Computational Profit Modeling, Inc. (CPM) has created a revolutionary new state-of-the-art method of dramatically boosting the bottom-line for Catalogers. CPM Enterprise Predictive Analytics successfully replaced the existing house profit model of one of the largest Catalogers in the nation. CPM Enterprise Predictive Analytics increased their bottom-line by $4,000,000 per year above and beyond their existing house model. Our proven solution can do the same for you! Fortune 500 companies have been secretly using our software for years – now it’s your turn.
According to Catalog Age, response rate optimization is a key issue. The majority of respondents in a recent Benchmark Survey admitted they did not model their in-house lists to improve response rates or average order size. CPM Enterprise Predictive Analytics simultaneously optimizes BOTH response rate AND average order size for every customer on your House List.
Studies have shown that 65% of catalog profits are attributable to "Who" gets the offer, 25% of profits are based on "How much" the offer costs, and the remaining 10% of profits are attributable to "What" creative artwork and wordsmithing are contained within the catalogs. Ironically, most catalog companies spend most of their time and money on the least profitable area rather than concentrating on "who" gets the offer, i.e., the most profitable area. That’s what profit modeling is all about – accurately predicting “who” to optimally send your next catalogs to. This is CPM’s core competency.
Graph of CPM Enterprise Predictive Analytics Predicted Net Profit vs. Decile for various models:
• Increase revenues by accurately predicting your best customers in the top 1 to 3 deciles – and mailing them more frequently
• Decrease costs of printing, paper, and postage by accurately predicting your worst customers – and mailing them less frequently, or not at all
• Maximize profits exceeding your company’s goals with CPM Enterprise Predictive Analytics including the mailing strategy analyzer
• Minimize attrition of your House List and stir up activity while gaining updated customer data with a single full list mailing
• Convenience – No per-run charge. · Run it on your premises (you are no longer at the mercy of an outside analysis house). · Run it whenever you like (days, nights, weekends, etc.). · Run it prior to every mailing to use the most current changes in your customer database if you like. · Most profitable model available (highest accuracy obtainable).
Most real-world modeling problems are not one dimensional or linear. In fact most real-world problems are multi-dimensional and non-linear with complicated interactions between the variables (columns in a table). If you are using multi-linear regression, logistic regression, or decision trees you are not capturing all of the knowledge embedded in your data. You need the powerful CPM Enterprise Predictive Analytics to squeeze all of the knowledge out of your data.
Surprisingly, if you are using all of the variables (columns) in your model, your model will not perform as well as using an optimum subset of the variables due to all of the noise picked up by superfluous variables. Determining the optimal subset is a very difficult mathematical problem known as combinatorial optimization. CPM Enterprise Predictive Analytics excels at combinatorial optimization. Another difficulty a lot of other algorithms have is when there are very many variables and some of them are very similar or even identical but have different names (surrogate variables). This is known as co-linearity. Another way of thinking about this phenomenon is that the incremental informational content of the additional variable only adds marginal value, yet is adding a lot more noise. The NGM doesn’t have a problem with co-linearity at all.
CPM Enterprise Predictive Analytics consists of over ten programs that are designed to be daisy-chained together, including two state-of-the-art predictive modeling tools, NeuroGenetic Modeling (NGM), and Genetic Binary Decision Trees (GBDT).
The NGM is composed of a hybrid of two forms of Artificial Intelligence: Artificial Neural Networks (ANN) and Genetic Algorithms (GA). In very basic terms the ANN’s think like a human brain to solve a very specific business problem. But unlike most other analytical modeling tools we don’t stop there. We literally employ a population of thousands of ANN’s to solve the problem. And we don’t stop there either. Using concepts of genetics and survival of the fittest, based on Darwinian evolution, we allow the population of “brains” to breed over many generations until they come up with a solution that is optimal or near-optimal. The result is a highly non-linear predictive model that has selected the optimal (or near optimal) subset of variables.
The user can select from a menu of different types of problems to optimize: maximization (response rates, profit, etc), minimization (cost, etc.), failure-prediction (capturing 30% or more of the failures at a predefined 10% passer level, etc.), classification models (A-B testing of good vs. bad, or changed vs. unchanged, etc.). Some of the manufacturing customers have used this type of modeling to see which of possibly many production variables have changed that are causing this month’s products to have lower quality than previous months products.
The GBDT is also composed of two algorithms: GA’s and binary decision trees. Unlike all other types of decision trees that use “greedy” algorithms, the GBDT is classified as a “non-greedy” algorithm. What this means is that the other types of decision trees will start at the top of the tree and as they work their way down the nodes they will “lock in” the best solution they see so far and then move down the tree recursively. This results in decision trees that are not globally optimized. On the other hand, the GBDT does not use recursion and holistically optimizes every node simultaneously using GA technology resulting in a better decision tree, which means more knowledge is extracted and more profits can be made. The GBDT solves the same kinds of problems the NGM does, but does it completely differently. Unlike other predictive modeling tools that only analyze a single decision tree, the GBDT plants a forest of thousands of decision trees, and then using Darwinian evolution breeds successively more fit trees over many generations resulting in a highly optimized yet robust decision tree.
The NGM always outperforms the GBDT, but typically by only several percent. And both outperform all other types of models. The NGM model is complex and is known as a black-box model. On the other hand, the GBDT models are so simple even a non-analyst can understand them!
The two methods are complimentary and can even be combined if the dataset is very large and the data is stationary. The GBDT can be used to break the problem down into a few large pieces and then the NGM can be successively run on each of those large pieces, resulting in a 1-2 knock-out punch! However this is only recommended for very large problems to avoid what is known as over-fitting.
CPM Enterprise Predictive Analytics is designed to work with 3rd party automation software (Automate) allowing all of the tools within the suite to run from start to finish unattended. Some programming would be required.
The NGM and GBDT are designed to handle industrial strength predictive modeling problems with up to 4,000 variables and over a million records. Both of these algorithms are computationally intensive due to the populations they are internally working on, so speed is always a factor. Fortunately they run on 64-bit computers and access all of the CPU cores on all of the CPU’s for the highest computational speed possible on a single computer. And if you need even more speed our cluster version runs on multiple computers “tied together” using distributed parallel processing to work together as one larger computer. Working in parallel each computer in the cluster simultaneously solves a different piece of the analysis and the results are internally combined.
CPM Enterprise Predictive Analytics, with its advanced algorithms, uses millions of combinations of variables to create the optimal customized profit model. On the other hand, most Catalogers use a comparatively rudimentary form of modeling whereby only one combination of variables is utilized. CPM Enterprise Predictive Analytics doesn’t confuse you with 1000’s of mediocre algorithms from which you have to choose from; instead we simply offer the Best-In-Class for each class of analysis.
1-seat / 4-CPU (up to 64-cores) 64-bit Windows annual subscription license, $39,995.00. Compare to SAS Enterprise costing 10 times as much and requiring programming…and the solution is not as profitable!
Cluster pricing available upon request.
Order your annual subscription license today! If you are not using CPM Enterprise Predictive Analytics you are leaving money on the table!