Control Yourselves, Big Data Dudes

Posted on Feb 15, 2012 | 4 Comments

According to a recent piece in the New York Times, we live in an era of “Big Data.” It’s a time of a “data flood” and “data-driven discovery and decision-making.” Some even call it a revolution.

“A report by the forum, ‘Big Data, Big Impact,’ declared data a new class of economic asset, like currency or gold,” the Times noted.

Interesting stuff, to be sure. But we were also taken by the published comment of a Times reader, Danny P. of Warrensburg, Mo. As he wrote: “‘Big Data’ that the article refers to doesn’t have any controls in place, and adequate controls are what allow quantitative study to determine cause-and-effect relationships with any real degree of accuracy. Without those controls, big data becomes nothing more than the bag of letters in Scrabble.”

This was a very Druckerian point to make. For as Peter Drucker saw it, measurements, information systems, feedback loops and the like–what he called, in general, “controls”—need to be carefully constructed if they’re to be effective. In fact, as Drucker explained in Management: Tasks, Responsibilities, Practices, controls should meet seven specific criteria:

1.     They must be economical.  “More controls does not give better control. All it does is create confusion. … The capacity of the computer to spew out huge masses of data does not make for better controls.”

2.     They must be meaningful. As we’ve discussed before, “the events to be measured must be significant either in themselves . . . or as symptoms of at least potentially significant developments.”

3.     They must be appropriate to the character and nature of the phenomenon measured.  This was supremely important. The controls must “bring out clearly what the real structure of events is.” Getting 50 employee complaints a month might seem better than getting 100 employee complaints a month. But if the 100 complaints are dispersed throughout the company and the 50 complaints are targeted against one abusive supervisor, then just looking at numbers can be fatally misleading.

4.     They must be congruent with the events measured. “A measurement does not become more accurate by being worked out to the sixth decimal when the phenomenon is only capable of being verified within a range of 50 to 70%.” (Economists, take note of that one.)

5.     They must be timely. “The time dimension of controls has to correspond to the time span of the event measured.” Don’t do a daily temperature update to gauge global warming.

6.     They must be simple. As we’ve explored in another context, “complicated controls do not work. They confuse.”

7.     They must be operational. Someone has to be able to use the information to do something. “It should never just say, ‘Here is something you might find interesting.’”

Does all the data coursing through your organization meet these seven specifications?


  1. Sergio
    February 15, 2012

    I understand “controls” (as referred to in this article) to embody the essence knowledge work, and the properties of effective controls are consistent with Drucker’s definition of knowledge where he said “data endowed with relevance and purpose. Converting data into information thus requires knowledge.”

    What makes applying this knowledge in the Big Data space challenging is that the knowledge worker (i.e. control) first needs to understand the domain (see Cynefin Framework) most applicable for the desired causality. If the domain is simple, for example, then the challenge can be relegated to that of simply scaling technology to meet the levels of Big Data (e.g. analyzing and storing TeraBytes). If the domain is “complicated”, or “complex”, the challenge becomes the fact that the quantitative data alone may not suffice, and instead a stream of qualitative information is needed. I’m certainly not an expert in Machine Learning or Artificial Intelligence, but implementing “controls” for qualitative information is a tough problem, but one that cannot be ignored.

    The NY Times article on Big Data had a comment by Professor Brynjolfsson “decisions will increasingly be based on data and analysis rather than on experience and intuition.” I can support this statement as long as we don’t become disillusioned to think the knowledge worker can be replaced by an automated controls.

    Before Big Data, I found it hard to spot these 7 properties in organizations I came across. With Big Data now a reality, things just got a lot more complicated.

  2. Daniel Pacheco
    February 18, 2012

    Convert Big data into Small data (chunk down) and then make decisions on root causes of the small chunk data. Easier said than done that’s why consultants like me can make a living. The operational guy has data flood the consultant converts this data flood into small chunk data – controllable buckets of water. How does the consultant do it ? He uses his intuition. What if the operational guy uses his intuition. Then the operational guy moves into the strategic mode and does not need the consultant. The consultant will have no work and will have to become the operational guy.

  3. steve wright
    February 21, 2012

    Seems we have a lot of confusion on the definition of “control”. The Drucker use of the term:

    Drucker (in Management: Tasks, Responsibilities, Practices, for example) points out that controls are concerned with measurement and information. In contrast, control is concerned with direction. Where controls focus on past facts – analysing what was and what is – control is founded on future expectations of what ought to be. In short, controls are a means to the end of better control. They are not – nor should they become – an end in themselves.

    And the Randomized Control Trial use of the term:

    Scientific control allows for comparisons of concepts. It is a part of the scientific method. Scientific control is often used in discussion of natural experiments. For instance, during drug testing, scientists will try to control two groups to keep them as identical and normal as possible, then allow one group to try the drug. Another example might be testing plant fertilizer by giving it to only half the plants in a garden (the plants that receive no fertilizer are the control group, because they are kept normal).

    The comment by the Danny P. in the NYT article seems to be referring to the latter definition of “control”?

  4. The Feedback | The Drucker Exchange
    February 21, 2012

    [...] to make sense of it all. Peter Drucker had advice on how to design effective controls, and so we presented his seven specifications and asked what our readers saw in their [...]


Leave a Reply