Deploying advanced analytics is a transformative process in any organization. Sometimes, new findings upend long-held beliefs and disrupt established business processes. This can engender a hostile reaction to the changes introduced by advanced analytics. When deep-held worldviews are threatened, emotional responses to defend the status quo are to be expected, even though they run counter to the facts. Such emotional reactions by stakeholders may block the successful adoption of analytics throughout the organization. How can analytics practitioners successfully press the case for change when emotions trump facts?
Data Versus Perceptions
Imagine a company using a combination of key metrics, pattern identification, expert insight, and ad hoc judgment to detect and investigate fraud to implement a predictive model to improve its fraud investigation. The analytics team develops the fraud model using data collected by the company and third-party sources, and deploys it within the existing business framework. In the context of this new paradigm the business leaders and investigators may draw new inferences about their company, its practices, and the world around them. Business leaders want the increased efficiency the new predictive model will provide. But, investigators who, before the model deployment, had a particular understanding of their investigative process may feel threatened that the new model will replace them and feel compelled to defend old practices. It is better to establish communication early between technical and business stakeholders to build a common understanding of, and support for, the analytics strategy.
A successful analytics strategy requires an organization to assess its unique business challenges, match those challenges with relevant data and resources, and establish processes that grow its capabilities. To build an analytics culture it is vital that technical and business stakeholders are involved in the analytics assessment and data discovery processes. In most cases the communication goes well. However, in rare cases valid, but emotionally charged, questions may arise:
- Are the data really as objective as we believe them to be?
- Do these findings threaten how we do business?
- Why does the picture presented by the model differ so much from our perceptions?
- Can we trust a model we do not fully understand?
Analytics practitioners must frame the conversations around these emotional reactions that could derail an analytics project or hinder widespread adoption of analytics in the organization.
The Six Step Communication Framework
A recent article by Michael Shermer in Scientific American accurately addressed the scenario described above. Shermer highlights two behavioral mechanisms that cause people to cling to belief over evidence: cognitive dissonance and the backfire effect. As humans, we struggle to simultaneously hold conflicting thoughts. We also have a bias toward coherence, and any information that contradicts a coherent worldview can be threatening. Repeated attempts to correct misperceptions using facts often backfire, leading to deeper and deeper entrenchment.
Shermer proposes a six step framework for communicating information from facts when emotions have come into play that may help analytics teams successfully channel communication to build project consensus:
- Keep emotions out of the exchange
- Discuss, and do not attack
- Listen carefully, and try to articulate the other position accurately
- Show respect
- Acknowledge that you understand why someone may hold that opinion
- Try to show how changing facts does not necessarily mean changing worldviews
Each of the six steps in Shermer’s framework highlights the complimentary role of emotional intelligence in communicating scientific findings to non-experts.
Applying Analytics with Emotional Intelligence
Data Scientists are employed as experts to speak intelligently about technical matters. Having a high-EI (Emotional Intelligence) can mean the difference between bridging a gap in communication and losing an audience. Beck and Libert argue human capabilities will become more and more prized over the next decade and analytic (and non-analytic) workers should focus on developing skills like persuasion, social understanding, and empathy as human differentiators in an increasingly automated world.
Elder Research has learned many lessons over more than two decades of helping clients to successfully navigate all phases of an analytics project—including helping to build or expand internal teams, developing analytic strategies, building and deploying models and visualization tools, and enhancing team capabilities through analytics training and mentoring. In addition to the advice from Shermer, Beck and Libert, we would like to share some insight from those lessons:
Be aware of the limitations of data – because of its quantitative nature, data can gain a false air of objectivity. All data, even very good data, has biases and limitations. No data are perfect. Data Scientists should clearly state the limitations of the data they used to build a predictive model. They should take care not to misappropriate their findings based on an imperfect understanding of the data or the model. By acknowledging these limitations, Data Scientists can show sensitivity to the other position, and consider whether it fills in any of those gaps.
Analytics experts will have redundant strengths and similar blind spots – having agreement amongst quantitative experts is not a sufficient basis for declaring results to be factual. For example, even if several analysts agree based on metrics that a certain kind of fraud is being missed, there may be other arguments outside of the data that explains why this type of fraud is not currently captured. At Elder Research, we work to mitigate this issue early in the process by working collaboratively with the business and technical experts to make sure both perspectives are heard.
Balance accuracy versus interpretability – depending on the context, it may be necessary to compromise the performance of a predictive algorithm in order to preserve interpretability of the factors that led to a given recommendation based on model results. This tension between accuracy and interpretability is a fundamental one in analytics, but is especially important to consider in an environment that is new (and perhaps hostile) to the changes introduced by advanced analytics. Even if it is possible to build a better model, start small with a narrow, well defined scope, and add in complexity and sophistication as stakeholder buy-in increases.
Predictive analytics is a powerful motivator of change that can bring about transformation and capture real value, if it is applied appropriately and conscientiously. By having a well conceived analytics strategy, early participation from business leaders and end-users, and open communication, the results from an analytic model will be greeted with excitement and a willingness to adopt and act on the results. By focussing on emotional intelligence with as much care as they do machine learning, Data Scientists can foster a positive environment for analytics adoption.
 See Kahneman, D. Thinking Fast and Slow, pg. 114: On the subject of the human brain having a tendency to generate a confident picture from incomplete information, Kahneman states “. . .we are prone to exaggerate the consistency and coherence of what we see.”
 M. Beck and B. Libert, “The Rise of AI Makes Emotional Intelligence More Important,” Harvard Business Review (Online), February 15, 2017