real-time optimization, however, that approach won't work. Instead, we shoot for developing an "engagement" proxy. It's really the same idea as a conversion proxy. We try to find behaviors in the 1st visit that are reasonably predictive of the extended value. If we can do this (sometimes you just can't), then the "engagement" proxy can be used to optimize the SEM program in real-time.
The last type of site I'm going to consider is one where there is no clear success event at all. Sites that exist primarily for branding, information or relationship building may all fall into this category. For these sites, the most common error by FAR in SEM programs is to optimize to traffic. Optimizing to traffic is always a bad idea (unless, I suppose, you can charge more for impressions on your Landing Page than you are paying). Remember the basic rule about SEM? Bad traffic is cheaper than good traffic.
For sites that exist primarily to brand, you'll probably want to optimize with some form of engagement metric. That's a topic I've written extensively on before - and it's also yet another case where there is no one right engagement metric for every site. What I'd suggest first is trying a range of engagement metrics and seeing what percent of your population falls into each. I'd also look at which Search Terms and campaigns are doing well or poorly by each measure. You probably have a fairly decent intuitive sense of which search terms and campaigns are driving better qualified traffic. If your engagement proxies aren't close to that intuitive sense they're usually bad. In addition, you should look for an engagement metric that qualifies enough visitors to be statistically interesting and doesn't qualify too many visitors from almost every campaign. In general, you'll probably want to pick the most DEMANDING metric that still generates enough data to analyze.
What are some common engagement metrics? Simple metrics like 2+ page views can often improve optimization dramatically compared to "traffic." But such simple metrics are rarely good in-and-of-themselves. 2+ visits is a surprisingly useful optimization metric given its simplicity. Time on site is appropriate for some sites and can be pretty decent optimization metric. Total pages can also be useful. You can use Functional measures of engagement - tracking the percent of people who reach key pages on your site. This is especially valuable if you know that significant percentage of the traffic on your site isn't engaging in ways you care about (maybe it's career traffic for example). Ideally, you'd probably like to use a fairly rich combination of factors (like key pages, interactions, time on site, etc.) to really come up with a good index of Engagement.
That's no easy task in most tools (VS can do it - and check out the new WebTrends Scoring system - very cool btw - and something I'm going to write about shortly). You might well have to short-cut here and there to come as close as possible to a good answer.
How do you know you've done a good job picking an Engagement metric? The truth is, there isn't any behavioral analytic technique for absolutely proving this. Instead, you might want to think about using survey research to gauge interest and then tie those results back to your various engagement models and see which one is most predictive.
What does this all add up to? There are endless permutations of appropriate optimization goals. When you pick the one(s) you are going to use, these are the things to keep in mind:
- Do the optimization points capture all the important aspects of site success?
- Do they weight those important aspects appopriately?
- Are the optimization points biased in ways that favor specific campaigns or campaign elements?
- Can the optimization points be used for real-time optimization?
- Are there simpler proxies that would work for the same optimization points or that would provide real-time optimization?
- Can your tools (both WA and Bid Management) capture the necessary data?
If you've considered these questions, the chances are your optimization points are better than average and you are ready, at last, for some real SEM Analysis!
Other Posts in this Series: Introduction, Searchnomics Issues, Getting Setup for SEM Analysis, SEM Data vs. Web Analytics Data, High-Level Search Engine Reporting, Analyzing Search Traffic in more Detail and Measuring Search Effectiveness for eCommerce Sites.
Comments