Building a Yield Model, Part 3

This article is part three in an ongoing series on how to build a yield model. The hypothetical model we’re using is for an online store. Now that you know how many people saw your ads (see step one) and how many people responded by visiting the site (see step two), you’re ready for step three: determining how many of those site visitors interacted with the site in a relevant way.

Step three is a critical one. It’s just after someone has arrived at your site but just before she begins to generate revenue. Monitoring this step in the yield model can help you figure out where you’re losing people, understand the impact of different marketing initiatives, and establish a benchmark for site changes over time.

Given the size of today’s sites and the myriad paths a visitor can take, identifying which user action is the right one to track in your yield model is important. Before they generate revenue, site visitors open shopping carts, review product information, read articles, ask questions, register, and more. Choose the action with the highest correlation to producing revenue. You may decide you need two steps here, while I’ve shown only one. Do what works for you.

Let’s look at how this step can be adapted to fit different business models. One ad-supported site may skip step three completely because site visitors are generating revenue as soon as they hit the home page. Another may generate only minimal revenue from the home page and decide to track subscription to an email list as the most important step leading to revenue. One online store may determine that starting a shopping cart is a better indication of revenue potential than registration is, while another finds that registration is better than a shopping cart. A third online store may decide to track both.

A word of advice: It’s more important to complete an entire yield model than it is to perfect a single step. Perfectionists out there will get bogged down trying to document the infinite number of branches a yield model can take. Make your best guess, limiting the amount of side work you do for this determination.

(For those of you just joining this series, we’re assuming you’ve developed a way to store marketing campaign information at the user level, so you can track the effectiveness of your campaigns all the way through to purchase. See the article on step two.)

We’re using an online store as the example for our sample yield model. Let’s assume we’ve determined the most relevant interaction to monitor is registration. Our hypothetical model now shows:

  • 10 million people saw your ad.

  • The ad had a 2 percent (or 200,000 people) response rate.
  • You had a 50 percent (or 100,000 people) registration rate (lucky you!).

Let’s pretend, for comparison purposes, that historically your banner campaigns on two different sites have had identical click-through rates. So, if you purchase an identical number of impressions on each site, you’ll get the same number of visitors to your site from each banner. That covers steps one and two.

When you get to step three, you discover that although the click-through rates were the same, Campaign A had twice as many people click through and go on to register as Campaign B did. Why? I’m sure this calls for a meeting, but for today I’ll throw out a few possibilities. Could it be the demographics? Are the banners pointing users to different entry points? Perhaps the answer is hidden further down in the yield model. For example, are the two groups interested in different types of products? Yield models are very thought provoking.

But why do you need a yield model to monitor registration rate? Because you’ll get a more complete answer by looking at results in their entirety instead of piecemeal. Don’t forget the blind men who examined the elephant. Examining changes in registration rates along with initial response rates can help you identify the impact of site redesigns, changes in the marketing mix, and so on. When you carry the model out further — to capture purchase rates or purchase sizes, for instance — you’ll get even more ideas for how to improve your results.

For example: If the response rate stayed the same from one period to the next but the registration rate went down, what are some potential causes? What if the response rate went down but the registration rate went up? See how this can get you thinking? It gets even better as the next steps are added.

The point is, the yield model highlights the issues and forces everyone who sees it to wonder why the results are what they are and why they change from one period to the next. The questions that arise generate cross-departmental thinking and help people who might otherwise never interact understand how their efforts affect each other.

Don’t limit yourself to comparing yield models for one specific campaign over time. Compare your email campaigns to your snail mail postcards. Compare search engine results to word of mouth. You’ll learn something every time and will begin to quantify the impact of changes to your marketing mix.

In my next column, we finish our hypothetical yield model. That’s when the fun really starts.

Related reading

site search hp