Many large companies now have knowledge management systems. These usually consist of reports categorized for easy access, plus "communities of practice" so that subject specialists can share arcana and best practices -- and so that subject experts can be found. The objective is for people inside a company to discover what someone else has already discovered and not "re-invent wheels."
The Toyota "A3 Paper" system has become well-known. It records a problem, its analysis, and the countermeasures taken using a concise, rigorous format. When faced with a problem, the first step is to check whether it has been seen before, and if so, what was considered root cause, and what was done. For this to work, the last step of any problem-solving episode is to condense the thinking into A3 format and enter it into the system. This system is now so huge that classifying entries for global access is a problem. However, for example, assemblers can see detailed A3s from other assemblers at similar stations in all plants world-wide. But the power of the system is that using it reinforces Toyota's logic of problem solving, simplifying internal communication using a common "science-based language."
Internally, this is much better than local-format (or free-form) records in disconnected areas. However, it may not help deal with external problems, fuzzy problems that don't fit the format, or "wicked problems." It obviously failed in the now famous cases of sticky accelerators and squirrelly brakes. There, early-on evaluation of low-frequency failure from the field may not cause people to agree that a problem is "real." All companies with complex products have a similar dilemma: when does skimpy evidence warrant action?
Complex, integrative software presents other problems. Just checking its integrity requires software and simulations -- models checking models. The Toyota learning system appears to have never quite mastered this stage, but no one else claims world benchmark status either. The learning to assure software integrity and interoperability mandates using computer-generated data and analysis.
Beyond that, learning systems have to extend further, linking core organizations with partners, and with "life cycle records" from the field. Computer companies have done this for some time. In Compression, such systems must improve the efficiency and effectiveness of life-cycle operations on the fly -- a new world of operational challenge.