Long before Scott Davis made the self-service ETL tool he calls Lyza, he tried to find out how analysts really work. He remembers in particular the woman in a focus group who said, “I want to stay close to the data.”

He didn’t understand at first. The data was right in front of her, neatly summarized. But she meant all of the data, every little bit of it. She wanted to snap open a zillion-row-long window that she could scroll down to see the figures flip by. (Yes, you can; I saw it yesterday.) She wouldn’t try to read them, she’d only see their shapes. She could say, for example, “Hmm, I see that just two thirds are under 1000.” Davis calls that visualization with browse—as legitimate a use of “visualization” as any I’ve heard of.

He also thought about how people use Excel. In fact, it helps explain’s Excel’s popularity. They have the data, and they have the formulas, and you can reveal either one. If a number shows up that doesn’t look right—say it’s six figures instead of five—you just look at the formula. You say, “Oh, that’s the annual figure. I forgot to divide by twelve.”

Something similar goes on at all levels of analysis: a rapid back and forth from question to answer, back to a rephrased question, and back to an adjusted answer.

Forget the flow charts. Forget the “data train,” a metaphor I admit to having used. Analysis is more like what my labrador does when she knows there’s something good nearby. She sniffs in what looks like a random pattern until you realize she’s narrowing the range.

What drives analysts crazy about working with IT, he says, is that the data’s taken away. The conversation goes like this: the IT guy asks what the analyst wants; the analytst describes her best guess; the IT guy goes away and does it. But that may not be what the analyst really needed, and the anallyst may not realize it until the first data’s tried and proves inadequate or suggests yet another path.

I can relate, because it’s like writing. I do a lot of scribbling and writing over, and I don’t have time to explain it. If I had to tell a typist what to write, I’d write much less.

Visualize the bumper stickers: “free the analysts” but also “free IT.”

Now, Larissa T. Moss has her doubts. Perhaps she’ll sit for a demo. I’d like to hear what she says.

At 8 o’clock Monday morning, a few hundred attendees at TDWI conference in Chicago will hear the organization’s former education director Dave Wells give his keynote, “People First: Creating a Business Intelligence Culture.” He’ll say something startling: there’s much more to BI than data.

BI is no more about data alone, he has said, than Van Gogh’s “Starry Night” is only about brush strokes. For those of you whose brains itch with a sense that BI is ultimately a social practice, this is for you.

Among other things, he will propose a new definition of business intelligence and talk about cultivating a BI culture, “not exactly a science…something more like an art form.”

I can’t be there, but this story is just beginning.

Is the data clean enough to give us meaning? If you are one of those who insist that the war on dirty data must be won first, forgive me. I’m tired of that conversation. It sounds like backyard mechanics comparing fuel injectors when all that really matters is the commute.

I talked recently to TDWI’s just-departed education director Dave Wells. He’s been thinking, and he says, “Tweaking the details probably doesn’t make any real, significant contribution to the overall value of the insights I can gain by understanding the patterns inherent in data.”

Is it really, really going to make any difference if you can get your data from 96 percent to 98 percent pure? Will the meaning really change?

My old boss in market research had a rule of thumb: with a good sample, results usually stabilize after about 25 percent of the data have been counted.

Are we there yet? Wells says the more interesting problem these days how to soak the meaning out of what you do have. Let those who scrub keep scrubbing while the rest of us move on.

To get to the next level in BI, he says we’ll have to understand better how people think. That could involve systems thinking, how one part of an operation affects a different one. Other areas he’s looking into are design thinking and critical thinking.

Notable quotes

• One of the real looming questions I think in BI as it becomes mainstream is what do you do when analytics contradict conventional wisdom. Then it becomes a political and religious debate.
• The flaw in our current approach to analyzing things is best described by an analogy that says, “you cannot understand the impact of Van Gogh’s Starry Night by categorizing the brush strokes.”
• Maybe the value of real high impact metrics, for instance, is not in the measuring of things, but in the measuring of relationships.
• There will always be something technological to fix. Let’s go back to Starry Night. Do I need to fix that brush stroke? Or do I need to stand back and look at the big picture?

One person I’ve talked to about his ideas thinks Dave is “full of crap.” He says, “If you talk to the people who deal with data, you hear that the problem is nowhere near solved.” OK, there’s still work to be done. But rejecting Dave’s reasoning is like saying that because car design is still not perfect we can’t yet talk about driving from San Francisco to Lake Tahoe.

The full interview will be at the May 7 issue of BI This Week.

Links he suggests, with his notes:

• Systems thinking
• Design thinking: The weblog Functioning Form’s “Defining Design Thinking” post gives a good starting point, he says, with links to articles and papers. He writes, “Design thinking is a less mature discipline [than systems thinking], which makes it less robust but more interesting to explore.”
• Critical thinking: “The critical thinking community is more focused on education than on business today,” he writes, “so much of the interesting stuff there is in discovering how it does apply in business. I have no doubt that it does apply because much of critical thinking is simply another perspective on decision theory.” See the Strategies for Success page and the Foundation for Critical Thinking site.

After about three quarters of an oatmeal stout, my old friend Sam the BI developer wondered aloud, “What is the ideal dashboard?”

There was no need to call Steve Few or Edward Tufte. I had the answer right away. (I had been sipping an oatmeal stout myself.)

The ideal dashboard, I said, is like the human face. You can read the big, obvious signs at a glance: whether the person is smiling, frowning, snarling, or foaming. You know right away whether things are OK.

If things are not OK, you know you’d better ask. That is, you drill down.

If you can’t be direct, you might go on a hunch. “Hmm, looks confused and maybe a little resentful.” Or, in business, you might say, “Hmm, old Charley’s department’s in a muddle about that new hire.”

Even with today’s dashboards, you might have the same insights. But with the face-based display, your intuition would pick up where your rational brain left off—thanks to finer, more subtle, perhaps even ambiguous signals on the first level.

In this model, the dashboard as a metaphor gives way to the face as the guiding light for designers.

Now, hold your applause. There are problems.

For one thing, a face-based display would require a ton more bandwith. Even after designers figured out how to build such a display, engineers would have to pack a lot more data onto the screen.

The bigger problem, I think, is that the business world doesn’t like such soft, gooey metaphors. It prefers metaphors from sports, war and machines.

But once accepted and once the technical part works out, the face metaphor would fit better than the dashboard. After all, what is a business? Is it a machine? Or is it a group of people—with all the flesh and blood and goo and subtlety and amibiguity that that implies?

Late last month on the Juice Analytics weblog, they were talking about Stephen Few’s new concept, the “faceted analytics display.“

I like the idea, and I’m sure FADs are important. I just hate to see Few resort to a new term because inept designers have spoiled “dashboard.”

Dashboard is a valuable metaphor and should be defended. I’m afraid FAD will be forgotten.

Isn’t a FAD just a dashboard with extra features? When they added tachometers to auto dashboards, did dashboards become something else? If you add new software or a new peripheral to your computer, isn’t it still a computer?

Perhaps we could call it a “faceted” dashboard—but still a dashboard. That name is still stronger than FAD.

Using “faceted” might force a modifier onto dashboards with just one face. How about “dumb,” as in dumb dashboard? Then we’d call the really bad dashboards just “dumber.”