Three Ways to Track ROMI

In keeping with this week's "three ways" theme, here's a framework to track ROMI (return on marketing investment) that can pretty much solve any problem. It's high-level, but it's been very helpful for me over the past couple of months in thinking about a very complex, multi-channel marketing measurement problem.

First, Test vs. Control. This means for every direct tactic that you launch, hold out a control group. The control group can be small, but should be selected from the exact same list that you used to generate the campaign. Otherwise, bias can (and will) creep in. Hold out the control list in a "stimulus table" somehwhere in your database. When the tactic has been in market for a long enough time, take a look at the performance of test vs. control customers against a baseline. A good baseline to use is "90 days prior". You can also use "vs. last year" but here you have to watch to make sure that there isn't a disproportionate percentage of customers in either list that weren't active in that time period.

Second, Opt for Customer Lifetime Value or Customer Data When Measuring Return. It's always better to take a look at customer dynamics vs. simply total sales. For example, if I'm doing test-control analysis, I'm much more interested in understand how many customers I acquired, lost, what my impact was on average transaction size, and ultimately what my impact was on CLV than just total sales for the population for that period. Disaggregating gross sales performance into customer-specific data yields huge insights. This is particularly true of subscription or repeat-order businesses.

Third, Econometric Forecasting with a Marketing Component. Putting all of your data into a time series database and understanding the contribution of each marketing tactic to return is the final step. This can work in a direct marketing, advertising, or "mixed" environment. The first step is to leave the marketing stimulus out--build a forecasting model for key independent variables, namely total revenue, new customers, lost customers, and average transaction size (and potentially price as well). Explain as much of the variance as possible using autoregressive terms (seasonality, etc.) and extrinsic data (competitive actions, GDP, business confidence, etc.) When you have a good model, add in marketing stimulus, using appropriate adstocks / decay rates. This should yield a good model for understand what elements of marketing are driving what elements of return.

Taken together, these three marketing measurement techniques can enable a "test and learn" culture at your organization.