Now, as ever, profitable pay-per-click (PPC) campaign structures lie waiting somewhere on the other side of opaque customer intentions, challenging competition, and confusing interface quirks. Like writing the great American novel, PPC is hard. The chances of the average PPC manager outperforming tough competition in the auction are about on par with the likelihood that I’ll write a novel as good as The Dead Zone.
In On Writing, Stephen King expresses his belief that a story is like a fossil, buried somewhere waiting to be found. You don’t have to “create” it. If you search hard enough, King believes, you unearth it. Easy for him to say. But that does feel a little like PPC.
Other times PPC feels a bit more like sculpting, in the sense of the methodological rumination attributed to Michelangelo. When asked how he knows what to cut away, he supposedly said, “It’s simple. I just remove everything that doesn’t look like David.”
You can get an awful lot of mileage out of that technique. Surgically removing the parts that won’t achieve your conversion goals is a fantastic approach to simplifying daunting PPC problems.
The “sculpting” method is, arguably, uncomplicated and uncontroversial. If a certain query never, ever (or, hardly ever, if you prefer) converts for you and yet it is being mapped to a productive broad match keyword, it’s really quite simple: you negative out that query.
To be sure, sculpting requires some subtlety; you don’t literally have to cut away parts in all cases. Many times, maximum precision is achieved by, say, bidding 68 percent lower on a parameter.
Let’s take an example of a particularly challenging and complex area of PPC – attempting to make display network ads perform to cost-per-action (CPA) targets – and go through how I recently sculpted my way toward the target. The keywords and line of business have been disguised to protect confidentiality.
Our goal is to generate bookings for “Chicago sightseeing tours.” The CPA target is $100, slightly more relaxed than our target for search. We craft and place into rotation a mix of text and image ads, varying key elements of those ads with a testing purpose.
- Yikes! Our initial CPA performance is frighteningly poor, in the $400 range. Funnily enough, it’s when I see that $400 when I typically get excited. Complete nonperformance leaves you no choice but to stop. But the hope here is: where there’s smoke, there’s fire.
- We’ve been testing a variety of keywords for the contextual advertising (these help Google match ads to pages around the Web, using semantic matching technology), but only some of them appear to result in conversions. We begin to de-emphasize the non-working keywords, pausing the worst ones outright. This is just the first of numerous opportunities to chip away anything that “doesn’t look like David.”
- Powerful math: Every time we weed out the worst parts, we give ourselves more scope to get more aggressive, bid-wise, on the rest of the effort. In the context of a large display network, that gives us more reach. That gives us a chance to discover more quality placements, and to accelerate our learning.
- We discover that conversion behavior varies more than expected by geography. No countries outside of the USA (even nearby countries) are converting, so we all but give up on them.
- We begin to realize that 85 percent of our conversions are coming from greater Chicago! Of course this makes sense, but we didn’t realize the extent of it. We bid up 25 percent (for argument’s sake) on that region.
- We keep the surrounding states in the mix. Two other states, and the rest of Illinois, are performing OK. The entire rest of the USA is probably going to be bid down by a whopping 50 percent the way things are going. Keep in mind our analysis always has to do with purchase behavior. Chicago isn’t automatically better because more people are interested in the ads; it’s only better if they buy.
- We love mobile and click-to-call for search for this client, but we forgo it entirely for display. Just too much junk to worry about. -100 percent on mobile (where it isn’t serving your needs) should be a no-brainer, but I can’t tell you how many times I’ve seen accounts failing because people neglect to address this channel.
- Unfortunately, tablets are the worst channel, but due to Google’s continued refusal to offer advertisers direct bid control over that channel, we continue to endure $700 CPAs on that channel with no means of getting out of the way of this non-converting click shrapnel.
- On average, by day of the week, we ultimately reach a bid adjustment of -30 percent on traffic between midnight and 6 a.m. We continue monitoring all other dayparts under consideration. Wednesday seems to do well. Monday morning typically disappoints.
- We discover that only a handful of websites (publishers or placements) are consistently driving 80 percent of the conversions, with a single publisher (we’ll call it Chicago.com for argument’s sake) generating fully half our conversions! We bid up on the good ones, including well-known properties such as About.com, LonelyPlanet.com, Wikihow, and so forth. It’s not ideal that we’re dependent on a single publisher for half our conversions in the display network. But at least we know what works, and the methodology is leading us straight toward hitting our target. We continue to aggressively exclude or impose negative bid adjustments on many low-quality long-tail publishers.
- Google now offers demographic information for the display network! We try not to prejudge; we watch the real-world performance data. Age 18 to 24 does a bit better than expected, perhaps due to their adept online transacting skills. Overall, performance seems to be fairly even across age groups, though major fluctuations fooled us in the early going on spotty data. Given enough time, we realize that there is a lot of complexity to the causation. For example, retired travelers might put off their trip until September, but they do more searching (and non-buying) in August. On various demographic fronts, including gender and parental status, we’re able to weed out some poor performance with modest bid adjustments. It all adds up!
- We’re not done yet! Remember the ad creative? We’re now starting to see some fascinating feedback on which ad formats and creative versions are working best. (The winning image creative so far is something like “30ish couple under an umbrella looking happy.”) For image ads, attribution can be particularly tough. It’s hard to weight the influence of multiple impressions of an ad, especially when an early impression isn’t properly credited. Still, with enough data, we can learn plenty about the impact of our ad creative. A point worth mentioning here is that the major PPC ad platforms don’t allow you to weight your bid on various ad creative versions in rotation; nor do they allow you to control the proportion of delivery. Such control would be preferred by advanced sculptors.
High five! We were able to hit our target of $100, exactly as hoped. For those who believe in share of voice and the like, it’s possible that our various efforts in search, display, organic, and social are adding up to a whole that is greater than the measurable sum of the parts.
Did we get the volume we wanted from this initial display network launch? Not yet, but it’s not bad. In the past month, the effort has driven a 4.2 percent increase in conversions, not counting unattributed conversions. To put it another way, that’s now going to be viewed as an unwelcome drop in conversions if you shut that part of the account off. These are the types of small wins you need to eke out these days in mature accounts in a competitive marketplace. Worth the effort? Only if six or seven figures in new annual revenue matter to the business.
This method of building a response asset takes a lot of persistence and precision. “Spray and pray” is becoming more outdated with every passing day.
Image via Shutterstock.