First Look: Experian Marketing Solutions

June 24, 2013

in Analytics, Decision Management, Optimization, Product News


I got an update on Marketswitch Optimization as part of my series on marketing decision management solutions. Experian positions Marketswitch Optimization for the most advanced stage of decisioning– as clients move from profiling and segmenting their customers, to predictive scoring and business rules to optimization. They see some interesting numbers for improvement through this lifecycle with incremental progress at each stage concluding with optimization where very complex decision problems are solved at the individual customer level. Some projects just add the optimization layer whilst others include predictive modelling for a combined benefit. Marketswitch Optimization is one of Experian’s more broadly deployed capabilities with customers around the world in many different industries (though mostly in financial services or telecommunications with some focus in media and utilities). It is applied across the lifecycle but prospecting and new business, cross-sell/up-sell dominate with some collections and account management also.

Marketswitch Optimization takes all of a group’s metrics and creates optimal, personalized customer treatment decisions to maximize these organizational objectives. Its original focus was on marketing and customer treatment, often adding optimization to existing contact management and content systems, and it is still primarily used for next best action or next best offer. Marketswitch Optimization aims to deliver a sophisticated solution based on proprietary mathematics for very complex decisioning problems that is easy to use and understand. Its graphical user interface allows business users to build and validate the optimal strategies produced. It’s also fast and flexible enough to be deployed in batch or real-time, connected to a database or flat files and delivering optimal answers to campaign management and decisioning tools. Experian talks about solving problems at the individual customer level, which they say is a major challenge facing businesses that have very complex customer decisioning challenges with many trade-offs to balance.

The basic usage for Marketswitch Optimization is a closed loop from business problem definition, to strategy design and optimization, to execution and performance review before re-evaluating the business problem etc. Functionality includes the optimization engine, handling uncertainty, assessing sensitivity of solutions, an ability to produce decision trees or strategy trees from the optimization results, PMML export for integrating with other decision engines, and reporting and analysis tools.

Marketswitch Optimization is a mature product and is currently at version 6. With version 6 the core workspace is a client development environment aimed at a fairly typical analytical marketing user – not a mathematician but someone familiar with analytic marketing. Projects within the tool manipulate a set of business objects, data, from files or databases. Samples can be defined and processing tasks can be defined to manipulate the data (sampling, de-normalizing, creating attributes) as it is pulled in from external sources. As part of the data management a wide range of statistical analyses can be performed on the data showing distributions, overlaps, frequency etc.

Users can define optimization templates in a graphical user environment which allow for multiple strategies to be defined without writing code. These templates allow the user to define the constraints and variables, identify the metrics and pick various goal functions to generate a set of outputs (goals, constants and results). Marketswitch Optimization ships with a range of marketing and contact optimization templates. Besides these specific productized templates for the core marketing functionality Experian also has a library of templates created over the years and the ability to build custom ones.

Users can also define decision structure to specify eligibility conditions and economics – how to calculate the value of an offer for a customer, for example. It also allows peer groups to be defined such as all the offers of a certain type, targeted at a segment or available in a channel. These can then be constrained as groups also.

Scenarios themselves consist of a goal function, a set of constraints both on offers and customers, a set of contact rules and a set of targeted metrics. Scenarios can also be built that use uncertainty, applying worst case, average or probabilistic constraints on values. Uncertainty can also be applied to data coming in about customers.

Once a scenario is solved a report is generated to show how these metrics are being met by the scenario and allow analysis by offer, by peer group etc. Reporting shows what has been constrained and allows slicing and dicing. A tool exists also for scenario sensitivity analysis, allowing a key input or constraint to be varied to see how sensitive the output (in terms of goals and metrics) is to the value of the input value. Swapset analysis allows users to compare scenarios and see how the optimal actions change between two strategies. Users can deploy finalized scenarios as a set of defined actions to take or they can use the Strategy Tree Optimization function to produce a decision tree or strategy tree that mimics the optimal answers. These can be widely deployed using PMML. Marketswitch solutions to scenarios can also be deployed in real time with customer records being processed one at a time as they’re passed from a customer interaction layer (website content management system, POS terminal, ATM, etc).

Ongoing model monitoring is important too and Marketswitch allows you to monitor model quality in the context of the sensitivity of the model – making sure the model quality and up to date ness is acceptable given how sensitive the model is. Impact analysis is also provided through reporting on the possible impact of improvements in the model. Model calibration is also available and this provides statistical calibration of the model to the most recent results, allowing it to adjust its behavior within defined bounds automatically removing unjustifiable accuracy and making small changes as data about results comes in.

Other elements include

  • Rollout to create batch jobs of actions
  • Statistics on data and decision structure
  • Detailed reporting on scenario performance against production data
  • Scenario statistics
  • Multi-goal optimization

You can get more information on Marketswitch Optimization here.


Previous post:

Next post: