Does a Business Rules Management System (BRMS) lower overall project costs?

In summary, it can, but generally not if a conventional architecture pattern is used.

A clear-cut case is DROOLS, the open-source java based rules engine. A typical architecture is to pass transactions via SOAP to DROOLS to be evaluated.  The DROOLS rules are typically written in java code (well almost – see out tab on Rule Patterns), so there is no obvious cost advantage over coding the rules as a normal application.

In addition the analysts have to separate the rules from the use-cases and express the fact model, the developers have to create the SOAP interface, and implement the fact model, as well as splitting the flow of their applications to include the interaction with the rules engine.

The nett result is more expense than a conventional system: a price has been paid for separating the rules.

That is not always the case: if the rules form a dominant part of the system, then a rules-centric architecture can be used.

A case study: 30 institutions are obliged to implement a stringent and ever-changing regulatory process for client applications.  Those applications are the subject of inspections by the regulatory body, and the institutions are subjected to large fines where application are found to be incorrect.  The scale of the fines is a prime driver of project budgets.

Typically all the systems have the same features, with a Logical Data Model (LDM) of 200-300 entities. As a rule of thumb, for the Enterprise RedHat or .Net stack, multiplying the number of entities by $100,000 gives the ballpark project cost: $25M.

The institutions are obliged to publish their performance and most fit the pattern. Larger institutions have attracted larger fines so their systems are bigger, and less able to cope with ongoing change: a vicious circle.

One medium sized institution stands out, recognised as best practice by the regulator, but also by their miniscule budgets (< $4M).
Their architecture is minimalist: the prime driver is the rules engine (actually ILOG/Rete, but no special features are used, so almost any BRMS would work as well) sitting on top of a handful of very generic java processes. Even the DB access is abstract, meaning the rules can pick and choose the tables/columns they wish to access.

This light-weight solution has given them the flexibility to keep up with very fast changes, as witnessed by their extremely low level of fines, and a client service rated as the best-of-breed by the regulator.

Currently the institutions are moving to online entry and management of applications for their clients.  In theory, we might expect the light-weight solution to stumble at this point: to implement a one-page, ajax driven, MVP, secure website running against an enterprise database requires a substantial architecture.  They may be limited to an HTML equivalent of their paper forms, with simple validation, and in-depth validation only on form completion.