Category: DM

I’m squeezing in one last session before flying out: Kevin Chase, SVP of Implementation Services at ING, discussing how to use rules in a multi-client environment, specifically on the issues of reuse and reliability. I’ve done quite a bit of work implementing processes in multi-client environments — such as a mutual funds back-office outsourcing firm — and the different rules for each client can make for some challenges. in most cases, these companies are all governed by the same regulations, but have their own way that they want things done, even if they’re not the ones doing it.

In ING’s case, they’re doing benefits plan administration, such as retirement (401k) plans, for large clients, and have been using rules for about six years. They originally did a pilot project with one client, then rolled it out to all their clients, but didn’t see the benefits that they expected; that caused them to create a center of excellence, and now they’re refining their processes and expanding the use of rules to other areas.

They’re using rules for some complex pension calculations, replacing a previous proprietary system that offered no reuse for adding new clients, and didn’t have the scalability, flexibility and performance that they required to stay competitive. The pension calculator is a key component of pension administration, and calculating pensions (not processing transactions) represented a big part of their costs, which makes it a competitive differentiator. With limited budget and resources, they selected ILOG rules technology to replace their pension calculator, creating a fairly standalone calculator with specific interfaces to other systems. This limited-integrated approach worked well for them, and he recommended that if you have a complex calculator as part of your main business (think underwriting as another example), consider implementing rules to create a standalone or lightly-integrated calculator.

In their first implementation phase, they rewrote 50+ functions from their old calculator in Java, then used the rules engine to call the right function at the right time to create the first version of the new calculator. The calculations matched their old system (phwew!) and they improved their performance and maintainability. They also improved the transparency of the calculations: it was now possible to see how a particular result was reached. The rules were written directly by their business users, although those users are actuaries with heavy math backgrounds, so likely don’t represent the skill level of a typical business user in other industries. They focused on keeping it simple and not overbuilding, and used the IT staff to build tools, not create custom applications. This is a nice echo of Kathy Long’s presentation earlier today, which said to create the rules and let the business users create their own business processes. In fact, ING uses their own people for writing rules, and uses ILOG’s professional services only for strategic advice, but never for writing code.

After the initial implementation, they rolled it out to the remainder of their client base (six more organizations), representing more than 200,000 plan participants. Since they weren’t achieving the benefits that they expected, they went back to analyze where they could improve it:

• Each new client was still being implemented by separate teams, so there was little standardization and reuse, and some significant maintenance and quality problems. It took them a while to convince management that the problem was the process of creating and maintaining rules, not the rules technology itself; eventually they created a center of excellence that isn’t just a mentoring/training group, but a group of rules experts who actually write and maintain all rules. This allows them to enforce standards, and the use of peer reviews within the CoE improves quality. They grow and shrink this team (around 12-15 people) as the workload requires, and this centralized team handles all clients to provide greater reuse and knowledge transfer.
• They weren’t keeping up with ILOG product upgrades, mostly because it just wasn’t a priority to them, and were missing out on several major improvements as well as owning a product that was about to go out of maintenance. Since then, they’ve done some upgrades and although they’re not at the current release, they’re getting closer and have eliminated a lot of their custom code since those features are now included in the base product. The newer version also gives them better performance. I see this problem a lot with BPMS implementations as well, especially if a lot of custom code has been written that is specific to a current product version.
• They had high infrastructure costs since each new client resulted in additional hardware and the associated CPU licensing. They’ve moved to a Linux platform (from SUN Solaris), moved from WebLogic to JBOSS, and created a farm of shared rules servers.
• Since they reduced the time and expense of building the calculator, they’ve now exposed other areas of pension administration (such as correspondence) that are taking much longer to implement: the pension calculator used to be the bottleneck in rolling out new products, but now other areas were on the critical path. That’s a nice thing for the calculator group, but had them start to recognize the problems in other areas and systems, pushing them to expand their rules capability into areas such as regulatory updates that span clients.

This last point has led to their current state, which is one of expansion and maturity. One major challenge is the cleanliness and integrity of data: data errors can lead to the inability to make calculations (e.g., missing birthdate) or incorrect calculation of benefits. They’re now using rules to check data and identify issues prior to executing the calculation rules, checking the input data for 30+ inconsistencies that could cause a failure in the calculator, and alerting operations staff if there needs to be some sort of manual correction or followup with the client. After the calculations are done, more data cleansing rules check for another 20+ inconsistencies, and might result in holding up final outbound correspondence to the participant until the problem is resolved.

He wrapped up with their key lessons learned:

• A strong champion at the senior executive level is required, since this is a departure from the usual way of doing things.
• A center of excellence yields great benefits in terms of quality and reuse.
• Leverage the vendors’ expertise strategically, not to do the bulk of your implementation; use your own staff or consultants who understand your business to do the tactical work.
• Use an iterative and phased approach for implementation.
• Do regular assessments of where you are, and don’t be afraid to admit that mistakes were made and improvements can be made.
• Keep up with the technology, especially in fast-moving technologies like rules, although it’s not necessary to be right on the leading edge.

Great presentation with lots of practical tips, even if you’re not in the pension administration business.

For the first breakout this morning, I attended Gladys Lam’s session on organizing a business rule harvesting project, specifically on how to split up the tasks amongst team members. Gladys does a lot of this sort of work directly with customers, so she has a wealth of practical experience to back up her presentation.

She first looked at the difference between business process rules and decisioning rules, and had an interesting diagram showing how specific business process rules are mapped into decisioning rules: in a BPMS, that’s the point where we would (should) be making a call to a BRMS rather than handling the logic directly in the process model.

The business processes typically drive the rule harvesting efforts, since rule harvesting is really about extracting and externalizing rules from the processes. That means that one or more analysts need to comb through the business processes and determine the rules inherent in those processes. As processes get large and complex, then the work needs to be divided up amongst an analyst team. Her recommendations:

• If you have limited resources and there are less than 20 rules/decisions per task, divide it up by workflow
• If there are more than 20 rules per task, divide by task

My problem here is that she doesn’t fully define task, workflow and process in this context; I think that “task” is really a “subprocess”, and “workflow” is a top-level process. Moving on:

• If there are more than 50 rules per task, divide by decision point; e.g., a decision about eligibility for auto insurance could be broken down into decision points based on proof of insurance, driving history, insurance risk score and other factors

She later also discussed dividing by value chain function and level of composition, but didn’t specify when you would use those techniques.

The key is to look at the product value chain inherent in your process — from raw materials through production, tracking, sales and support — and what decisions are key to supporting that value chain. In health insurance, for example, you might see a value chain as follows:

1. Develop insurance product components
2. Create insurance products
3. Sell insurance products to clients
4. Sign-up clients (finalize plans)
5. Enroll members and dependents
6. Take claims and dispense benefits
7. Retire products

Now, consider the rules related to each of those steps in the value chain (numbers correspond to above list):

1. Product component rules, e.g., a scheduled payout method must have a frequency and a duration
2. Product composition rules, e.g., the product “basic life” must include a maximum
3. Product templating rules, e.g., the “basic life” minimum dollar amount must not be less than $1000
4. Product component decision choice rules, e.g., a client may have a plan with the “optional life” product only if the client has a plan with a “basic life” product
5. Membership rules, e.g., a spouse of a primary plan member must not select an option that a plan member has not selected for “basic life” product
6. Pay-out rules, e.g., total amount paid for hospital stay must be calculated as sum of each hospital payment made for claimant within claimant’s entire coverage period
7. Product discontinuation rules, e.g., a product that is over 5 years old and that is not a sold product must be discontinued

These rules should not be specific to being applied at specific points in the process — my earlier comment on the opening keynote on the independence of rules and process — but represent the policies that govern your business.

Drilling down into how to actually define the rules, she had a number of ways that you to consider splitting up the rules to allow them to fully defined. Keeping with the health insurance example, you would need to define product rules, e.g., coverage, and client rules, e.g., age, geographical location, marital status, and relationship to member. Then, you need to consider how those rules interact and combine to ensure that you cover all possible scenarios, a process that is served well by tools such as decision tables to compare, for example, product by geographic region.

This is going to lead to a broad set of rules covering the different business scenarios, and the constraints that those rules apply to different parts of your business processes: in the health insurance scenario that includes rules that impact how you sell the product, sign up members, and process claims.

You have to understand the scope before getting started with rule harvesting, or you risk having a rule harvesting project that balloons out of control and defines rules that may never be used. You may trade off going wide (across the value chain) versus going deep (drill down on one component of the value chain), or some combination of both, in order to address the current pain points or support a process automation initiative in one area. There are very low-level atomic rules, such as the maximum age for a dependent child, which also need to be captured: these are the sorts of rules that are often coded into multiple systems because of the mistaken belief that they will never change, which causes a lot of headaches when they do. You also need to look for patterns in rules, to allow for faster definition of the rules that follow a common pattern.

Proving that she knows a lot more than insurance, Gladys showed us some other examples of value chains and the associated rules in retailing and human resources.

Underlying all of the rule definitions, you also need to have a common fact model that you use as the basis for all rules: this defines the atomic elements and concepts of your business, the relationships between them, and the terminology.

In addition to a sort of entity-relationship diagram, you also need a concepts catalog that defines each term and any synonyms that might be used. This fact model and the associated terms will then provide a dictionary and framework for the rule harvesting/definition efforts.

All of this sounds a bit overwhelming and complex on the surface, but her key point is around the types of organization and structure that you need to put in place in your rules harvesting projects in order to achieve success. If you want to be really successful, I’d recommend calling Gladys. 🙂

All the usual suspects joined on a panel at the end of the day to discuss the vendor view of business rules: Pegasystems, InRule, Corticon, Fair Isaac ,ILOG (soon to be IBM) and Delta-R, moderated by John Rymer of Forrester.

The focus was on what happening to the rules market, especially in light of the big guys like SAP and IBM joining the rules fray. Most of them think that it’s a good thing to have the large vendors in there because it raises the profile of and validates rules as a technology; likely the smaller players can innovate faster so can still carve out a reasonable piece of the market. Having seen exactly this same scenario play out in the BPM space, I think that they’re right about this.

The ILOG/IBM speaker talked about the integration of business rules and BPM as a primary driver — which of course Pega agreed with — but also the integration of rules, ETL and other technologies. Other speakers discussed the importance of decision management as opposed to just rules management, especially with regards to detecting and ameliorating (if not actually avoiding) situations like the current financial crisis; the use of predictive analytics in the context of being able to change decisions in response to changing conditions; and the current state of standards in rules management. There was a discussion about the difference between rules management and decision management, which I don’t believe answered the question with any certainty for most of the audience: when a speaker says “there’s a subtle but important difference” while making hand motions but doesn’t really elaborate, you know that you’re deep in the weeds. The Delta-R speaker characterizes decision management as rules management plus predictive modeling; I think that all of the vendors agree that decision management is a superset of rules management, but there are at least three different views on what forms that superset.

As a BPM bigot, I see rules as just another part of the services layer; I think that there’s opportunity for BRM in the cloud to be deployed and used much more easily than BPM in the cloud (making a web services call from a process or app to an external rules system isn’t very different than making a web services call to an internal rules system), but I didn’t hear that from any of the vendors.

That’s it for the day; I know that the blogging was light today, but it should be back to normal tomorrow. I’m off to the vendor expo to check out some of the products.