ROAS in Paid Search Marketing - Marketing & SEO Discussion List

Written by Mark J. Welch
October 22, 2007

Tuning Paid Search Campaigns by Conversion

Nathan asked whether it's proper to design a campaign based only on conversion rates or cost-per-conversion. My answer is "no," to the exact question posed, although I suspect that Nathan probably already applies the same kind of analysis I'm about to share.

First, "conversion rate" and "cost per conversion" are merely proxies for the "real metric," which is "gross profit per marketing dollar spent." Our goal is always to generate sales that produce more than $1 in gross profit for every marketing dollar spent. (Depending on overhead and other factors, the goal might be $1.10 or $2.50 per dollar spent.)

If you rely on any "proxy" as your measure, and if your competitors really measure the "real goal," then over time you're going to lose out because your competitors will poach the profitable customers and you'll end up with the left-overs. For example, you might think it's fine to have a conversion rate of 10% or a cost-per-conversion of $10, but then you'd still be disappointed (and unprofitable) if your average order size gradually dropped from $35 to $10. (And you'd have a mistaken perception about the market, since your strategies have caused you to draw mostly from one segment of the broader market.)

Of course, we can never rely exclusively on "real data," no matter how "technically accurate" and no matter how precisely the data matches our goals. That's because we're dealing with statistics, which can be misleading if analyzed improperly. The most obvious example here would be "long-tail" keywords (those thousands of keywords which aren't in your top-100 volume keywords, but which are occasionally searched).

The Process

In September, I launched a large campaign with about 4,000 ad groups and 25,000 keywords.

Six weeks later:

More than 500 of those ad groups and 10,000 of those keywords have drawn zero impressions or clicks; I'll consider these campaigns "inactive."

Perhaps 2,000 of those ad groups and 10,000 of those keywords have generated "some impressions and a few clicks, but zero sales" -- let's call this the "low-volume dud group."

Another 200 ad groups (perhaps 1,000 keywords) have generated "some impressions and a few clicks, with one or a few sales" -- let's call this the "low-traffic-with-sales group"

And of course 500 of the keywords have "lots of impressions, many clicks, and many sales" -- let's call this the "primary group."

For this campaign, I have an overall conversion rate of 1%, so I generally look for "at least 100 clicks and at least 2 sales" before I even consider putting any keyword into the "primary group."

Within the "primary group," I generally enforce my ROI goals fairly strictly -- if I see ROI below my target for an ad group or individual keyword after 100-200 clicks, I'll reduce bid rates or pause the campaign. (Typically, in this particular campaign, I am seeing about $2 in gross profit per $1 spent, for ROI of +100%. If ROI drops below +25% I'll drop the bid rates.) Within this "primary group," my overall conversion rate is about 1%. Most of my time, energy, and "testing" is done within this "primary group," which accounts for about 85% of total activity (clicks and sales).

Within the "low-volume-with-sales group," most ad groups and keywords have "incredibly high ROI" (often +500% or +1,000%) because the conversion rates may be 10% or higher (many keywords have 1 sale from fewer than 10 clicks). I certainly will maintain those ad groups (and I may consider increasing bids), but I know that generating 1 sale from 5 clicks might just be a "fluke" and I will probably draw zero sales from the next 50 clicks, so any response must be cautious and tentative.

Within the "low-volume dud group," my ROI is -100% (awful), but the data is inconclusive because I haven't reached a statistically significant number of clicks yet for any individual ad group or keyword. I'm not going to cancel an ad group or keyword that has 15 clicks but zero sales, because over time, it might still generate a positive ROI. However, I will look at the aggregate data, and if I discern a pattern (for example, if I have 50 ad groups related to "widgets" and they've drawn a total of 1,000 clicks but zero sales), I might make some decisions about that broad category.

Within the "inactive group," I see no immediate reason to purge these keywords, and I expect that 50% of these groups will see activity in the next month.

If I used a "blind, mechanical" analysis, I'd delete every campaign with negative ROI, which means I'd delete those 2,000 "dud: ad groups with "some clicks but no sales," even though I don't have statistically significant data. And a "blind, mechanical" analysis would suggest that I should increase bids on the 200 "low-traffic-with-sales" ad groups, even though I don't have statistically significant data. That approach would cause me to overspend on "fluke" keywords, and would lead me to "drop" thousands of keywords that might have generated sales next month.

Mark Welch
http://www.markwelch.com