Predictive Modeling and Display Ads

Often times we’re approached by direct marketing clients who want to scale up their online sales, but have exhausted or saturated their successful channels. They’re bidding on all the keywords that perform, they hit their e-mail list as much as they can, and they’re applying all the focus in their power to SEO (define). The one channel they haven’t successfully cracked open is display advertising.

Unlike search, where you’re limited to the volume of consumer queries around your performing terms, display ads, particularly banners, have a lot more scalability if done right. The amount of inventory and opportunities in some cases (if your target audience is broad enough) is unlimited. The challenge is to crack the formula of what sites, placements, rates, creative (offers, images, messages, etc.), and conversion environments (landing pages, conversion paths, etc.) will ultimately form a scalable and reliable channel that grows your business.

The problem is that many companies have tried to use display advertising with sporadic bursts of banners, often in high CPM (define) and premium placements, without a foresight of predictive modeling and a sustained effort that applies an ROI (define) optimization process and methodology. They then fail and condemn the whole category of online display.

Predictive Modeling

If they just applied some simple media math before starting their planning process, they would have cut out a lot of discovery cost right away. Seems obvious right? Well, guess what — I see tons of companies fail because they didn’t start with this simple step.

Often, I look at the failed campaigns of clients and think, I could have told you this media plan wouldn’t work before the first impression was even purchased. The math just doesn’t work — no common sense was applied. You don’t need to be a genius to know that $15 CPMs aren’t going to work if your average sale is $160, your profit margin is 50 percent, a banner’s average click rate is 0.15 percent (that’s 1.5 tenths of a percent) and your site’s historical average conversion rate from advertising is 3 percent. In this scenario, you’re pretty much looking for a cost per sale of $80 or less just to break even, and $15 CPM won’t get you there on day one. (Of course, there’s a huge value to a new customer — but let’s assume we need to at least break on the first sale.)

Also, when doing a predictive model, don’t assume the best and don’t assume the worst. Don’t be over confident and assume a 10 percent conversion rate, and don’t stack the deck against yourself with a 1 percent conversion rate. Start from the middle and, here’s the big thing, start your channel development by entering races you can win. Then, optimize your click and conversion rates up from there, and start buying higher price CPMs and CPCs (define).

Predictive Modeling Example: Setting Your Media Rate Thresholds

So, using the assumption above, let’s create a predictive modeling table and identify the most we could pay for a CPM or click, based on historical data and assumptions — in other words, set our media rate thresholds to achieve a cost per sale for around $80 or below. By just doing this, we’ll cut out testing and heartbreak for placements that, barring some anomaly or miracle, won’t perform in our first phase. By using these thresholds as buying parameters, we’ll of course save tons of cash in the optimization process.

Below, you can see a predictive modeling table in chronological order of key performance indicator (KPI) metrics, from budget to cost per sale. The green zone represents CPMs and CPCs we can pay in this first round to come in below our $80 target. Pay any more than the highest rate in the green zone ($4), and it’s unlikely we’ll be successful. So you or the client may want the home page of CNET, but you won’t be successful if you can’t get it for a $4 CPM or less. (Notice the green zone goes up to a cost per sale of $88.89 — well, no one said we couldn’t push the envelope a little and hopefully lower CPMs — and CPC deals will pull the average down.)

Note: “Cost Per Sale” is “Budget” divided by number of “Sales,” e.g., conversions.

After You’ve Achieved Success in the Green Zone, Scale Up!

Now, once things work in the green zone, the objective is to raise our banner click rates and site conversion rates. When we do this, we’ll be able to buy higher priced CPMs and CPCs and thus, our channel becomes wider and more scalable, we move into premium positions, and our client’s business grows.

After you perform the above process and have a viable display ad channel based on clicks and easily measurable ROI, you can start to do longer term analysis and ROI modeling on your display campaign, by:

  1. Looking at overall lift in sales and performance of other channels from display, using view based cookies and general trend analysis.
  2. Applying a branding value to a portion of your display media budget to either bring up your CPM and CPC threshold (the maximum amount your can pay for media) to open up new opportunities or bring down the direct marketing cost of your campaign.

However, first we have the hard task of proving the display ad channel, and the first step is realistic predictive modeling combined with a client who is willing to invest the time and money in cracking the success code for their online display ad channel.

Related reading