Bid Management Automation: Fraught with Questionable Assumptions

Today, there’s no viable “con” position in the “paid search bid management” debate. Either you sing the praises of automated bid management, or you’ll be asked to sit in a dark corner (or cave) and contemplate what possessed you to enter the digital marketing arena. Unfortunately, there are few real debates on the advantages and disadvantages of different approaches to bid management automation. This is mainly because most of the dialogue — be it on a panel, in online content, or in casual conversation — is driven by bid management software vendors. So, in this article, I’ll try to open a dialogue to address this often one-sided debate. I’ll begin by highlighting some of the (often incorrect) fundamental assumptions underlying current bid management tools. In future columns I’ll address additional aspects of the debate.

Large or complex campaign – that is the question

In the bid management space, there are 50 or so bid management vendors, but only a handful of good ones. For marketers trying to make a rational vendor selection, this is a problem in itself, similar to the confusing task of shopping for a web analytics solution in a crowded marketplace.

It’s no secret that paid search is complex. But how you should respond to that complexity?

Counterintuitively, as uncertainty rises, your may need to simplify your actions to a broad steering function. On the surface, weather patterns are incredibly complex, if by complex you mean taking every molecule and wind gust into account, and trying to predict precisely where your helium balloon will land if you release it in downtown Pittsburgh. But if your task is actually, say, setting a cruise ship’s course for Belize and adjusting to stay on course, the mechanism for reaching your goal is certainly within the realm of low tech. Current know-how and available technology are adequate to handle such nautical challenges.

One large vendor in the space uses a catchy Flash animation to portray complexity spiraling out of control. The animation starts with a growing keyword list. But come now, that isn’t complex. It’s just large.

Then the large list is paired visually with many ads and bids. That does seem to be getting more complex. Actually, if you add in a bunch of the other pertinent variables to that mix, suddenly the degree of complexity has gone from being one that is appropriately handled by the vendor’s technology, to one that the vendor’s technology isn’t anywhere close to being able to address. Whoops. Maybe they should have stuck with “large.”

Most bid management tools are helpful in dealing with “large.” They’re often pitiful in dealing with the true complexity inherent in large accounts. The number of possible outcomes, and reasons for those outcomes, is nearly immeasurable. Paradoxically, that’s why something in the realm of “low tech,” combined with deep human marketing expertise, is often best suited to this particular task.

When it comes to the advanced machine learning systems that some vendors claim to be developing, it’s dubious that a mid-six-figure investment in complex system development, data gathering, and mathematical modeling can address the behavioral complexities of search. That lets out 95% of the vendors in the space. And don’t forget Google, the multibillion-dollar company that sells you the clicks. Unlike third-party vendors, they rule the planet’s search data, and the data in your account as it relates to the planet’s data.

Assumptions behind third-party bid management systems: primer

Let’s address some fundamental problems with the assumptions behind most current third-party bid management systems. They all come back to the problem of the “allowable CPA target.” Allowable CPA targets don’t lead easily to pinpoint bidding strategies. In a small minority of paid search accounts (10% or fewer), the cost-per-order or cost-per-acquisition (CPA) target is a useful guide to action and can be measured accurately to guide algorithmic bidding decisions. In other cases, a host of complications conspire to make mincemeat of the concept of an allowable CPA.

1. Attribution. Brand keywords, product codes, and highly specific searches might typically be the last click prior to a completed sale; as such, they typically get “credit” for the sale. But the demand creation load shouldered by earlier “research” keyword searches, to say nothing of display ads, media awareness, and offline advertising, need to be given due credit.
2. Variability and unpredictability of order size. Even more complicated, variable propensity to repurchase and lifetime values of different customers in different contexts. Some sales and customers are worth a lot, some are worth less, and again, it’s difficult to accurately attribute this to a given keyword, especially if the variance is wide. If attribution is inaccurate, then automated bidding guesses wrong, sometimes wildly wrong. Meanwhile, it has lulled you into a false sense of precision. This is reminiscent of Nassim Taleb’s recent comment that “stock market analysts are worse than useless.”
3. Purposes of campaigns. These are often multifaceted and may shift based on internal corporate politics. Direct marketing objectives are only one piece of the puzzle. Many high-spending advertisers are also interested in a higher-order theory of (say) saturation, repetition and recall that works best above a certain volume threshold. Working together to build credibility, a combination of outdoor display, radio, and several forms of online advertising, all targeted to a specific geographic region, might help a company hit an overall sales target in a given quarter. Seasoned marketing strategists understand that the massive complexity of consumer attitudes and behavior exceeds the cognitive powers of algorithms designed for situations of moderate complexity. So they proceed with looser (but still real) gauges of campaign effectiveness as their guide, and quarter-to-quarter financials as a confirmation of effectiveness.
4. Offline and latent conversions. These measurements are a small problem for some digital advertisers, and a massive one for others. An “allowable CPA” sounds good in a meeting, but cobbling the latent online and offline conversion data together and accurately attributing it to keywords isn’t likely to sync up too well with the methodologies of automated bid tools that rely on easy-to-measure online conversion events.

Conclusion: the “correct” bid on a given keyword at a given time is not a precise number, but rather, falls into a range. That range can be much wider than bid management tool vendors imply. Bid changes in such a scenario needn’t always be frequent, especially in a stable auction where ad positions don’t fluctuate all that much.

We’re just getting warmed up. Next month, I’ll address several additional inconvenient truths for bid management automation. Down the road I’ll provide guidelines for selecting a bid management vendor.