Item identification rules
"Intelligent" identification schemes are obsolete
Many manual identification processes are simply incompatible with the best PLM practices. To take full advantage of your new PLM system, throw out the excess baggage.
Traditional "document control" identification practices originated over 40 years ago; although they're costly and cumbersome, they were reasonable compromises given the lack of available technology. At the time, a fundamental assumption was that finding information is very difficult, and it was therefore necessary to overload item identification properties with search-specific "helper" data.
There are (among other holdovers) "intelligent part numbering" schemes, change requests that share identifying numbers with change orders, pre-release document revisions that must be updated upon release, and rigid item naming conventions.
All of this is akin to embedding your car's color, motor oil weight, and tire size in the registration license number, or encoding in the street address the number of rooms in your residence.
Schemes for overloading information in an item's identifying numbers and revisions can be surprisingly expensive to create, manage, and interpret. Unfortunately, the cost is often widely distributed across the organization over time, and therefore never seriously examined.
Suppose each 12-character significant identifier takes an average of just 5 seconds longer to type or write down, correct transposition errors, and otherwise manipulate compared to an 8-character value. (In fact, for many tasks a short number won't need to be written at all, and transposition error rates exponentially increase with identifier length.) If your PLM user works with 100 item numbers per day, that's an extra productivity tax of 7 hours (almost an entire work day!) per year, per person. Make sure your number scheme is really cost-effective.
There are substantial and unavoidable costs in initially designing a significant numbering scheme, entering and checking the longer character strings, canceling and re-releasing items incorrectly categorized, repairing the scheme when it breaks (as it inevitably does), and merging it with other companies' conflicting schemes after an acquisition. (In a merger of PLM systems, the non-significant system typically wins because it's simply less work.)
The essential characteristic of an item number is that it is guaranteed unique and absolutely unchanging for the life of the item, even as the item and its environment evolve. Anything less means there's expensive rework in your future.
With virtually unlimited standard and custom attributes available in a PLM system, an item identifier need not encode common attributes such as document type, part category or product family.
Therefore: Put your confidence in the extensive search capabilities available in PLM databases.
Distinct items do not share identifying numbers
In the old days, it made perfect sense to have documents and parts share numbers; after all, researching these relationships was difficult and expensive. The document number typically was embedded within the part number (e.g., document 12345 described parts 12345-01 and 12345-02). However, documents and parts have different lifecycles. Part interchangeability rules may force a new part number where only a document revision is required; likewise, for clarity a new document may be created to describe one or more existing parts.
Another obsolete method is when related design document types share a common root, and extensions indicate a particular type. For example, a circuit schematic might be 45678-01, printed circuit layout 45678-02, and PCB drill pattern 45678-03. But you may later need to "upsize" the schematic with a new document number to cover an entire product family with other PCB layouts and drill patterns, and the original relationships would no longer exist.
All PLM systems let you make document and part relationships explicit, independent of their identifying numbers. Don't try to maintain these relationships through item identifiers.
Item identifiers are short
Item numbers exist as "handles" for our parts and documents. The human mind has limited ability to remember long sequences of characters. Using a part number longer than 7 characters will generally require more than half your workforce to write it down before they can act on it.
The ideal numbering system can be astonishingly simple, requiring only a single sequence (or perhaps a separate sequence for documents, parts, and changes) of 5 to 8 characters, providing a unique identifier for up to 100 million items. In practice, it's a very sparsely-populated region.
For the psychological foundation of this rule, begin at the source: G. A. Miller, "The magical number seven, plus or minus two: some limits on our capacity for processing information." Psychological Review (1956)
Item identifiers are numeric
Use only numeric characters (0..9) in your document and part numbers, as letters can be difficult to read in some typefaces (the worst culprits being BIOQSZ).
Furthermore, people who spend a lot of time working with item identifiers (purchasing agents, production analysts, receiving clerks) get quite good using their PC's numeric keypad, and letters can significantly reduce their productivity.
If you ignore all of the above, use chunking
If you must, say for historical reasons, have long item identifiers, split them into smaller segments ("chunks"). Humans handle chunks of data better than unbroken strings, though the limit is about 3 to 5 chunks. Most PLM systems allow static prefixes and suffixes, as well as a computer-assigned sequential base (or root) value. So, a common part number scheme consists of a category prefix of 2 or 3 characters, followed by a sequential value, with a human-managed iteration or style suffix: 302-03597-002. It's not terribly friendly, but it is far superior to an unbroken 30203597002.
Forty-five years after Miller, this human-factors rule was described in Cowan, Nelson, "The magical number 4 in short-term memory: A reconsideration of mental storage capacity." Behavioral and Brain Sciences (2001)