If you don't know where you are going, any road will take you there. — Lewis Carroll
Although this advice was originally intended for “Alice in Wonderland,” it could just as easily apply to direct marketers who are unsure of where their customer data is leading them.
Fortunately, charting a course for DM campaigns isn't rocket science. But it does require common sense, statistical skills and vigilance in removing uncertainty from the equation. The more you minimize risks, the more quickly you'll arrive at relevant and useful campaign intelligence.
So before moving further down your current campaign path, why not confirm your coordinates? Here's how data analysis can help.
DATA ANALYSIS DO'S
Preparation is essential to ensure the right message reaches the right people at the right time, and through the right channel. The best way to prepare is to plan thoroughly.
Do: Set measurable goals. This seems like an obvious place to start. But surprisingly, marketers often skip this step as they fast-forward to execution. Big mistake. By taking time up front to identify precisely what you want to measure and why, you can focus resources exactly where they'll have the most impact. Your goals should describe the campaign's overall objective, target audience, offer and call to action, as well as creative tone and message.
Do: Define your terms. How will you recognize success when you see it? A predictive model is based on an “objective function” that isolates and optimizes success criteria across a particular data set. So the more specific your goals the better. Do you want to maximize response, revenue or reactivation? Or would you prefer to minimize customers defecting to the competition? Regardless, be sure to select no more than a few criteria of greatest importance. Spreading your interests too thin can dilute results.
>Do: Engage your data pros early and often. Who is closest to your customer data? Your IT staff? A marketing specialist? A service bureau or an external consultant? Regardless, your data experts can help shape your analysis appropriately. What's more, the earlier you bring them into the picture — and the more deeply they get involved in planning — the more efficient the entire analysis process is likely to be.
Do: Build a business case. Campaign analysis isn't free. Before pursuing any course of action, you'd be wise to weigh its costs and benefits to verify the potential rewards. Fortunately, your assessment doesn't need to be perfect — even a back-of-envelope justification is better than none.
Let's say you want to build a campaign revenue model. At what response rate will the model ultimately pay for itself? And how likely are you to achieve that response given the expected size of your target segments and their readiness to buy? Don't forget to estimate the cost of developing the model as well as ongoing execution.
Do: Keep stakeholders in the loop. Give yourself a chance to set expectations by sharing your plans with those who have a need to know. A “no surprises” approach to campaign analysis enlightens others about your data's possibilities and limitations. It also prepares others to support you — no matter what your data might reveal.
DATA ANALYSIS DON'TS
Even with careful campaign planning, you still could be blind-sided by the unexpected. To minimize this possibility, testing should play an integral role in your preparations. But as you test, stay clear of these common pitfalls.
Don't: Confuse a test with a rollout. By definition, testing is a small-scale way to predetermine the relative merits of key campaign components. It intentionally exposes the weakest links to help you discover what works best. But too often, tests also are expected to generate immediate revenue. This is counterproductive.
For instance, it's common to test several variations of an offer to learn which one prompts the best response. At the outset, you know at least one offer will yield less than optimal results. So why force ROI expectations on this scenario? Instead, recognize the large-scale revenue improvement that your nominal testing investment can contribute to the overall campaign.
Don't: Consider too many factors at once. If you try to examine multiple variables simultaneously, you won't be able to interpret the impact of any individual factor. Statisticians call this “confounding.”
Let's say you want to determine the effectiveness of a creative treatment in two cities, Cleveland and Miami. But knowing Miami has many Hispanic residents, you add a Spanish translation. How can you compare responses if you don't know whether differences are due to location, language or both? Confounding is common, but it's easy to avoid if you remember to test only one element at a time.
Don't: Forget to verify tests before large rollouts. Most of the time, when you build a predictive model based on tests, your rollout results will be slightly off. This is natural. But sometimes the difference is significant. The environment may have changed or the model may be flawed. To increase your confidence, add an interim step — test the test with a mini-rollout prior to any full-scale effort. Often enough you'll affirm your solid foundation. But you also might expose serious irregularities and avoid substantial costs.
Don't: Assume too much. Think you can replicate past success without making adjustments? Think again. Even if the model is based on the same product, message or segment, there's no guarantee it can be replicated. Most environments are too dynamic. As you test repeatedly over time, you'll develop a feel for what's prone to work. But gut checks never compare to hard numbers. A predictive model is as unique as the situation driving it. As long as you can make an economic case for testing, trust — but verify. And keep an open mind.
TODD KING (info@database-marketing.com) is the director of analysis services at Database Marketing Solutions, Los Gatos, CA.
| DO… | DON'T… |
|---|---|
| • Set measurable goals. • Define your terms. • Engage your data pros early and often. • Build a business case. • Keep stakeholders in the loop. | • Confuse a test with a rollout. • Consider too many factors at once. • Forget to verify tests before large rollouts. • Assume too much. |
