In several panels and conferences lately I have heard of the benefits of “data democracy”.  This is the notion that we will see much greater benefits from big data and advanced analytics as a broader set of managers have access to the data and the analytical tools.  A large number of companies are pursuing solutions to provide just such ease of access.

I do agree that the potential upside from more widespread access is significant for several reasons. It can solve part of the supply challenge of data analysts and their more highly paid cousins, data scientists.  It can reduce translation problems between operations and analytics. It can improve speed and efficiency by reducing the time from question to answer.  Perhaps most importantly, it can help improve the fundamental insights due to more appropriate questions being asked.

These benefits, however, come with equal if not greater risks to businesses.

• Analysis-Paralysis: Although it is great that we can now ask questions of any sort we want, it doesn’t mean that we always should.  In my experience serving clients I found that more data often led to more questions, but not necessarily more or better decisions.  For instance, how does a retailer decide on pricing and marketing?  Marketing managers have access to lots of data on what promotion worked in a similar category last week or month or year.  They have analyses that show if you spend extra money on advertising, not surprisingly you can charge higher prices.  The merchants have data that show the impact of price moves relative to competition.  The pricing “committee” meets to decide on price structure plans and with different analytical perspectives sends the teams off to do more analysis to understand alternative trade-offs.  In many cases each group uses its own analytical tools to bring new perspectives to the problem, often-times making the decision even more murky.

Yes, I know this is a governance problem, not a data problem.  No one has structured the critical question to answer.  No one has led the team to work together towards a conclusion.  There is no clarity in terms of who “owns” the decision.  All true.  However, I have seen this same pattern albeit with different metrics, questions, and industries countless times.  Humans tend to believe their own analysis more than someone else’s.  And in many cases they are incented differently driving a different set of questions and objectives.

• A little learning is a dangerous thing: As Alexander Pope wrote, “A little learning is a dangerous thing….shallow draughts intoxicate the brain…”  This point is even more true today with our access to tremendous amounts of “shallow draughts” as it was in 1709 when Pope wrote.  When we conduct an analysis and it “proves” our hypothesis, our beliefs become more entrenched.  Humans are naturally overconfident, but when supplied with a piece of data that supports our original bias, our confidence is at its zenith. This is known as the “confirmation bias” among psychologists.

As a result, people become ever more stuck in their positions.  Due to a biased search for information, people then tend to do analysis that will further support their hypotheses.  This in the end leads to a less flexible team and more room for narrowing of perspectives rather than what we hope for from data access…widening perspectives.

Access to data does not keep us from our human psychological limitations.  In fact, it may only reinforce them.

The conclusion, however, is not to put the genie back in the bottle and hide the data behind the wizard’s facade of data scientists.  There are several key elements to manage the democracy of data to enable it to deliver its benefits with a minimum amount of risk.

1. Training.  One of the biggest selling points of many of the new data analytics tools is the minimal amount of training necessary to use the software.  “Just click here to download and in 15 minutes you’re up and running.”  How true.  But what (and where) are you running?  The necessary training is not just in the techniques and statistical fundamentals.  It is in training people of their own biases and how to adjust for them.  It is training on what questions can and should be asked rather than just on how to come up with answers.  It is training on “so what” development or how to develop the implications of any analysis.  As with Excel, people are expected to “figure it out” or take a course on line.  But in a world where the data tools are ever more powerful…and the reliance on their outputs ever more accepted…it is incumbent on leaders to ensure proper holistic training of how to use the power of democracy appropriately.
2. Clarity of governance.  In some ways, this data “babble” may force clearer governance in companies as it makes more obvious the lack of decision-making authority.  To take advantage of the possible efficiencies from greater data access, enterprises must establish much clearer guidelines on decision-rights.  Who has final accountability? Who is authorized to select one model’s outcome rather than another’s? What are the consequences for that person or team to ensure they are incented appropriately?  Without clarity of governance, data will lead to less rather than greater efficiency.
3. Robust quality control on modeling.  Quality control typically lives in the realm of the operations or manufacturing organization.  In a world of much greater access and reliance on data models, “six sigma-like” quality control procedures are as important in your modeling as in your production line.  Unfortunately, in very few enterprises today are these processes in place.  And in fact, in most they are not even discussed as necessary.  Just as in manufacturing where quality has become the expected responsibility of everyone on the assembly line or upstream, now in modeling everyone must have the same level of responsibility for calling out problems and stopping the process when a mistake is encountered.
4. Embed a test & learn mindset across the organization.  The risks of data democracy increase when the decisions are black and white/all or nothing.  A test & learn mindset encourages managers to regularly test their data insights and hypotheses in real world settings against a control group.  Test & learn reduces the analysis paralysis because it creates real world outcomes.  While this approach to management is typical in many tech start-ups, it is not second nature outside that community.  Changing the cultural dynamic in a long-standing organization will require leadership to set the example and begin to move towards a more “learning” environment.

The future opportunities from the data and analytics revolution are too large to ignore.  But like many other such opportunities for enterprises, the gain is not without some pain.  If one approaches it as just the next step in making decisions, the risks may outweigh any perceived benefits.  Take a holistic step back and decide what in the organization needs to change to really tap into the upside available.