By Don Holtz, President, Interlocking Analytics

Using Analytics and Measurement to Increase Marketing Performance

This case study describes how a company that has data available and makes a commitment to measure historical marketing performance can accurately use the insights it provides to dramatically increase sales and profits. An example includes an appliance manufacturer that had a customer database that was unused by their mass market brand group. Since only about 20% of end customer purchases were registered, senior management believed that the database was of limited or no value. The original assignment was to develop both a Customer Base Analysiswhich is a deep-dive data mining process that delivers purchase behavior insights and a Marketing Profitability Analysis. In addition to internal data from registration cards, customer service calls, and warranty registrations, external data was appended (300+ demographic and lifestyle data points). An external segmentation was also appended that provided insight as to the core drivers of each segment. We included all of the company’s brands in the analysis, including their mass market and premium brands.

The process of the Customer Base Analysis included:

  • A Knowledge Sharing Session that brought together all the brand marketing staff, IT, and all of their agencies. Each agency was required to present their marketing programs and the measurements that were used to determine the success of their marketing programs.
  • A Data Extraction document was developed with IT that led to the extraction of granular level tables which were provided to Benchmarketing Analytics.
  • A Data Audit was conducted and presented which enabled all stakeholders in the project to certify the data was accurate.
  • An analytical view that combined all transactions for a customer into a single record to be analyzed by our SAS statistical experts was developed.
  • A customer behavioral profile was developed and presented. Below are some of the questions that were answered.
    • How many customer households were in the database?
    • How many appliances did they purchase?
    • What was the timing of their purchases?
    • What was the product order of purchase?
    • What is the value curve?
    • Are customers loyal to a brand or do they split brands?
    • Do customers come back after years from their initial purchase?
    • Are there demographic differences in behaviors and brands?
    • Were customers that received solicitaions more likely to buy? After 1 solicitation?, After multiple solicitations? Does the time gap between solicitations make a difference?
    • Are there geographic differences?
    • Does trade area and retailer mix make a difference?
    • When did customers buy within silos and across silos?
    • What marketing programs would provide a high ROI?

A behavioral segmentation was completed that provided the company with a clear picture of customer groups that behave similarly and the customer characteristics for each group. The segmentation used unsupervised techniques (factor analysis combined with k-means clustering).

An opportunities assessment and economic ROI analysis were performed.

Mass Market Brand Next Steps

The analysis was presented and led the company to value the database. Specific recommendations were made:

The mass market brand SHOULD NOT use the database for any marketing purposes. Their margins did not support a sufficient ROI to directly use the database to communicate directly with customers. The response and conversion rates were projected to be far too low to overcome the low gross margins. However, the information provided in the analysis was important enough that it was presented to one of their larger retailers. This retailer agreed to supply point-of-sale transaction data to determine if it could be used to increase the manufacturer’s and the retailer’s sales.

After receiving the data from the retailer, a shorter version of the Customer Base Analysis was developed. This analysis further enhanced everyone’s understanding of brand loyalty, time gaps in purchasing, etc. In addition, it was discovered that there was a significant number of customers that purchased appliances in a pattern that was more indicative of being a business than being a consumer. A model was developed that was able to identify a segment of customers as a businesses, despite receiving consumer name and address information in the registration.

Once the analysis was completed, a series of “Look Alike” models were developed by appliance type. An experimental design was constructed that would include some customers receiving single solicitations and some receiving multiple solicitations. These were developed by appliance type. As solicitations were deployed and responses (purchases) were measured, response models were developed to supercharge the results and the ROI. A complete series of planning and measurement reports were developed and produced including lift versus control groups and ROI. The reports measured the financial return at a SKU, product class, all appliance, and total store level. The results were very favorable.

Premium Brand Next Steps

The premium brand, which had been using the database, had a revelation. Due to the timing of registrations, the mail to solicit the purchase of a second appliance was actually sent after the purchase was already made. They did not know their solicitations had limited effectiveness until receiving the insight from the Customer Base Analysis. This is the danger of not thoroughly understanding the process of how data is collected or how calculations are performed to measure marketing performance. In order to solicit customers to sell a sufficient volume of incremental appliances and achieve a good marketing ROI, Logistic and Ordinal Logistic regression predictive behavior models were created and an Experimental Design developed. These techniques permitted the deployment of a series of solicitations tests that were followed up by Response Models that significantly increased the effectiveness of their marketing efforts.
Mass Market Brand Results

The retailer became the second highest seller of appliances in the US and the mass market brand’s share of sales at the retailer went from 19% to 32% for the one market category and from 39% to over 50% for a second market category. The brand became the category captain for appliances for the retailer.

Mass Market Brand Follow Up Analyses

A number of initiatives were launched that continue to increase appliance sales and profits for the retailer.

Proprietary Salespeople – An experimental design was created to test the introduction of salespeople that would sell only the manufacturer’s brand at the retailers. A statistical analysis was completed and even though seasonality was a negative factor, it showed that the program was a success.

Planogram and Signage Changes – An experimental design was created for a directional test deployed in 16 stores. A pair of glasses with a small camera on the bridge was given to customers that were entering the store if they were there to buy appliances. The recordings help design a new display set up, color scheme, signage, and hero model process where the goals were to a) have customers find what they want easily and b) increase each appliance sale by a minimum of $50. The overall average gross margin for appliances was 32%. The gross margin above the average selling price was nearly 70% (discovered during the Customer Base Analysis). The experimental design offered enough variety that it could be concluded that a combination of display, color, and hero SKU’s would meet the objectives of increase sales and profits.

Promotional Analysis – The manufacturer spent a large amount of money on circulars on Sundays and Thursdays. A series of econometric models were developed where the time series of the sales of SKU’s, product classes, appliances were disaggregated into their cyclical, seasonal, trend and economic factors components. Other variables in the models included competitive ads, price points, types of appliances being advertised, etc. By removing the effects of non promotion variables, an “irregular” component to the sales volumes would be calculated. If this irregular component exhibited lift that was statistically significantly higher during promotional periods than non-promotional periods, the lift and ROI on the promotions could be calculated.

Inventory Stocking based on Trade Area Demographics – The sales and stocking levels at new stores was analyzed and matched against the demographics of the trade area. The results of the analysis demonstrated that if the merchandisers took trade area demographics with an analytical approach into account when stocking the stores, sales would dramatically increase. A further analysis was performed at the SKU level to develop a series of probability distributions. These distributions included customers likely to come into the store to buy, the appliances they were likely to buy, inventory turns, delivery times, and the likely price points. A Monte Carlo simulation was developed based on the differing classifications of trade areas. The result was a change in the way merchandisers looked at trade area demographics before stocking the stores. A secondary benefit is that reordering came much closer to what the trade area wanted rather than what was initially available for purchase.

Customer Service Customer Classifications based on Potential Life Time Value – A complete strategic plan to first differentiate levels of service to customers and then to use customer service as a marketing channel was developed. The differentiation would be based on Life Time value and “Potential” Life Time Value of the customer calling in for service. Classification models were developed for both. To fund this efforts, an analysis of all of their models was completed to determine when it saved money to exchange an appliance early in the customer service process rather than their current method which was to delay the exchange as long as possible. It was shown that for a number of appliance types and models, it was not only less expensive to exchange the appliance right away, it also led to higher sales and loyalty for high potential life time value customers.

In summary, a process that begins with accurately understanding customers purchasing patterns, characteristics, Life Time Value and marketing program performance combined with correct marketing return on investment calculations can be used to continually increase company sales profitably.

Don Holtz is President of Interlocking Analytics, a provider of 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 dholtz@ultraia.com