Setting Expectations
A successful data modeling engagement needs to be based on a set of expectations that are A successful data modeling engagement needs to be based on a set of expectations that are understood and agreed upon by all stakeholders. Easy to say, but in the haste to get moving, we understood and agreed upon by all stakeholders. Easy to say, but in the haste to get moving, we can forget the lessons from our programmer brethren (and indeed the advice we would give on can forget the lessons from our programmer brethren (and indeed the advice we would give on the database side) and fail to put a proper specification of work in place. We will look at some the database side) and fail to put a proper specification of work in place. We will look at some key questions to ask — and
key questions to ask — and who to ask — in establishing such a who to ask — in establishing such a specification, but before wespecification, but before we bury ourselves in the detail, we need to establish the big picture.
bury ourselves in the detail, we need to establish the big picture.
Understanding
UnderstandingContext Context
It's sometimes called "asking the next-higher question." "Why are we doing this? What does the It's sometimes called "asking the next-higher question." "Why are we doing this? What does the client / project manager / sponsor want
client / project manager / sponsor want it for? And, consequently, exactly whait for? And, consequently, exactly what sort oft sort of deliverable is needed to serve that
deliverable is needed to serve that purpose?" I would suggest that failure to upurpose?" I would suggest that failure to understand andnderstand and address this
address thiscontext context in which work is being p in which work is being performed is the single greatest cause of problems inerformed is the single greatest cause of problems in data modeling assignments.
data modeling assignments.
A simple example: a team was assembled to
A simple example: a team was assembled to undertake a classic data mining project, undertake a classic data mining project, looking atlooking at an operational database for "interesting correlations". The high
an operational database for "interesting correlations". The highly-qualified team included a dataly-qualified team included a data modeler to reverse engineer the database, as well as a statistician to interpret results. After some modeler to reverse engineer the database, as well as a statistician to interpret results. After some preliminary work, they produced a steady stream of "interesting correlations", but the client preliminary work, they produced a steady stream of "interesting correlations", but the client
remained consistently unimpressed. At this point, I became invo
remained consistently unimpressed. At this point, I became involved. My first question to thelved. My first question to the
"difficult" client was, "why are you doing this?" The response: "Because our organization has
"difficult" client was, "why are you doing this?" The response: "Because our organization has been criticized in a public report, and we wanted to know whether any of the claims in the report been criticized in a public report, and we wanted to know whether any of the claims in the report could be refuted by our operational data." It took no more than an hour with a copy of the report could be refuted by our operational data." It took no more than an hour with a copy of the report and a highlighting pen
and a highlighting pen to identify the statements that might be corroborated or to identify the statements that might be corroborated or refuted by therefuted by the data, and only a day for the data mining team to produce answers — to the satisfaction of the data, and only a day for the data mining team to produce answers — to the satisfaction of the client.
client.
In the review that followed, the mining team was quick to point out that the client had not In the review that followed, the mining team was quick to point out that the client had not explained the reason for the work, that they'd never heard of the report, and that the work explained the reason for the work, that they'd never heard of the report, and that the work
actually required wasn't data mining anyway! I also suspect that having been given a brief to do actually required wasn't data mining anyway! I also suspect that having been given a brief to do exactly the sort of work they
exactly the sort of work they found professionally satisfying, they were not inclined to ask toofound professionally satisfying, they were not inclined to ask too many questions.
many questions.
In data modeling at the project level, an understanding of context is essential in establishing In data modeling at the project level, an understanding of context is essential in establishing quality, cost and timing requirements. The two
quality, cost and timing requirements. The two most common complaints about data most common complaints about data modelersmodelers are that they are impractical and that they take too long. I'm well aware of the defenses to these are that they are impractical and that they take too long. I'm well aware of the defenses to these accusations, having used them many times myself, but we have to acknowledge the perceptions.
accusations, having used them many times myself, but we have to acknowledge the perceptions.
And they are
And they are context context issues. Behind the complaints is the suspicion that some data modelers are issues. Behind the complaints is the suspicion that some data modelers are more concerned with the intrinsic quality of the model than with its contribution to the success of more concerned with the intrinsic quality of the model than with its contribution to the success of the project: "the operation was a success but the patient died".
the project: "the operation was a success but the patient died".
Like other professionals, data modelers want to produce
Like other professionals, data modelers want to produce high-quality results — but need tohigh-quality results — but need to remember that quality is
remember that quality is fitness for fitness for purposepurpose rather than an absolute standard. We are well aware rather than an absolute standard. We are well aware
indeed are sometimes not willing to co
indeed are sometimes not willing to compromise) in order to meet overall project goals. If thempromise) in order to meet overall project goals. If the project is time-critical, then timely delivery may be more critical than intr
project is time-critical, then timely delivery may be more critical than intrinsic quality factors.insic quality factors.
We should also remember that data modeling has no natural end. As with any design product, a We should also remember that data modeling has no natural end. As with any design product, a data model can always be
data model can always be improved — indefinitely. It's easy to be caught improved — indefinitely. It's easy to be caught in the perfection trap:in the perfection trap:
pursuing the perfect "correct" model when all we are doing is refining a design beyond the point pursuing the perfect "correct" model when all we are doing is refining a design beyond the point
of worthwhile returns.
of worthwhile returns.
At least in applications development, we have a broad idea of the purpose of the data model. In At least in applications development, we have a broad idea of the purpose of the data model. In strategic work, such as development of an enterprise data model, we cannot take even this for strategic work, such as development of an enterprise data model, we cannot take even this for granted. Time and again I have seen enterprise models developed without a precise
granted. Time and again I have seen enterprise models developed without a precise understanding of how they will be
understanding of how they will be used. The result is frequently a model with unused. The result is frequently a model with unnecessary detail,necessary detail, or perhaps unfamiliar structures that do not sit easily with an organization's strategy of buying or perhaps unfamiliar structures that do not sit easily with an organization's strategy of buying off-the-shelf software. Some advocates of enterprise models will argue that they need to be off-the-shelf software. Some advocates of enterprise models will argue that they need to be developed in "excruciating detail" - fine, as long as we have established exactly how and by developed in "excruciating detail" - fine, as long as we have established exactly how and by whom that expensive detail is going to be used.
whom that expensive detail is going to be used.
Remember — there is no such thing as a good data model independent of its purpose. Data Remember — there is no such thing as a good data model independent of its purpose. Data
modeling takes place in the context of a larger project, and will ultimately be judged by how well modeling takes place in the context of a larger project, and will ultimately be judged by how well it contributes to that project's success.
it contributes to that project's success.
Identifying
IdentifyingthetheStakeholdersStakeholders
In the next subsection, we will look
In the next subsection, we will look at some techniques for establishing expectations in moreat some techniques for establishing expectations in more detail. But whose expectations? A common mistake in consulting is to identify a person or team detail. But whose expectations? A common mistake in consulting is to identify a person or team as "the client" only to find at the end of the assignment that other stakeholders' needs have not as "the client" only to find at the end of the assignment that other stakeholders' needs have not been met. The following list of potential stakeholders and some of their likely expectations is a been met. The following list of potential stakeholders and some of their likely expectations is a
useful starting point.
useful starting point.
•
• TheThe database designer database designer is a key "hands on" user of the final model. Many times I have is a key "hands on" user of the final model. Many times I have seen data models developed without any serious preliminary discussion with the DBA.
seen data models developed without any serious preliminary discussion with the DBA.
The model is handed over on the due date — and then the arguments start. The model is The model is handed over on the due date — and then the arguments start. The model is too generic, too hard to understand, won't perform. Setting — and arguing — the
too generic, too hard to understand, won't perform. Setting — and arguing — the
evaluation criteria at the outset, before substantial work has been done, can substantially evaluation criteria at the outset, before substantial work has been done, can substantially reduce this risk.
reduce this risk.
•
• TheThe process d process developersevelopers, from business analysts and process designers to programmers, from business analysts and process designers to programmers and testers, will need to be able to understand and work with the model. As in the case of and testers, will need to be able to understand and work with the model. As in the case of the DBA, we need to understand what they expect of the model.
the DBA, we need to understand what they expect of the model.
•
• More broadly, theMore broadly, the usersusers of whatever we are producing — be it an enterprise model, a of whatever we are producing — be it an enterprise model, a review, a reverse-engineered database. We need to identify, at least at a group level, those review, a reverse-engineered database. We need to identify, at least at a group level, those who will be affected by
who will be affected by any change that will result from our work.any change that will result from our work.
•
• TheThe project man project manager ager and andsponsorssponsors, who have authorized the funding of the total package, who have authorized the funding of the total package of work or the data modeling component, will have their own expectations. Our basic of work or the data modeling component, will have their own expectations. Our basic questions to them are: Why have you asked for this? How do you plan to use it? Budgets questions to them are: Why have you asked for this? How do you plan to use it? Budgets and timelines may have some flexibility — or not. And the data modeling component and timelines may have some flexibility — or not. And the data modeling component may be seen as worthy of additional effort to get the best result — or it may not.
may be seen as worthy of additional effort to get the best result — or it may not.
•
• Subject matter expertsSubject matter experts who will be concerned about the time required of them and their who will be concerned about the time required of them and their staff.
staff.
•
• Technical stakeholdersTechnical stakeholders are people outside the project who require adherence to standards are people outside the project who require adherence to standards — perhaps use of particular tools or platforms. External consultants
— perhaps use of particular tools or platforms. External consultants are likely to have toare likely to have to meet the expectations of a purchasing department, as embodied in a legal contract.
meet the expectations of a purchasing department, as embodied in a legal contract.
•
• TheThe consultancy manager consultancy manager is typically your line manager — perhaps the head of a central is typically your line manager — perhaps the head of a central data modeling or enterprise architecture group. They may have their own agenda and data modeling or enterprise architecture group. They may have their own agenda and expectations — typically, that work conforms to
expectations — typically, that work conforms to enterprise standards and thatenterprise standards and that documentation is added to a central repository. If you're working for an external documentation is added to a central repository. If you're working for an external
consultancy, there is likely to be an agenda of building a reputation and relationships that consultancy, there is likely to be an agenda of building a reputation and relationships that will lead to further work.
will lead to further work.
There is nothing novel in the above analysis of stakeholders and needs; the challenge is to act on There is nothing novel in the above analysis of stakeholders and needs; the challenge is to act on it and actually establish these and
it and actually establish these and any other stakeholders' needs at the oany other stakeholders' needs at the outset of the exercise. Thisutset of the exercise. This is also a good time to think about the level of involvement that these players should have in the is also a good time to think about the level of involvement that these players should have in the work. One of the great lessons of change management is that people will work very hard to work. One of the great lessons of change management is that people will work very hard to implement something that they have helped to design — and conversely will be resistant to ideas implement something that they have helped to design — and conversely will be resistant to ideas and designs imposed upon them. Perhaps it's worth involving that "difficult" DBA in the
and designs imposed upon them. Perhaps it's worth involving that "difficult" DBA in the modeling effort — right from the start.
modeling effort — right from the start.
Asking
AskingKeyKeyQuestionsQuestions
At the outset of a data modeling assignment, we need to establish the expectations of all of the At the outset of a data modeling assignment, we need to establish the expectations of all of the stakeholders (clients). We recognize that these may change as work proceeds, and will commit to stakeholders (clients). We recognize that these may change as work proceeds, and will commit to monitoring and documenting such changes. Our motto should be "no surprises".
monitoring and documenting such changes. Our motto should be "no surprises".
Here are a few questions that can
Here are a few questions that can help clarify expectations, both in terms of deliverables help clarify expectations, both in terms of deliverables andand method of working.
method of working.
•
• What does the final product look like?What does the final product look like? The client should understand The client should understandin concrete termsin concrete terms what they will be ge
what they will be getting. Words liketting. Words likearchitecturearchitecture,, modelmodel,, strategystrategy, and even, and evenreport report are are ambiguous, and we can have no confidence that a client who signs up for one of these has ambiguous, and we can have no confidence that a client who signs up for one of these has the same thing in mind as
the same thing in mind as us. If we can't show them the us. If we can't show them the deliverable from a previousdeliverable from a previous exercise as a sample, we will need to generate a sample in an early project phase. Until exercise as a sample, we will need to generate a sample in an early project phase. Until the client has seen and agreed to a concrete example, we cannot claim to understand their the client has seen and agreed to a concrete example, we cannot claim to understand their expectations. In the case of a report, it makes sense to outline section headings and
expectations. In the case of a report, it makes sense to outline section headings and approximate number of pages as indications of depth and effort.
approximate number of pages as indications of depth and effort.
•
• What does "success" look like to the stakeholder?What does "success" look like to the stakeholder? Sometimes the answer, to our Sometimes the answer, to our surprise, is very different from our understanding, although the
surprise, is very different from our understanding, although the stakeholder is unlikely tostakeholder is unlikely to tell us directly. You might be brought in to "improve" or "coach" a team, or review their tell us directly. You might be brought in to "improve" or "coach" a team, or review their performance, sponsored by a more senior manager. The team or individuals may take a performance, sponsored by a more senior manager. The team or individuals may take a
surprise, is very different from our understanding, although the stakeholder is unlikely tostakeholder is unlikely to tell us directly. You might be brought in to "improve" or "coach" a team, or review their tell us directly. You might be brought in to "improve" or "coach" a team, or review their performance, sponsored by a more senior manager. The team or individuals may take a performance, sponsored by a more senior manager. The team or individuals may take a