First look at JRules Scorecard Modeler

This entry is part 5 of 5 in the series First Looks at JRules 7

I recently have been able to use the new add-on to JRules called Scorecard Modeler.

Unlike the name may imply, Scorecard Modeler does not help you model scorecards at all. In other words, it is not a statistical analysis tool that would be used to create scorecard models, but simply an editor that allows easy creation of scorecards within the JRules environment.

What is a scorecard? I have recently written a post title “A quick introduction to scorecards” to go over some of the details of what a scorecard is. I suggest you start there before continuing reading this post.

Scorecard modeler is simply a new editor that will be added to your list of choices for adding rules to a rule project and in some ways it looks similar to a decision table editor.

It is a Rule Studio only editor so as of this writing, there is no RTS editor available. How does it work? It simply generates the usual IRL rules (like all other editors) for deployment. I suspect that IBM will be working on an RTS editor for this, but that is only speculation at this time.

The Review

Installation

The installation of the scorecard modeler requires that you should have JRules already installed (verson 6.7.x, 7.0 or 7.1). The installation will ask you to select the directory of your existing JRules install, but otherwise it is pretty straight forward.

The Editor

Once you have installed the modeler, when you relaunch your Rule Studio, the new editor will be available to create scorecards in JRules. Some of the options will change the number of columns that need to be filled in for the scorecard but overall it is simple to understand how to fill the information in.

There are some GUI quirks in the current interface (entering negative values is quirky, moving attributes up and down, and making the scorecard properties page appear for example) but these do not prevent the scorecard from working properly and I am hoping that IBM will eventually address these minor irritants.

Some odd behaviours

Scorecard modeler has some behaviours that are not typical of the other editors in JRules. Namely, the modeler automatically creates new BOM entries in your project and it creates a variable set that is used to create variables that point to each scorecard created. Now these are odd behaviours, but are obviously required for things to work properly. Once you are familiar with those behaviours it makes troubleshooting possibly a bit easier.

These behaviours are even more prominent if you are using an architecture that uses multiple projects to hold XOM, BOM and Rules and especially if you have multiple BOM projects. The location where scorecard modeler chooses to set the variable set up and add the BOMs might not be the one you want or expect.

Another odd behaviour that is not common to the other editors is that the scorecard modeler actually allows the selection of “non verbalized” elements from the BOM. I am not sure why that is and it is contrary to the normal behaviour of all other editors in JRules so it is good to note that a user is actually able to choose from items that are not usually available as choices (until they are verbalized).

The limitations

There are limitations that are good to know about upfront.

Naming of a scorecard has more limitations than any other “rule” in JRules. Don’t use spaces, dashes or underscores in your scorecard name. I believe that this limitation is due to the automatic creation of variables (see above) and that as long as you stick to those “naming rules” you will be doing just fine.

Renaming of a scorecard is a little more work than what we might be used to. You can easily rename the scorecard, but the automatically created variable that goes with this variable is not renamed automatically. In some cases the old variable name gets deleted, but the new one does not get created. This can easily be fixed by manually creating the variable name that matches your scorecard name in the proper variable set and everything will be OK.

Documentation generated by the rule reports will not include a copy of the scorecard and there is no copy and paste to and from the scorecard modeler. So documentation will have to be created by using the age old technique of screen capture and pasting images in documents… Not ideal. I am hoping that IBM addresses this in future releases.

There are other limitations and all users should get familiar with them by reading the readme file that comes with scorecard modeler.

The issues

As of version 7.0.2 there are some issues that I have encountered in a specific environment for which IBM had to provide hot fixes for making the scorecard modeler work in that environment. Version 7.1 fixes most of these issues save 1  as far as I can tell (some attributes with a specific configuration can’t be selected at all to be part of the scorecard).

In IBM’s defence, IBM support has been extremely responsive to help resolve any of the important issues  encountered and the hot fixes provided make it possible to use the scorecard modeler.

The Documentation

The documentation provided with scorecard modeler is more of a relatively straight forward tutorial than actual documentation. Some of the features are not documented at all or so badly documented that even IBM has a hard time figuring out what they are for…

My suggestion on this is that new users should walk through the tutorials provided and then immediately try to make their own scorecards work in the modeler.

One of the good things from the documentation point of view is that at least the API documentation is included so that if customizations are needed for reasoning strategies and things of that sort, developers will have a starting point for working on the customizations.

The Verdict

The scorecard modeler in some cases still feels like an “early version” of the software, but with IBM’s help (through support and hot fixes) the scorecard modeler ends up being a great tool to enter scorecards into JRules.

Although a bit quirky, overall I think this tool has great potential to be a great addition to JRules.

A quick introduction to scorecards

Some may ask what the relationship of scorecards to Business rules is…

Well it turns out that IBM has created an add on to add support for scorecards in JRules and that before I go on and do my review I decided it was necessary to write a post on what a scorecard is so that I don’t have to do it as part of my upcoming review… 🙂

What is a scorecard?

For those of you who don’t know what a scorecard is, it is a Risk Management tool used mostly by banks and other institutions to calculate the risk they take by selling you one of their products.

  • How risky a customer are you?
  • What are the chances that you might default on payment?
  • etc.

Based on that information, they may adapt their specific offer. For example, a good customer might get a better interest rate, a higher credit limit, etc.

It works like this (more or less):

  1. The company performs a statistical analysis on historical data that they have on all their existing customers
    • They may break the data down into smaller groups that have something in common (called segmentation) so that the resulting analysis is more precise
  2. From this analysis, they will identify characteristics (attributes or pieces of information) that are “predictive”
    • That is, the characteristics show a statistically significant relationship to past results (positive or negative)
  3. Each characteristic is then broken down into ranges of possible values
    • For example, a characteristic based on family income may be broken down into ranges as follows:
      • 1-30000
      • 30001-60000
      • 60001-90000
      • 90001-120000
      • over 120000
  4. Each range is then given a score and the attribute is also usually assigned an expected score. For example
  5. Characteristic Expected Score Range Score
    Family income 50 1-30000 10
    30001-60000 25
    60001-90000 40
    90001-120000 65
    over 120000 75
  6. This is repeated for many characteristics and combined together into a single scorecard.
  7. In addition to a score for each range, a reason code may be attached to a specific range value to provide some indication as to which attribute may have scored low (i.e. lower than the expected score).

In the end, when executing a scorecard the result will be a score calculated based on all the indivdual scores of each attribute and a list of reason codes (usually the codes associated to the attributes that score the most negative, if applicable).

This score can then be used as part of another calculation to calculate a risk rating, credit limit, interest rate, etc.

That is about it! Now you have a basic understanding of what a scorecard is. In the near future, I will post my review of the Scorecard Modeler add on for JRules.

In the meantime, if you have questions about scorecards or the post needs clarifications, feel free to comment and drop me a line!

#BRF Emerging Trends & Decisioning Panel – Follow up

This entry is part 24 of 24 in the series Business Rules Forum 2009

Last november I wrote many posts from the Business Rules Forum in Las Vegas, one of which was trying to capture the information from the “Emerging Trends and Decisioning Panel”. My original post can be found here: https://www.primatek.ca/blog/2009/11/05/brf-emerging-trends-decision-panel/

Well, BRCommunity has publish a full transcript of the panel which can be found at: http://www.BRCommunity.com/a2010/b547.html.

You may have to register to the web site to have access to the contents but if you are interested in this topic it is worth it.

Enjoy!