How Modeling Can Help

Modeling is complicated and expensive. It usually requires the appending of external data, such as age, income, and presence of children. It requires expensive software manipulated by skilled analysts. The minimum cost for a model is usually $25,000, and some companies have spent well over $200,000 for a complex model. So why does anyone do modeling? Because, for large projects, it may pay off and give you results that are better than those you could get without the model.

Summit Marketing Group in St. Louis did marketing for a major property and casualty insurance company. The P&C insurance market is very competitive. Hundreds of different companies offer automobile insurance, and customers continually shop for the best deal. The pressure to squeeze profits from every marketing dollar is intense. To help solve the problem, Summit developed an optimization model that was designed to maximize profits for each marketing campaign.

To acquire new automobile insurance customers, the P&C company sent mailings to approximately 1.2 million households in a 10-state area each month. The mailing list was drawn from a database of 9 million households that included response and sales history, demographics, and credit scores. To maximize profits, Summit focused on return on promotion (ROP). ROP is a discounted cashflow technique that is used to measure the net present value of a marketing investment. It is theoretically similar to ROI. ROP considers response probabilities, conversion rates, risk probabilities, acquisition costs, and other marketing decision factors. The goal of the model is to determine which factors are most important in creating the largest return on promotion, and to use these factors in selecting the 1.2 million names to receive the mailing each month. Since the company mailed every month, Summit had the previous months’ history to use as a base. The objective was to use the model to get better and better at selecting the names to be mailed to, in order to continually improve profits.

The modeling process produced some surprising results:

• The best candidates were not necessarily the best responders. The best candidates probably already had insurance somewhere else and were unwilling to shift.

• The best responders were often the least creditworthy, since they have trouble getting insurance.

• Since the model mixes lifetime value, creditworthiness, and probability of response, the top response deciles are not necessarily the most profitable. The goal was profit, not response.

Table 14-1 shows the results of one month’s mailing, comparing the previous month’s selection methods with those used in the current month. The control group is what they were previously doing each month. The optimized group is the new way of doing exactly the same thing with better results.

Table 14-1 : Return on Promotion

	Control group	Optimized group	Percent changed	Number changed
Total mailed	1,264,571	1,264,571	0%	0
Cost of mailing	$547,559	$547,559	0%	0
Number of responses	13,366	16,090	20%	2724
Response rate	1.06%	1.27%	20%	0.22%
Number of sales	1,599	2,323	45%	724
Sales rate	12.0%	14.4%	21%	2.47%
Total revenue	$2,605,603	$3,158,151	21%	$553,208
Revenue per sale	$1,630	$1,360	–17%	($270)
Profit	$95,896	$187,851	96%	$91,955
Return on promotion	18%	34%	96%	16.8%

Notice that the control group had higher revenue per sale ($1630) than the optimized group did ($1360). The optimized group compensated for this by having a higher response rate and a higher sales rate.

This allowed total revenue to increase by over $500,000 and profits to increase by over $90,000 for one month’s mailing. This translates into $6 million in added revenue and $1 million in added profit on an annual basis. If the model cost $200,000 to run, it was still a worthwhile investment for the insurance company.