≡ Menu

First Look – Predigy


Predigy is a technology originally developed by Intelligent Results (founded in 2001) that was acquired by First Data in 2007. It was originally focused on the military (particularly on the analysis of unstructured data) but has subsequently moved into commercial applications. Predigy is now a decisioning platform with some applications in banking, collections, telecommunications and utilities. These applications are primarily in marketing, customer care, loyalty, retention, risk, fraud, collections and recovery. The intent of the platform is to support data-driven decisions throughout the customer lifecycle.

Predigy supports decision trees, predictive models and strategy design. It provides an offline business environment with some simulation and a production engine. This is all provided in a hosted environment and is web based as part of First Data’s environment. Predigy has four main components:

  • Cluster
    An automated approach to finding natural groupings in portfolio accounts -Partitioning Around Medoid (an instance that best represents a cluster) is a proprietary clustering technique available only in Predigy. The use of the Medoid means that each cluster has an exemplar.
  • Modeler
    Develop and deploy new models. Strong points are fast deployment, comparison tools and the ability to use directly in strategies. The models can include unstructured data and, like a growing range of modeling tools, provide a lot of automation of the leg work involved in developing models.
  • Strategy
    Offline design of decision tree/model combinations – strategies – combined with the ability to use data to see business impact through simulation of strategies with various datasets.
  • Production Engine

The product is increasingly integrated with First Data data stores and this means less data work for a typical FD customer. The base environment offers datasets, variables, formulas, samples, strategies, models, clusters and reports. Some interesting points:

  • Variables can be explicit or calculated or even models – integration means that everything, even models, shows as a variable. Also supports formulaic clean up of data with extra variables -to fill in blanks for instance, transform and limit.
  • Modeling is all automated, can use structured and unstructured data without issue. The automated unstructured analysis simply looks for strings and maps to outcomes – explicit entity identification in unstructured data is manual. The environment is aimed at modelers and focuses on automating base tasks like data cleaning and finding variables.
  • Formulas allow you to define calculations about which you care in the tree. Formulae can be very complex and then built into decision trees. There is a function builder interface as well as java like language for modelers.
  • Models can be created in other tools and use Predigy as the execution engine – the user simply pastes the model definition into a formula. PMML support is under development
  • Sampling is support for building/testing, focusing in on a subset of population.
  • Strategies are a decision tree environment. The tool tries to simplify the trees required by using models as characteristics and cuts. The user can specify costs and benefits of different actions on nodes for simulation/analysis and can drill into the data to see cost/benefit of the strategy. The decision making leaves on the trees have codes for either the customer’s own downstream system or the FD systems/call center/letter shop. The trees provide strategy code and action code to support Champion/Challenger.
  • Clustering technology (PAM) is undirected model development – what are the groups in this dataset. Because medoids are real people the clusters can be described. The modeling environment is for targeting specific outcomes – directed analytics.
  • The simulation tool flows sample data through the tree. The user can ask what would have happened if I had deployed this in the past. They can apply factors/constants/formulae to see what the outcomes would look like – target response rates for a given profit or return given a known response rate for instance.

Predigy is a nice, all-in-one decisioning platform squarely focused on the customer lifecycle, especially the customer credit lifecycle. It’s hosted platform and integration with First Data datasets are interesting capabilities. Currently it has no other rule formats- just decision trees, though this may change.


Comments on this entry are closed.

  • Geert Van Brandt February 17, 2009, 8:18 am

    Interesting evaluation. Do you know what the pricing structure is for using Predigy?

    • James Taylor February 17, 2009, 11:01 pm

      I don’t but I will ask and see if someone will reply

    • James Taylor February 19, 2009, 5:55 pm

      Here’s the response from the folks at First Data
      “Pricing for our hosted service is based on volume processed in addition to a seat license fee per module (Strategy, Modeler, Cluster).”