New Techniques for Marketing Measurement
It’s fitting that the most famous quote about marketing measurement – “Half the money I spend on advertising is wasted. The trouble is, I don’t know which half.” - is itself unreliable: although it’s typically attributed to John Wanamaker, there is no definitive citation. Nor has much changed since Wanamaker’s time: to paraphrase what Mark Twain didn’t say about the weather, everybody still talks about marketing measurement but few do anything about it.
The problem isn't lack of data. Marketers today can track the messages received and actions taken by their customers in more detail than Wanamaker could imagine. The problem is fundamental: Even if we know every action a customer takes, we can't calculate how their actions are affected by our marketing. All we see is what a particular set of inputs will produce in aggregate. It's like following a recipe in a cookbook: You know what the combination of ingredients will produce, but not how each one contributes.
Measuring the impact of individual marketing programs requires breaking apart the sum of the inputs and testing the effect of changing them individually. This is why even complete data cannot solve the measurement problem. Only tests allow marketers to measure the incremental impact of specific activities.
There's nothing new about this insight. But, with the exception of direct marketing, testing has traditionally been expensive and inexact. As a result, marketers limit testing to a handful of questions and look for additional insights in statistical correlations, such as marketing mix models. A limit of this approach is that mix models are effective only in industries with very large customer bases and marketing budgets, and can only measure broad effects.
In today's data-rich environment, testing is much easier. Marketers can build customer databases more often and track many of the messages those customers see and respond to. Even formerly anonymous media, such as broadcast and print advertising, drive consumers to trackable behaviors on the Web.
A different kind of obstacle to testing is the time lag between marketing outreach and a customer's eventual purchase. Although revenue is the ultimate measure of success, marketing that occurs early in a lengthy buying cycle introduces delays in measurement and allows time for intervening events that could affect the final outcome. For both these reasons, marketers need a measure that can be read quickly.
Current innovation in marketing measurement is directed at finding such measures. The obvious approach is to divide the buying process into stages and then measure the impact of marketing from one stage to the next. This is similar to the sales funnel or pipeline monitored by sales automation systems, except it begins earlier in the purchase process. It's also similar to psychological models such as AIDA (awareness, interest, desire, action), which are often used to guide brand advertising.
What's changed from the old psychological models is that today's sources of data make it possible to track the status of individual consumers. In the past, companies rarely had more than basic demographic information about non-customers, which did not help marketers track movement through purchase stages. Tracking customer movement requires fresh behavioral information that might include survey responses or website visits. Behavior is a much richer source because there's more of it, and it's not limited by what prospects choose to provide.
In sum, marketers are finding they can greatly improve marketing measurement by using behavior to track consumers through stages in the purchase process and testing to correlate changes in this movement with changes in marketing programs. This creates several technical requirements, including:
Identifying individuals consistently over time and across channels in order to build a database of behaviors and marketing contacts. This database provides the input for tracking movement through the purchase stages.
Data mining and analytical tools that uncover patterns, such as the number of Web visits within a time frame that indicates a prospect's location in the purchase process.
Scoring systems that apply behavioral patterns and external information to identify the current stage of thousands or millions of prospects on a continuous basis as they move from one stage to another. (In practice, movement does not always follow a fixed sequence: some prospects will stagnate, move backwards or skip ahead several stages at once.)
Tagging content by stage to create the scoring system. While the volume of behavior is important, the specific information consumed gives much more insight into a prospect's state of mind. Three visits to the white paper library means they're just doing some research, while three visits to the contract terms means they're seriously considering a purchase. Although content tagging can be a major task, tags are already needed to help select the right message for each individual, meaning the measurement system should be able to use tags already in place.
- Maintaining a history of previous status, since this cannot be reconstructed from current information alone. This is a classic "slowly changing dimension" in data warehouse terms.
- Converting this data into actual measurements will inevitably lead to more:
- Testing features, including random selection, tagging of test group members and test/control reports.
- Stage movement measures, such as average time per stage and continuation rates from one stage to the next.
- Correlation of marketing contacts with prospect stages. This is used to calculate the marketing cost per stage and show the impact of different contacts on stage progression.
- Integration of purchase history with prospect profiles, needed for revenue-related calculations, such as ROI.
- Projections of future revenues from the current prospect pool. This is a measurement of the number of prospects in each stage, the expected continuation rates from one stage to the next and the expected value of ultimate purchases. It is used to forecast business results and estimate the financial impact of differences revealed by tests.
Few of these capabilities are wholly new to marketing systems and most are used for other purposes, such as offer selection and segmentation. Exposing these tools for measurement requires using them in different ways, so marketers and IT staff need to ensure they are available.
The benefits of stage-based measurement make it worth the trouble. Breaking the buying process into stages lets marketers understand what's driving results, identify problem areas and find opportunities for improvement. It lets them measure the true incremental value of individual marketing contacts, replacing arbitrary "revenue attribution" to the first or last touch or simplistic fractional weighting. Perhaps most important, stage-based measurement provides near-immediate feedback, allowing marketers to quickly reallocate resources to the most productive programs.
Stage-based measurement isn't a complete solution to marketing measurement problems. It has a short-term, incremental focus that may not capture the deeper value of branding programs. It measures each marketing program in isolation, making it difficult to assess interactions among several programs. And it often relies on the assumption that changes at one stage in the buying process ripple through to the end more or less undiminished. This last assumption is particularly dangerous, because the opposite is often true: positive changes at one stage often have negative consequences later on. (For example, a free introductory offer may attract more orders but fewer conversions to paid customers.) Sophisticated marketers will be aware of these issues and compensate for them. They are a reasonable price to pay for finally knowing which half of your marketing budget is wasted.