I was reading some discussion about data models the other day and came across the following:

“The relationships between entities define the rules about which entities relate to others, as well as the number of minimum and maximum occurrences on each side of the relationship.”

Rules is definitely the wrong choice of word here. The misconception that relationships per se in data models represent (business) rules goes all the way back to the 1980s. It’s simply wrong. Instead I would say:

The relationships between entities provide structure for the data model, specifically indicating which entities relate to which and how. Specifications for a relationship typically indicate the number of minimum and maximum occurrences allowed on each side of that relationship – i.e., their cardinalities.

Of the three typical cardinality values – zero, one and many – only one removes any degree of freedom, and thus could be said to represent a rule. In any case, cardinality per se is a specification device; the business rule is in the meaning – for example:

A customer may be related to at most only one sales area.

I get this question all the time, and it’s a painful one, so let me answer on the record. Question: In our enterprise architecture tooling, there’s a business dimension in which we define Business Concepts (the real business language), and an IT dimension containing Information Objects (data organization model). How can we solve the problem that business expresses rules as they relate to Business Concepts, while IT needs to translate these into rules related to Information Objects? We don’t want to bother business with IT model concerns, nor duplicate the rules in two places. Can you please shed light on an elegant approach to this dilemma? My answer: The standard SBVR[1] provides the ‘elegant’ approach, which is technology that can “read” language based on the business vocabulary (e.g., RuleSpeak) and/or dialog with people to disambiguate those statements. Until such technology is commercially available – and why not, look what IBM Watson can do! – two forms of each statement are unfortunately necessary. The key for your rule management regime is to maintain traceability between them. By the way, the mapping is almost certainly 1:m, not 1:1. I wish I had a better answer, but there just is none today. All I can say is that current implementation technologies for business rules are very, very primitive. ~~~~~~~~~~~ Acks: Tom Andries www.BRSolutions.com

In a data modeling discussion forum, a practitioner recently asked:

Has anyone seen an object super-type used to encompass both Party and System Component?

Examples of Parties: people, organizations

Examples of System Components: projects, proposals, computer systems

In over a dozen responses, not one person asked for a definition! My Response I’m surprised … no amazed … no horrified … by this discussion. It pretty well illustrates my frustration with data modeling. I should simply point out that in the real world, projects and proposal are not system components in the sense probably meant, then stop right there. Clearly the question mixes apples and oranges. But let’s work it through a bit more. What is the thing that that the proposed supertype (more general concept) would represent? … Which is to say, how would you define the notion in your head when you say the corresponding thing exists in the real world? If you can’t define the concept in any coherent way, that should be the end of the discussion. This is why I recommend starting with concept models, instead of data models. Since I’ve done a few data models in my time, I’ll jump to the other side of the fence and propose a definition for the sake of discussion.

more general concept x: an agent that can make and respond to requests

Systems, people, and organizations can all do those things. However, proposals can’t … unless you are talking about a software object representing a proposal. So here’s a revised definition …

more general concept x: a software agent that can make and respond to requests

But any software object can basically do that. Is that what is meant? Are we talking about (a) things that exist (can be jointly classified) in the real world, or (b) things that exist in software representations of the real world? My fear is that concept x is simply a design artifact that doesn’t really correspond to anything. It’s just an optimization for technology’s sake. Do we really want to go there?!

As part of a continuing dialog, I recently asked these questions: What does the term “conceptual data model” really mean? Is it the best term for what is meant? To me, it sounds like “conceptual data model” might be about “conceptual data”. Surely not(?). What exactly then? (Some of my thoughts on the matter: http://goo.gl/8GX5o.) A practitioner responded with the following 3-part challenge below. With each challenge is my response. Challenge part 1: Which of the following, if any, would you see in a conceptual data model: primary keys, foreign keys, abstract keys, constraints, nulls, non-nulls, 1:1s, 1:Ms, optionality, cardinality, etc.? My response: Wrong approach. Bad definition practice one three counts. 1. Things should be defined in terms of their essence, what they are, not in terms of what are allowed or disallowed. 2. Suppose none of the above were allowed. You’d end up simply with a description of what the thing is not, not what it is. Since you can’t possibly list everything it’s not, the definition is woefully incomplete. 3. The list attempts to describe something in terms of an implementation or design of the thing, not in the thing’s own terms. That’s simply backwards. Challenge part 2: So let’s turn the question around. What would be the minimum types of objects you need to explore a concept, produce a mini model, and convey knowledge about the concept and its relationship to other concepts? My response: To develop a concept you need: 1. A concise definition for the concept, which distinguishes the concept from all other concepts and conveys its essence to the business. 2. A business-friendly signifier (term) for the concept that has only that one meaning (or is properly disambiguated by context). 3. A set of facts that indicates the place of the concept within a structure of other concepts (e.g., classification, categorization). 4. A set of verbs that permits the business to communicate unambiguously about how the concept relates to other concepts (e.g., customer places order). 5. A set of structural rules for the concept and its relations to other concepts, as needed. Why the verbs? Because you can’t write sentences without verbs and business rules can always be written in sentences. ‘Requirements’ should be too(!). Challenge part 3: When you are finished do you call that a ‘conceptual data model’? My response: You notice that I didn’t say anything, directly or indirectly, about “data”. That term is irrelevant. Taking away “data” leaves “conceptual model”. But that term suggests what the model is made of (concepts), not what it is about. Of course the kind of model we’re talking about is made up of concepts – it’s certainly not made of physical material (e.g., wood, plaster, etc.) or through use of math (some formula). That leaves “concept model” (also sometimes called “fact model”). Bingo. Is there a standard for this kind of model? Indeed there is: SBVR (Semantics of Business Vocabulary and Business Rules). See the SBVR Insider section on www.BRCommunity.com for more information.

In the real world, we communicate with sentences … just like we are doing now. If a “data model” can’t support writing meaningful and consistent sentences, then surely something is amiss. Business rules are always sentences, so you can see where I’m going with this. So I come to this radical and provocative suspicion: The reason data models remain so often disconnected from other things in the business and IT, and why data modelers generally so under-empowered, is that data modeling practices have been removed them from direct engagement with business logic. Shouldn’t be that way. Won’t ever work well. I’m sure there are a whole lot of people who disagree with me. So I am going to write a series of posts this week on the subject … and call this ‘data model’ week.

One of the first rules of business analysis should be never waste business people’s time. One of the fastest ways to waste their time is not knowing what they are talking about … literally … and do nothing about it. So you end up just wasting their time over and over again. Unacceptable. Is there a way to avoid it? Yes, by taking the time to understand exactly what concepts the business people mean when they use the words they use.  I believe business vocabulary should be job one for Business Analysts. If you don’t know (and can’t agree about) what the concepts mean, then (excuse me here for being blunt) you simply don’t know what you’re talking about. (And sometimes, unfortunately, neither do the business people … which is something important BAs should find out as early as possible.) So structured business vocabularies (fact models) are a critical business analysis tool. How else is there to analyze and communicate about complex know-how in a process-independent way?! Looking at the issue the other way around, you can make yourself look really smart about a complex area in a relatively short time by having and following a blueprint. We’ve had that experience many, many times in a wide variety of industries and problem areas. (Try jumping between insurance, pharmaceuticals, electricity markets, eCommerce, race care equipment, credit card fraud, trucking, taxation, healthcare, banking, mortgages, pension administration, ship inspections, and more! We do.) There’s no magic to it – like contractors for the construction of buildings, you must have or create structural blueprints. For operational business know-how, that means bringing an architect’s view to structure the concept system of the problem space …  just a fancy way of saying develop a well-structured business vocabulary. Then a whole lot of things will fall right into place for you. P.S. By the way, I’m not talking about any form of data modeling here. Also, there’s no real need to use the ‘S’ word (semantics) for it.  

I have been involved in a great on-going discussion on LinkedIn about data models. I posed the question: Is there any proven way to demonstrate data models are correct, complete, and stable with respect to the operational business and its needs? You might enjoy joining in: http://goo.gl/MsnXu It was literally 25 messages into the discussion that I realized “data model” was being used in two distinct ways in the discussion. And even then it had to be pointed out by a participant who seemed to know one of the other people.

• I always mean “data model” in the ‘old’ way, in which the data model supports real-time business operations (or close thereto). In that world, you must design for integrity, which generally means ‘highly normalized’ in the relational sense.
• In the old-but-not-nearly-as-old world of OLAP, real-time operations and updates are not a concern, so de-normalization (and redundancy) are presumably acceptable. (I’ll leave that question to the experts.)

That’s always the problem with vocabulary – deeply buried assumptions that prevent you from hearing what you need to hear. From experience, I know the trap oh-so-well, but here I fell right into it myself. What’s the answer to the question I posed for “data models” (of the kind I meant)? Focusing on the meaning and structure of business vocabulary, not data, as a core part of business analysis.  Note to self (a rule): When you enter any discussion, be clear what you mean by the terms you use – even (and maybe especially) the ‘obvious’ ones.