The Data Driven Organization

Executives know that one must effect a variety of organizational changes in a timely fashion, to support a technology change. Otherwise, the organization may resist or reject the change.[i] These changes may involve the formal and informal reward systems, organization structure, resource allocations, and cultural norms to transform into a data driven organization.

These truisms apply equally to data science technologies. However, while new information technologies most often affect daily work at the bottom of the organizational hierarchy, data science aims to change how every level of the organization works. Thus, one hesitation executives frequently express about using data science is about how it will and should change the way they work. This blog post centers on the executive role in promoting data science strategy.

When to Use Data Science in a Data Driven Organization

Most decisions, including most executive decisions, don’t require or deserve data science. Executives need not build an analytical model to decide whether to hold a meeting. Often when executives require data to inform decisions, workaday business intelligence (BI) technologies provide adequate insight. Data science thus leaves much of an executive’s work intact. In fact, determining when to use data science is the first decision that data science requires of executives. This decision, more than most, benefits from a great deal of systematic delegation.

There are three criteria that distinguish decisions an organization should make using data science.

  1. The decision has high total economic impact per time period. Formally, the product of the number of decisions per time period and the average economic impact per decision is high. Dividing these high-impact decisions into three common classes can help recognize them; see Table 1.
FrequencyIndividual ImpactDecision MakerExample
Low (strategic)HighExecutive A chief marketing officer must decide where to locate a new store.
Medium (tactical)MediumMiddle Manager An operations director must decide how to route and schedule deliveries from one city to another.
High (operational)LowIndividual contributor or decision automation system A retail sales clerk or Web site must decide what cross-sell offers to make to a retail customer.

Table 1: Decisions with High Total Economic Impact

What “high average economic impact” means varies with the maturity of the organization’s data science function, and with time. As an organization’s data science capability matures, doing data analysis becomes less costly. Likewise, over time data science tools become more robust and less expensive. Data costs also generally decline over time. So an organization practicing data science can expect the economic threshold above which data science is useful gradually to decline. The limiting case is a “data-driven organization” that applies data science ever more proactively.[ii]

  1. Unaided human reason may not make the best decision, even with the proper incentives. That is, there is a significant chance that the decision maker (in spite of doing their best) will make a sub-optimal choice. The decision maker may not know which variables matter, how the variables affect the outcome, how the decision is structured, or even what the alternatives are. In some cases the decision must occur too quickly for a human to make the decision at all.
  2. The information necessary to analyze the decision is practically available. The organization either has, or can procure at reasonable cost, the input data that a data analysis requires. Often making a rational decision requires sifting through a great deal of information.

In these situations organizations that apply data science to make sure they make the best possible decisions will have a substantial competitive advantage over organizations that make the same decisions manually.[iii] They’ll “do more with less.”

Because data science creates competitive advantage, executives should assume responsibility for providing the initial impetus to use data science. This does not mean, however, that executives should decide when to use data science informally. On the contrary: the decision when to use data science also satisfies the three criteria above. One should make the decision to use data science, scientifically.

Organizing the Search for Data Science Opportunities

Data-driven organizations (and those striving to become so)

  1. teach all employees the three criteria, and
  2. reward all employees for
    1. identifying strategic data science opportunities and
    2. communicating them to the proper decision makers.

These decision makers, in turn, evaluate the opportunities and hand off the most favorable of them to the data science function.[iv]

There are many ways to organize data science opportunity triage. For example, organizations with lean programs rely on routine analytical practices such as value-stream mapping to identify waste-reduction opportunities (but often gloss the analytical work required to make the most of them).[v] Likewise organizations with six sigma programs rely on executives and champions to select six sigma projects.[vi] We doubt there is a single right answer.

Regardless of how data science opportunities are identified, the organization should quantify the business case for each. Quantifying the business case means quantifying answers to the questions involved in the three criteria, notably

  1. How often do we make the decision?
  2. How much is the decision worth?
  3. How many alternatives are there?
  4. How much uncertainty is there about the outcomes?
  5. How much will it cost to get the data we lack for a formal analysis?

Answering these and similar questions lets the organization determine how much benefit to expect by approaching a decision scientifically.

Once the organization has quantified its data science opportunities, it should select the most favorable according to the organization’s financial decision-making standards (for example, by prioritizing opportunities in descending order of internal rate of return). One subtlety in choosing and maintaining a portfolio of data-science opportunities is recognizing where opportunities have overlapping costs or benefits. When two opportunities share costs but have distinct benefits, they may be more attractive in combination than individually (and conversely). Thus a degree of centralized awareness and evaluation of data science opportunities across the enterprise may lead to a more globally optimal data science portfolio than an organization will achieve if data science activities are fully decentralized.[vii] This is one argument favoring the emerging chief data officer role.[viii] Chief data officers are charged with determining data strategy, which includes deciding how best to leverage data science to support business strategy.

The Short Answer

In summary, the executive’s role is to

  1. establish systematic organizational triage of data science opportunities, using the three criteria,
  2. maintain and prioritize a portfolio of data science opportunities,
  3. sponsor an appropriately contoured[ix] data science function, and
  4. oversee execution of the data science portfolio by the data science function.

In the process, some executive decisions will themselves be identified as data science opportunities. The executive’s role in these decisions will shift from actually making the decisions to overseeing the analytical process and consuming the results. The responsibility for the decision will not change. But data science will mean more delegation, as well as better results!

Like what you are reading? Check out the rest of our blogs here!

[i] The idea of methodically reorganizing to support technology change dates back to Eric Trist and Ken Bamforth, “Some Social and Psychological Consequences of the Longwall Method of Coal Getting,” Human Relations 4 (1951), pp. 3-38. For a recent treatment of the issue in the context of information technology, see Luis F. Luna-Reyes et/al, “Information systems development as emergent socio-technical change: a practice approach,” European Journal of Information Systems (2005) 14, pp. 93–105, (visited March 9, 2014).

[ii] The Internet is now fairly buzzing with discussions of practices and cultural features of a data-driven organization. See for example, and (visited March 9, 2014).

[iii] (visited March 9, 2014).

[iv] Some writers characterize a data-driven organization by the degree to which all employees use data to inform decisions. We agree that employees make better decisions when they focus on the fact. But we argue that because most employees are not (and will not economically become) trained data scientists, employees should be trained to recognize when they can make a good decision themselves, and when they should hand off a decision to data-science experts. Part of the executive function is to make the sorts of organizational changes we list in this blog’s introduction, to encourage the organization to triage decisions well.

[v] Mike Rother and Joyn Shook, Learning to See: Value-Stream Mapping to Create Value and Eliminate Muda (Lean Enterprise Institute, 2003).

[vi] (visited March 9, 2014).

[vii] We know of organizations that use software applications to manage opportunity portfolios.

[viii] See for example (visited March 9, 2014).

[ix] Our white paper “Standing up a Data Science Group” explains exactly what “appropriately contoured” means.