(A chinese translation of this post is available here)
I had the opportunity to incorporate a CLIPS expert system in one of my recent projects for bill plans logics and monitoring of health of our system modules. Both of them are classical sore-points for OO/procedural programming methodologies. We had made an attempt to implement the bill plan logics in an initial version using python, but it ended up in a meshed-up code with half-dozen levels of nested-if-then-else control structures. The system would have eventually ended up to become an excellent case study for project maintenance failure if we had persisted in using an OO/procedural language for the implementation.
Even though CLIPS has worked well for us, I still feel that it may not be suitable for all projects.
Requires radical change in programming paradigm.
Instead of executing your operations in a procedural manner, you have to "train" yourself to re-think your operations as separate rules; these rules operate in tandem in a series of recognize-act cycles. When a group of rules matches, a series of actions can be performed which may change the conditions of the rules in some manner. This may lead to other rulesets (or even the current ruleset) being triggered.
Another issue you would face is how to get these rulesets to "fire" in the order that you want without imposing too much restrictions on them. The conditions in each ruleset should not depend on previous states of other rulesets, i.e. they should not be coupled to each other.
Requires deep understanding of the knowledge domain.
To model the rules, you will need to arrange your knowledge on how things operate in your problem domain into a series of rules/patterns. Usually, you can attempt to perform this knowledge modeling process yourself, or to engage a knowledge modeling expert to assist you.
Execution should be data-driven or pattern-driven.
If you encounter a need to perform lots of if-conditional statements on many variables, the execution process is most likely data-driven or pattern-driven. Any non-trivial knowledge domain usually involves more than a dozen variables to work with. In OO/procedural, this means your source code will end up in meshed-up manner with multi-levels of nested if-else statements. CLIPS language syntax allows you to specify the conditions (i.e. rules) for a group of tasks to be executed in an organized manner.
However, one must be careful to plan and organize the data structure of the CLIPS facts and classes. Every datum in each CLIPS class and facts should be well-encapsulated using the same data-encapsulation principles in OO development methdology).