By Don Holtz, President, Interlocking Analytics
Modeling Marketing Performance as Conditions Change Over Time
Marketing Mix Models are a proven and reliable way to improve marketing ROI (return on investment) through allocating marketing resources more effectively. In addition to a better allocation of funds, these models help rearrange the timing of ads and promotions to maximize marketing performance. These models are excellent when applied to a stable sales environment such as food consumer package goods where price performance of frequently purchased items is relatively stable even though in hard economic times, there is a shift to lower price goods. When approached properly, these models can also work to assess more complex B2B and B2C marketing with longer sale cycles and customer relationships.
However, what if there are trends in the business environment that would significantly alter how well a marketing program works today compared to past experience? How does a marketer take into account these conditions that can change significantly over time? This is where Time Series Analysis serves as a good alternative. Both Marketing Mix Modeling and Time Series Analysis are powerful analytic tools, each with unique strengths that work best for different conditions.
Marketing Mix Models typically use two or more years of data in order for the resultant recommendations to be statistically valid. Conditions for many companies can change significantly over that period of time and in the future. The results of a Marketing Mix Model provide performance-boosting insights, but further expanding the model to include other data sources can help refine the results to factor in business and economic changes that happen over time.
Including sales drivers that change over time during model development can be a way of adding their effect on sales into the model. External sales drivers can range from economic indicators, such as the Consumer Price Index, to industry-specific drivers such as changes in competitive advertising or market share. However, because of the methodology used to develop Marketing Mix Models, it cannot fully take into account how marketing programs will behave in the future if there are permanent trends that can impact sales in a way not seen in the past. Here are two examples.
For a long time, cable companies were alone in supplying television services. Then a new competitor entered supplying the service via satellites and cable companies started to supply online and telephone services. In the current environment, they are faced with very stiff competition from telecommunications companies that can offer similar services at comparable price ranges. At the time, the simple reality of a strong new entrant into the business environment meant that developing a Marketing Mix Model using data from the prior two years would not capture the true ability of a cable company to increase sales through marketing programs. Inevitably, a portion of their market will be sliced away so add-on sales will be more difficult while sold over a smaller base.
Financial Services Companies:
The economic changes that have taken place over the past nine months have had a dramatic affect on these firms. There are new laws that are changing the way consumers will be able to get credit and how companies will be able to target the sub-prime market. Since promotions in some firms were heavily targeted at the sub-prime market and going forward will not be, this historical data will not longer be relevant to predict improved performance.
Time Series Analysis is the type of marketing analytics that can help marketers maximize the performance of marketing programs in a changing environment. The difference between Marketing Mix Models and Time Series Analysis is that time introduces new drivers of sales. This analysis takes into account how marketing driver effectiveness changes over time due to factors that are not internally controlled.
For example, a financial services company has been promoting its credit card to consumers with an average FICO score of 650. The growth in the number of credit cards primarily came from consumers with FICO score of less than 650. The most successful marketing program has been a balance transfer program with 0% interest on the transfer for 1 year with a 19% interest rate on new balances. A Marketing Mix Model would indeed detect this as a program that brought in the most new customers. The model would not detect a downward trend in new accounts as the foreclosure rate started to rise. Using Time Series Analysis, the company will detect that these programs are not working as well as before and that other programs that have had a lower priority are now producing better results. Time Series Analysis not only detects this change, but also provides the company with an accurate way to project sales as foreclosures continue to rise.
Technically, the methodologies for Time Series Analysis are not very different from Marketing Mix Models using a form of regression. The difference is how the driver and outcome data is organized. In a Marketing Mix Model, it is arranged by attaching the attributes of marketing programs at any given point in time to the resultant sales (also measuring the lag affects, such as TV ads that will impact sales in future weeks). Time Series Analysis attaches the driver data over time to the outcome. The data from past sales drivers and actual sales are used to project trends and other predictive time based behaviors so the change in driver effectiveness can be used in the model. There are some very sophisticated methods such as Vector Auto Regression that can help the marketer understand the evolutionary interplay of all marketing drivers and outcomes to very accurately project future sales.
If products and services are being sold are in a relatively stable environment, Marketing Mix Models tend to perform better. While these models can also work in less stable conditions, Time Series Analysis certainly has advantages. In particular, Time Series Analysis is ideal when the business environment is trending in directions where the marketing environment, economic conditions, competitive structure, or rate of growth is changing constantly. Both methodologies are effective at improving the marketing performance and predicting sales volumes based on scenarios of potential market conditions.
Don Holtz is President of Interlocking Analytics, our strategic partner for business-driven modeling, analysis and data-mining integrated with advanced ROI processes to assess and guide more profitable marketing decisions. Don can be reached at firstname.lastname@example.org.