“Data Management Isn’t Working” – An Interview with Graeme Simsion
Graeme Simsion is an Australian with a world-renowned
reputation in the field of data modeling, and more generally as an advisor
to CIOs and CEOs about their IT strategies. Recently, at the DAMA
International Symposium and Wilshire Meta-Data Conference in San Antonio,
he told the audience of one panel session that “data management
isn’t working.” As you might expect, the reactions of the crowd varied
wildly. Tony Shaw, the moderator of that panel and program chairman
of Wilshire Conferences, asked Graeme to explain himself.
Wilshire: Graeme, you have a reputation for your no-nonsense, pragmatic views. Often irreverent too. Recently you said that “data management isn’t working.” Now, admittedly this remark was in the context of a “playful” conference panel, but do you really hold this opinion? Or was the statement just for the sake of provocation? And if you stand by the assertion, what exactly do you mean by it?
Graeme Simsion: Not everyone there took it playfully. We had a panel of experts preaching to the choir, and I wanted to shake it up a bit; and got the full range of reactions. However, I’ve been saying essentially the same thing for a number of years: the traditional data administration model – corporate data model, meta-data repository, policies and policing, broad targets, long-term goals – seldom delivered on its promises in the past and is even less likely to do so in the future.
I’ve been proposing a more tactical “guerilla” approach, focusing on improving data management in highly targeted, measurable areas, and for the first time at this conference I admitted that it hadn’t always succeeded. I caught a bit of flak along the lines of “I wouldn’t hire you as a consultant if your projects don’t succeed”, but to tell the truth I’m getting wary of success stories that haven’t been monitored over any extended period, and told by people with a vested interest. Just look at the number of organizations that are downsizing or closing their data management functions – that’s the evidence of how the discipline is regarded.
Wilshire: I can’t believe you’d have been in the data business for so long if you thought it was all a waste of time. What has shaped your current perspectives?
Simsion: From very early on I believed that data management addressed a really important issue – and that organizations that succeeded with it could gain a real advantage. On the other hand, it patently wasn’t easy. An eye-opener for me was a tour I took back in ‘88 as part of a consulting assignment, looking at the DM function in banks in the US and UK. Standing back from the optimism of the data managers, I just didn’t see value being delivered to the business – at least not in the terms that had been promised. Lots of optimism about the future – not many points on the board.
Wilshire: So what’s the bottom line? Are you advocating the complete abandonment of the DM function now?
Simsion: Not at all. Since the early assignments I just mentioned, my interest has been in what you can do to make it work. Being involved with business people has helped me see the goals in better perspective: these days I’d start a data management assignment by getting business people to identify key areas in which poor data management is hurting them – or where there are opportunities to gain advantage from better data management – and then put some real dollars on them. I’d then be looking at the cost of tackling them, and wanting to establish a contract with the business on that basis. This is a long way from “first we develop an architecture” and so on. But it’s still hard, and one of the obstacles is the ingrained ideas of what data management teams should do. I’ve seen situations where DM teams are cruising along on this “business project” model, then they say – “good, now we can use the money to build a model, a dictionary, etc” and next thing they’re out of business.
From where I am now, I’m not religious about data management. If I believe that the ideas can help a business, I’m all for it – but if I were a CEO, I’d want a pretty good idea of the approach being proposed and the sort of people doing it before automatically endorsing a DM function.
Wilshire: All right, let’s hone in on the topic of data modeling then. At the Enterprise Data Forum coming up this fall, you’re talking about the “Big Issues” in Data Modeling. You’re covering a variety of good stuff, so let me ask you about just a couple of the topics. How about “The role of data modeling in a packaged software world.” Give us your take on that one.
Simsion: The glib answer is going to be “data modeling is still vital, yada yada yada”, but the reality is that there’s less data modeling being done, simply because of the steady move from in-house development to buying packaged software, including ERP. There’s obviously a role for data modeling with the package developers – and that has to be very good data modeling indeed. And in package evaluation – I’ve seen data modeling used very effectively, and equally, I’ve seen data modeling used inappropriately and good packages rejected because the underlying data model doesn’t conform to some other modeler’s idea of the best design. One of the greatest areas of need for data modelers is user-developed applications. In some cases these guys don’t know the first thing about data structure, and the single most useful input they can get is advice on a good model – real high-leverage stuff.
Wilshire: You’re also talking about “Modeling and Business Rules.” What pearls do you have on that one?
Simsion: My key position is that business rules are not set in stone – they’re open to challenge – vigorous challenge – particularly in an organization in which information systems are an intrinsic, critical part of the business. I think there’s been too much of an attitude that the business is “out there” rather than that we as modelers are helping to construct a new business.
Wilshire: OK, and what about the role of UML? What’s your advice to data modelers on when and where (and whether?) to use UML?
Simsion: Let me start with a huge qualification: I’ve never used UML in practice – I mean, I know the theory, but unlike many people out there I haven’t got my hands really dirty with it, and therefore I don’t have the experience that gives you a deep understanding. It came a bit late for me. Over the years I think there’s been too much focus on languages for data modeling at the expense of more important issues – and to relatively little effect. Until UML came along, the conventions we were using hadn’t changed much since the Bachman diagrams that came in the late 1960s. It wasn’t till the upsurge in OO development that there was a real need for something different, and I think you have to see UML in that context.
That said, it seems to me that UML’s strength is as a pretty comprehensive specification language – and there’s certainly a need for that – at least some of the time! And of course it’s gained a substantial level of acceptance. What you need to ask is whether the subset you choose to use is aligned with your development and implementation environment – in other words, will you be able to implement the things you specify? See, an attributed ER model is fine as a conventional relational database spec – and if that was my implementation environment, I’d be dubious about specifying class hierarchies etc. On the other hand, in an OO environment…
There’s also a need for communication in the other direction – with the business – and I think some aspects of UML are much better than others here.
Finally, you do need to ask some broader questions about the methodology you’re going to be using – you don’t make the choice of UML in isolation from that. The recent interest in Agile programming has reignited the debate about waterfall vs iterative methods, and – related to that – the “light or heavy” documentation issue. Mountains of UML diagrams may not be the best answer…
Wilshire: Consider a couple of hypothetical consulting assignments please. In the first, suppose a company calls you in and says “we only have the resources to do a few things well in our DM organization, what should they be?”
Simsion: Well, Tony, you have to look behind that question. It’s probably not that the company lacks resources – or money to buy them – in absolute terms; what they’re saying is that they only want to risk so much on data management at the moment. If they were totally confident, they’d spend what they needed to spend. And I’d really want to explore that with them before taking their budget as a “given”.
OK, given all that, I’d sit down these guys with the money, and talk to them in specific terms – with war stories – about what sort of benefits data management can bring. So I’d talk about data quality, metadata, data integration, data access etc all as separate topics. And as I went, I’d ask them to tell me about real examples of these sorts of things in their business. Then we’d move – outside the meeting – to quantify the benefits and estimate the costs of achieving them. So I’d have their sign-up to the benefits – and the data manager’s sign-up to the costs. So, there’s no generic answer, and I think that’s important. I’d certainly not be saying “first, you need a model”.
Wilshire: Assume a different case, where the basic data management functions are humming along and the organization decides it wants to graduate to becoming a “world-class data organization”. Where would you start on such an assignment? Assuming they give you reasonable (but not unlimited) funding, what priorities would you set for them?
Simsion: I’d be asking why they were talking to me instead of the consultancy with the line about “world class data organizations”… I mean really, this is not how general managers talk unless someone’s got in their ear. I guess my first question would be “why?” Data is my business, and it’s a means to an end. I want to know what business benefits they’re trying to achieve – and then help them put a data strategy in place that will get them there as effectively – cheaply – as possible, doing whatever it takes. If I just said “hey, all my Christmases have come at once; let’s roll”, I’d expect to come unstuck down the track when the “why” questions inevitably got asked.
Wilshire: So, to revisit the earlier question, if traditional data management isn’t working, what IS working? (i.e. what are the high-payoff areas for data people to focus on?)
Simsion: OK, OK, you really want me to pick some generically high priority areas. Historically the thing that’s been most successful is central data modeling consultancy. Making sure that all databases are constructed on a sound data model (ideally a consistent one, but let’s not go there yet) is one of the best and simplest things you can do. I’d be looking at data quality in a high-impact area – names and addresses are a classic one. Management information – review what they’re getting, what they really need, and don’t assume you need a three-year data warehouse to solve it. If you’re customer-oriented and haven’t got integrated customer data – well that goes to the top of the list.
If you’re an organization with a branch structure – multiple business units doing essentially the same thing, like branches of a bank, or hospitals within a public health service – look at information flows between the branches and head office. There’s often an opportunity to rationalize these with a big impact because of the number of branches.
Wilshire: You pretty much said earlier that you thought metadata repository and metadata management seldom delivered on it's promise. So what would you discuss in terms of "metadata" with your client.
Simsion: My strong guess is that we’ve spent more money on metadata than we’ve got back in returns. It’s a central icon of traditional data administration, but it often seems hard to find a balance between doing enough to provide real value, but not so much that it’s prohibitively expensive and wasteful. So, I'd ask the client whether they had problems finding data, understanding its meaning etc. But then I'd ask them to be more specific - where EXACTLY was the pain. So it might turn out that they had specific problems knowing what was available in the HR and payroll systems, how to interpret it, what the privacy rules were etc. So we might commence a project to document that specific metadata - as distinct from a project to document everything!
Wilshire: Do you have any “career” advice to data practitioners today? What new subject areas would you suggest DM’ers get involved in for their own future job security, growth, opportunity etc.?
Simsion: Don’t assume that there will be ongoing jobs in central data management teams. I think this becomes a very personal thing – you need to do a personal SWOT analysis – maybe some of the other sort of analysis too! – and decide how you can meet your personal needs by doing something that people are prepared to pay you for (assuming that money’s a driver…). I think data management people fall into two groups: those who are focused on the goals of data management – so for example, they’d like to see an organization increase its profitability thanks to better integrated customer data – and those who are focused on the tools and techniques – so they enjoy modeling for example. The first group, which I think is the smaller one, may find a move into more general management with a data flavour to make sense, whereas the second group need to ask – where is this skill needed?
Wilshire: Let me ask you about some of the “hot button” issues in IT today and get your gut reaction, opinion, whatever. Short questions, short responses...OK?
Simsion: OK, fire away.
Wilshire: XML
Simsion: Great innovation, but it doesn’t take away the need to understand data representation and intrinsic structure.
Wilshire: Web services
Simsion: Big issue is the management of the non-record-based data that is now so much a part of business and life. Data managers haven’t done enough here.
Wilshire: Business process re-engineering
Simsion: One of the fads that actually had some substance. You should never be building an application without first reviewing the business process. And the BPR toolkit is a valuable part of that. Key problem is that end-to-end processes in an organization are seldom “owned” by anyone much below the CEO, so sponsorship for anything except the highest value core business processes is problematic.
Wilshire: Dimensional modeling
Simsion: Highly appropriate – and well established – for data marts. The warehouse – feeding multiple marts – is a different matter.
Wilshire: “Agile” methods/Extreme programming
Simsion: Pretty much the way highly talented people in small teams have always done it (which is not to detract from the value of documenting this and thinking about it). The hard thing’s always been finding ways of working that don’t require everyone to be highly talented. I’ve always been concerned about the impact of mucking around with data structures too much – I like to get them reasonably settled early (which is not impossible) and let the iteration do its thing on the processes.
Wilshire: Universal models
Simsion: Word association: David Hay, Len Silverston – who’ve done a lot to share common patterns through their books in particular. And of course you have industry-specific ones. The need for this was something I talked about for years – and was once controversial – so it’s great to see it happening. I wrote a foreword for Len’s second edition, and in that I said that this publication of patterns was an important part of the coming of age of the profession.
Wilshire: Knowledge management
Simsion: Important, but I never want to see that Data-Information-Knowledge-Wisdom pyramid again. I think the connection with data management is tenuous, and it shouldn’t be an automatic pick-up for data management teams. I see it coming in where BPR stops – i.e. in knowledge-based rather than “factory” processes.
Wilshire: E-business
Simsion: What can I add on a topic that’s been done to death. I did a whole lot of work looking at best practice in the area. Basically, it comes down to having a business model that works, and electronic communication may be a key part of that. And at the consumer level, uptake is a big issue – exponential growth inevitably levels out.
Wilshire: Software quality
Simsion: The quality movement has done some good things – but when I see Software Quality in capitals I think of unimaginative fascists who believe you can proceduralize creativity. The quality ideas are great for production – but I have real problems when they get adapted to design activities – one of which is data modeling.
Wilshire: Data standards
Simsion: Vital – within and across industries – and promulgating them is a really good and helpful role for a data management team.
Wilshire: Zachman Framework
Simsion: An excellent framework for classification of the key concepts and deliverables that we have to work with. But its strength is in description rather than prescription. Has suffered from being extended beyond its intended purpose by some.
Wilshire: Outsourcing
Simsion: It makes sense to outsource something you can prescribe and measure – much harder to outsource on a “partnership” basis in which both parties need to work for the common good.
Wilshire: In the dozen or so years I’ve now been involved with the DM community, the most consistently voiced problem is “how do we get senior management commitment?” (or support, involvement, or some similar version of same). What’s your reaction, and answer, to that question?
Simsion: First it helps to recognize that everyone with a brilliant idea in their area of expertise wants senior management support. Sometimes they get it – supposedly – but it turns out to be uninformed or flaky support that doesn’t stand up when it’s needed – typically to win a fight with someone else. If you start by listening to what senior managers want, and offering focused support for that, instead of telling them what they need – well that’s a good start. Short to medium term deliverables that management values will build a track record that can turn early support into something that lasts.
Wilshire: At the start of our conversation I mentioned that you’re known for your pragmatic point of view. Can you give us some larger context for where this comes from? In the US we mostly only know you as a data modeling guy, yet your consulting experience worldwide has been far more strategic.
Simsion: I started out as a computer operator, then as a programmer, and did quite a bit of time as a database administrator, so in that sense data modeling has been just one stage in my journey backwards through the systems life-cycle. When I started the consultancy, in 1982, through to the early 90s, I did a lot of modeling, but as the business grew, I had to step back from the hands-on work, and take more of a management perspective – and get more in touch with what clients wanted from modeling.
At the same time my own professional interests were changing, and I got involved in data management, then information systems planning, business process design, and by the time I sold the consultancy most of my work was in straight business planning. So in terms of a “big picture” view, because of the DBA background I’ve always seen models first and foremost as database specifications – which is a pretty pragmatic sort of view of them. Managing data modeling has taught me a lot about what clients value – and doing only what needs to be done to meet their needs. And the business-oriented stuff has taught me a lot about the real value of modeling and management to the business.
Wilshire: You’re also known for your sense of humor and your funny accent. Is this why Australians have a world-class reputation as good data modelers?
Simsion: I’m slowly learning that it’s “Day-tah” not “Dah-tah”, but it still comes out as “Dy-tah”. I’m not sure about the relationship between sense of humor and data modeling, except that a sense of humor is usually a good sign of not taking things too seriously – and some data modelers do take themselves too seriously. That’s not to say that data modeling isn’t important, but it’s basically a technique for designing databases (OK, and planning data strategies too) – which are important, but not that important. I had someone come on an introductory data modeling course once, and on the second day she’s telling us that she stayed up all night with her partner drawing on the ‘fridge with a whiteboard marker using data modeling concepts to understand their relationship.
But back to the question about Australians. Australia being a relatively small (in population) and geographically isolated country, we tend to look beyond our own shores for ideas and don’t consider we’ve “made it” until we’ve gained acceptance internationally – traditionally in the UK and US, and more recently in Asia. So in the early days of data modeling, we were looking to both the US and UK. Very broadly, the US was doing great things with the relational model, and UK practitioners were stronger on the semantic side. I think we were well placed to draw on both schools. And we had a few really strong players – Clive Finkelstein with the information engineering approach and Nijssen who was based in Australia pushing fact-based modeling – so there was a lot going on… There’s a strong interest in the universities in data modeling – I’ve seen two PhD theses on the subject in the last few months…
Wilshire: Last question then. You sold your business and recently made the move to academia. What have you discovered since you got there? Is it fun to be rich and only obliged to work 10 hours a week?
Simsion: Well, when I sold the consultancy, there was this non-compete clause in the contract… Actually that was part of it, but it was a good excuse to take a bit of time out to reflect. My career’s been pretty much about doing things and then trying to make sense of them and learn something for next time, and sometimes that means taking a bigger chunk of time out. I enjoy the lecturing – you learn more from what you say than what you hear!
I’ve discovered that in information systems, academia often suffers from lack of relevance and industry from lack of rigor. There’s a big difference between proper research-based findings and guru-centered knowledge, but often the latter is all we have. In short, as we all know, we need to work more closely together – my short time in academia has just reinforced that.
Join
us for the
Wilshire Meta-Data Conference
and DAMA International Symposium
May 2-6, 2004 • Century Plaza
Hotel • Los Angeles, California USA
The
World's Largest Vendor-Neutral Data Management Conference
The 16th annual DAMA International Symposium and 8th annual Wilshire Meta-Data Conference will be held May 2-6, 2004 at the Century Plaza Hotel in Los Angeles, a beautiful venue adjacent to Beverly Hills. Hear 40 case studies outlining strategies of companies that have implemented successful data management projects. There will be more than 120 speakers in all, covering meta data, enterprise architecture, data and process modeling, unstructured data, business rules, data integration, XML, business intelligence, data warehousing, information stewardship, and more. Keynote Speaker Chris Date. Click here for details.
Click here to sign up to receive future editions.
For sponsorship information, contact Rick Froton at 603-305-0660.
©2002 Wilshire Conferences, Inc. May be quoted with full attribution.