Managing to Metrics: 3 Pitfalls

I spend a lot of time trying to convince people that of all the actions we can take to improve performance in the keyword auction, bid management is possibly the most routine and non-subjective part of the job. As such, your organization (or you, as a professional) needs to continue to strive to keep that activity as close to rule-based as possible. While you shouldn’t automate the whole job, it’s simply too important to leave it as a haphazard part of a broader job description that might or might not get executed well.

Bid automation to predefined rules is so important, there are at least four well-known ways campaign managers are tackling this today:

  • By doing a semi-manual sort (in AdWords, Excel, AdWords Editor, etc.), using filters, and taking action in response to the biggest problems and opportunities.
  • Setting up a bid rule and executing the rule using AdWords’ “automation” feature.
  • Using a third-party app to do something similar.
  • Using predictive intelligence “black box” technologies to aim for a certain cost per acquisition not only based on bids, but contextual factors that may impact that user session at any given time.

Yes, this stuff’s important. But by executing in pure rote-like fashion without sufficient reflection, we can make significant blunders. So, yes, there is room for “you” in the equation. The above methods – in whatever form suits the situation – aren’t optional. But campaigns get better when you understand how to customize. Here are three pitfalls you can run into, and how you might want to address them:

  1. Don’t forget that ad tests are at least as important as keyword bids. Few tools and commonly-used methodologies excel at the ad-testing game, and yet all keywords are in fact a pairing of a “keyword and the matched ad.” Winning ads should be pursued using rule-based (but human-driven) methodologies, too. If you’ve got nearly-completed ad tests running, the data on the related keywords isn’t necessarily gospel; it’s “work in process.” The future return from that keyword should improve if a winning ad is going to be favored soon.
  2. Don’t let volatile return on ad spend (ROAS) skew your decisions in ad tests. Ad testing is a long march where you need to strike a balance between click-through rate (CTR) and return on investment (ROI) to achieve the elusive “double winner” wherever possible. We can look at cost-per-acquisition (CPA) or raw conversion rate, but the trend now is to factor in revenue, thus ROAS. ROAS is a really important metric, especially as a broad bellwether of how larger clusters (like campaigns or ad groups) are doing. But in more localized tests, you can be fooled by an “impostor ad” that lucks into a couple of large orders. Look closely farther to the left in your ad test stats, and at the conversion rate stats, too. Chances are, if your test is asking something meaningful about different kinds of search intent, then the long-run ROAS winner will be the one that started out of the gate with the high conversion rate. Although there are exceptions, it’s conversion rates on ads that tend to be meaningful; they tend not to be responsible for long-term, significant divergences in average order size. Let those tests run fully before you jump to any conclusions.
  3. Don’t manage keywords out of context. Across tens of thousands of keywords in an account, you’ll have some odd, relatively random winners in any given ad group. Broader match types and keywords without clear buying intent may win out temporarily over keywords with more granular intent. The sad truth is, it’s more convenient to let these big talkers soak up more of the volume and just manage them, while you starve the more granular and possibly better keywords of volume. It’s less work to manage! You need to come in with some a priori judgments on match appropriateness and the value of “buy words,” giving these lesser-volume, but higher-potential keywords a real shot at glory. That’s where simply managing to a rule can lead to suboptimal results. Taking keywords in isolation from their cousins (similar match types and variations), you can wind up managing adequately to a smaller universe of keywords, rather than perfectly to the full extent of available intent. Even the “mighty redwood” will remain a sapling (or die) if there’s too much shade when it’s just getting established.

Related reading

Parthenon on the Acropolis, Athens, Greece
Man chatting with chat bot on smartphone vector illustration