Tough Love: Optimizing An Online Campaign

If nothing else, an online advertising campaign produces a pile of data. There’s no need to be intimidated or cast it aside. A little effort in the right direction transforms those stacks of campaign reports into useful information: Find out sooner which banners to keep and which banners to kill. Determine whether buying more impressions on your favorite site is a good idea. Learn how to optimize your media plan by making tough choices between similar sites.

How Many Is Enough?

At the risk of neglecting statistical measures, anecdotal experience would suggest that click-through rates stabilize after roughly 10,000 impressions on any give site for any given creative. At least that’s been my observation after placing a few hundred million impressions across several dozen different sites and networks. (Whether CTR is the best measure of banner performance is another discussion entirely .)

Setting the CTR argument aside, assume that click-throughs are a reasonable proxy for measuring the relative efficacy of various banners. (And within the confines of a limited number of similar creatives on similar sites, experience suggests CTR is indeed an acceptable proxy).

If your test matrix consists of only testing banner A on site X and banner B on site Y, it’s obviously not an apples-to-apples measure. However, testing banners A, B and C on sites X, Y and Z should allow you to determine the “better” creative. And in some situations the better creative varies by site.

Not Convinced? Test It Yourself

Whenever you start a new campaign, simply ask for performance reports measured every 500 or 1,000 impressions (good luck getting them!) for the first 10,000 or 20,000 impressions. Or go online and get them yourself at even tighter intervals.

You’ll likely notice erratic click-through rates in the early going. Make note of the average CTR achieved at 10,000 impressions. Measure again at 100,000 or 500,000 or whatever impression level you’ve purchased (assuming the site is large enough to support a buy of this size).

In most cases, you can expect the final click-through average to be within 10 percent of the CTR observed at the 10,000-impression level. If not, look back through your records and find the impression count that corresponds with the ending average CTR.

Sometimes visually graphing the information helps clarify the exercise. Plot CTR on the y-axis. On the x-axis, plot cumulative impressions served, ranging from the start of the campaign until the finish. At some point, the graph usually goes from wacky to basically flat. Where this happens, you’ve found the impression count at which you reach stability.

A Final Caveat to Banner Testing

If your chart never really flattens out — but instead tails off after 10,000 impressions — your media buy may have reached over-exposure before it had a chance to stabilize. While burnout is more common on small sites, or when buying a specific content area, large sites and large banner buys are not immune.

Another common culprit is web sites with a high visit frequency and visit depth.

How Much Is Too Much?

While limiting yourself to 10 percent of available inventory is a good rule of thumb, the number can vary by site. Again, a graphing exercise helps visualize the extent to which a given creative or a given site has suffered burnout. Instead of tracking impressions on the x-axis, track the cumulative percentage of unique users exposed to your banner.

Not only is this a mouth full, it is also a difficult data set to develop — often requiring the use of estimates. To calculate, divide the daily count of your impressions served by the daily count of unique users that saw your banner. Next, divide this figure by the total number of unique monthly users for the site or content area in question. Repeat the calculation for each day in the measurement period.

While this helps you evaluate the past, perhaps you’d prefer to determine what impression levels to buy in the future. Consider the three sites below:


Unique Visitors

Avg Days Viewed

Avg Page Views

Stock Site A

9.3 MM




Stock Site B

2.9 MM




Stock Site C

1.9 MM






The 10 percent rule would unfortunately lead to purchasing 930,000 impressions at Stock Site A, which turns out to be an average of seven impressions per unique user. All things equal, this campaign would likely experience burnout more quickly than buying 10 percent of impressions at Stock Site B — an average of just over one impression per unique user.

Expect the CTR on the 930,000-impression buy to begin to fall after roughly four days or 133,000 impressions. By the ninth day or 280,000 impressions, your campaign will likely experience further CTR erosion. (Based on the average days viewed of 3.3 for Stock Site A, the average unique user visited the site on 3.3 different calendar days during the month, or once every nine days.) The result is a framework for evaluating media buys and determining appropriate impression levels.

What If I Buy Multiple Sites?

Getting unduplicated reach is tough. Subscribers to MediaMetrix and @Plan are in luck the rest of us are relegated to begging the reports from our online advertising vendors.

Once in hand, look at the unduplicated reach for the basket of sites you are evaluating for any given campaign. The extent of the overlap and the disparity in CPMs between the respective sites goes a long way towards clarifying the site mix. The optimal outcome achieves the largest reach at the lowest effective CPM.

Clearly this type of analysis pushes the limits of the media planning tools available on the Internet, and in many cases is nearly impossible. Not all the sites you want are ranked. Some sites are so large that the results are meaningless because you’re only interested in a specific content area. The list goes on .

Nonetheless, my experience has shown it to be a useful final layer of media planning analysis. Again, this is no exact science, but it provides a mechanism for comparing a variety of sites in similar content categories.

Still have questions? Drop me an email or check out, my small but growing Internet marketing soap box.

Related reading

Overhead view of a row of four business people interviewing a young male applicant.