In a recent YouTube video, ‘Google AdWords Bidding Tutorial‘, Google’s chief economist Dr. Hal Varian describes a method for calculating what he calls “optimal or near-optimal” bids for keywords. His method requires first determining your Value Per Click and then selecting the bid whose incremental cost-per-click (ICC) is equal to, or less than, that value. To facilitate this process, Google is testing an update to its Bid Simulator feature that includes the ICC for each possible bid (The version of Google’s Bid Simulator you see might or might not include this additional column. If your version does not contain this column, it can be calculated from the difference in Cost between any two adjacent rows, divided by the difference in Clicks between them.):
But clicks actually have multiple different values, not just one, so here I list a few of the factors that internet marketers should take into consideration when bidding in pay-per-click auctions.
The first factor is exactly what Dr. Varian describes; that is, the direct revenue-generating value of a click (or, what I call the ‘base value‘) calculated for each keyword as the ‘Profit Per Conversion’ times the ‘Conversion Rate’, both of which are typically determined from recent performance data. For retailers of tangible goods (and some intangible goods), the profit per conversion is simply the revenue per conversion minus the cost of goods sold (COGS) per conversion. For CPA-targeted accounts, the profit per conversion is often based on the lifetime value of a customer. Many subscription service websites (dating sites, investment newsletters and the like) fall into this category.
A second factor to consider is what I call the ‘influencing value of a site visit‘. Growing attention is being devoted in internet marketing to cross-channel influences, like the tendency of television and radio ads to drive website visits and for website visits to drive off-line conversions via phone calls or in-store purchases. So, even a non-converting site visit has a value insofar as it builds brand awareness, educates the visitor about product offerings, prices, availability and so forth (and, ultimately, insofar as it drives purchases through off-line channels or through subsequent searches). For industries like real estate and auto sales where conversions occur almost exclusively off-line, the base value of a click will essentially be $0 (because the conversion rate is nearly 0%), but the influencing value can be quite high.
A closely related concept, not quite worthy of being enumerated as a separate factor, is the ‘unique value of an impression‘, which for text ads is probably negligibly low, but for the more visually rich ad formats (like banner ads and video ads) is probably a higher value. Especially for established brands, simply showing the company’s name or logo has some value per impression which must be factored into the value per click by means of the click-through rate.
A third factor to consider is the ‘information value of a click‘ resulting from the marginal reduction in uncertainty that additional data brings. If a keyword has gotten 100 conversions from 400 clicks, then it is quite certain that the conversion rate is near 25%, but if it has gotten 0 conversions from just 3 clicks, the per-click conversion rate is not nearly as well known. A coin (which has a per-flip heads rate of 50%) might show no heads in 3 flips, but if the 4th flip brought the first heads, we wouldn’t be confident that the coin had a 25% heads rate (for the same reason that we would not have been confident that its head rate was 0% when we had only seen 3 flips). Similarly, a keyword that has gotten 1 conversion from 4 clicks might actually have an inherent conversion rate of 25% like the 400-click word, but it instead might have an inherent conversion rate of 50% like the coin, or maybe even just 1% and we just happened to see one of those rare conversions by fluke. Our estimate of any uncertain quantity, like conversion rate or profit per conversion, is actually a range of numbers, from the lowest-reasonable guess to the highest-reasonable guess, and not typically a single value. As we get more, recent data about the performance of a keyword, that range narrows until it becomes essentially a single value. Thus, the information value of a click for a low-traffic keyword like the 3- or 4-click word described above is higher than the information value for a word which already has a large amount of recent data, like the 400-click word. Each click thus provides one bit of information about the conversion rate, regardless of the number of conversions that happen, and perhaps some data about the profit per conversion (if one actually does happen) and therefore each click has an information value simply from the reduction in uncertainty it brings.
In a recent blog post, Google revealed that they think that the per-click conversion rate does not change much with the position of an ad. So, rather than try to determine the conversion rate at position 1, the rate at position 2, and so forth, dividing our data up into 10 little bins, if we know that the conversion rate does not change much with position, we can greatly reduce our uncertainty about our conversion rate estimates by pooling all of that data together in just one bin. We can reduce our uncertainty in our estimates of the profit-per-conversion the same way, giving us a much better estimate of the base value per click. (In fact, it is likely that this two-fold reduction in uncertainty – and the resulting increase in the ability of advertisers to derive value from AdWords – was a strong motivating factor behind why Google released this information.)
The increase in the uncertainties associated with our estimates of conversion rates and profit-per-conversion that results from subdividing data is one of the often unspoken dangers of geo-targeting keywords. For example, one account manager geo-targeted keywords related to pet products in North Dakota separately from keywords related to pet products in South Dakota. Of course, by doing this, it became nearly impossible to reliably determine the conversion rates and profit-per-conversion in either state, even though there were no significant observable differences between them.
Fourth, the manager should consider what I call the ‘exclusionary value of a click‘. Getting a click on a given impression excludes a competitor from doing the same, and even getting an impression that does not result in a click still drives one competitor off the SERP. Because available ad space is a Darwinian environment, there is rationally some positive amount that each account manager should be willing to pay per click simply to reduce their competitors’ traffic (and thus, conversion volume) even if that traffic does not have a direct revenue-generating value to that account.
In the figure below we see a typical profile of the net margin per week expected at various bids for a keyword whose click volume and cost-per-click (CPC) depend strongly on position. Using Dr. Varian’s approach, we can determine the profit-maximizing bid for this keyword to be about $6 per click, which would give an expected net margin per week of about $400. To bid less than this amount diminishes the number of clicks received, and thus the net margin. And to bid more brings in traffic at a higher marginal CPC than the profit per click, thus also decreasing the net margin.
Because the profit is maximized at this $6 bid, it is also nearly maximized at bids near this value. So, we can see that, in this example, if we are willing to tolerate a net margin that is about $40 per week less than the maximum, we can bid up to about $7.20 per click. That is, by sacrificing just 10% of our expected profit, we can drive up the cost-per-click for the bidders in the next position higher than ours (who are likely getting a larger absolute number of clicks as well) by about 20%. (There is another bid around $4.50 per click that also generates a net margin that is 90% of the maximum, but since this one reduces our competitors’ CPCs, we’re not as concerned about it here.) In addition to forcing the bidder in the position above us to pay more, since each ad position in sponsored results as a general rule of thumb tends to get about 70% of the traffic of the slot above it, elevating our bid above the profit-maximizing level also causes each bidder in every ad position pushed below us to have their traffic cut by about 30%. Provided that the exclusionary value per click of this sort of financial sadism exceeds its cost to us – which is likely in highly competitive industries – then we can bid at the higher level knowing that our profits are still nearly maximized and, further, that any extent to which we ‘back off’ in our aggressiveness (down to the profit-maximizing bid) should only increase our net margin as a result.
Of course, there are other components to consider, like the vanity value of each click, which is important when account holders insist on appearing in the first position regardless of other considerations. But by taking each of the major different values of a click into account (the direct revenue-generating value, the influencing value, the information value, and the exclusionary value) online marketers can appropriately assess how much they should pay per click for each keyword.
- AdWords Position Preference is Dying. Good Riddance. - April 7, 2011
- Is Google Exploiting Neuromarketing in Reporting Quality Scores? - March 21, 2011
- Does Google Reward High Quality Scores with More Impressions? - February 14, 2011
- Like a Rock: The ‘Bid-CPC’ Relationship - January 19, 2011
- From Business Intelligence to Bathtub Insights - December 30, 2010
- Google’s New “Automated Rules” - December 9, 2010
- Braking the Rules - December 6, 2010
- Google Rich Snippets for Shopping Sites: A New Dilemma - November 4, 2010
- Quality Score Never Shined My Shoes - October 19, 2010
- Ad Auctions are Not Auctions - August 24, 2010