The missing link in many data science projects: Decision intelligence
Blog: The Enterprise Project - Digital Transformation
Digital transformation is the flavor of the season. Every company has accelerated its efforts to digitize operations, gather intelligence, and rapidly respond to a changing market.
McKinsey senior partner Kate Smaje says that organizations are now accomplishing in 10 days what used to take them 10 months. With data powering better and faster decisions, she says, the road to recovery is paved with data.
As a result, most organizations are trying to adopt data-driven decision-making. They are hiring data scientists, buying the best tools, and greenlighting big-bang analytics projects.
However, none of these efforts alone will deliver results. They can lead to a build-up of activity, expectations, and expenses, but the business outcomes will not just magically happen. A whopping 80 percent of data science projects fail. Wonder why? There is a critical element missing from these initiatives: Decision intelligence.
Decision intelligence is the application of data science within the context of a business problem, and it’s achieved by factoring in stakeholder behavior to influence adoption and decision-making.
[ Get answers to key digital transformation questions and lessons from top CIOs: Download our digital transformation cheat sheet. ]
Decision intelligence augments data science with two disciplines that are often ignored when it comes to data: social science and managerial science. It’s only when you combine all of the principles and skills from these three disciplines – data science, social science, and managerial science – that you can unlock business decisions. It is essential to contextualize data insights with social behavior in an organizational context to enable decision making.
How can decision intelligence help you?
The reason so many data science projects fail is that IT leaders are not applying all three of these disciplines. Decision intelligence brings together these three disciplines to address failure points.
Here are the roles each discipline plays:
To help your target audience make the right decisions, you must first strive to understand them. Go beyond their direct asks and find out how they think, feel, and act. This attempt to understand user behavior in a social context will help you gather the right data. The knowledge gained will help you tailor your insights and adapt them to the specific social and organizational context. Remember that this is a continuous journey that calls for softer skills, and it goes much deeper than gathering user requirements at the start of a project.
With a strong understanding of your users and the business context, data science equips you with a toolkit of techniques to help you ask the right questions of data. Whether you choose to apply simple descriptive analytics or deeper predictive analytics, you must ensure that the business insights are useful and actionable. Finally, tap into the principles of information design to make the insights consumable in the form of visual data stories.
Your job doesn’t end with providing relevant business recommendations to your target audience – you must also help them act on these recommendations by adopting the solution. This often calls for changes in business workflows and necessitates coaching of stakeholders to manage organizational change. Plan a variety of interventions on an ongoing basis to manage, monitor, and course-correct from a managerial perspective. Often, these are entirely overlooked or are seen as an afterthought in organizations.
Improving commodity price trading outcomes with decision intelligence
How does this all play out in the real world? Consider this example from a large conglomerate that was trading in agricultural commodities. The company had a great use case for data analytics: It wanted to predict future prices and improve the quality of its trading decisions.
Our consulting firm spent several weeks collecting data, curating it, and building a variety of machine-learning algorithms. These ranged from simple forecasting techniques to more sophisticated algorithms such as neural networks that could predict the expected price in the following week.
After crunching the data, the models were delivering over 95 percent accuracy, and our visualization layer was recommending relevant actions for the business users. The solution looked promising and the teams were excited about the results.
However, once the business users reviewed the recommended actions, the excitement began to fade. The verdict was that the recommendations were not actionable and the proposed plan was unusable because the solution’s precision was not good enough to make a trading decision.
While this came as a shock, the teams resisted the urge to spend more time tweaking the algorithm for accuracy. They went back to the drawing board.
What was missing was decision intelligence.
Incorporating social and managerial sciences into the solution
Our team took a step back and studied the business processes, historical trades, and past decision outcomes. They interviewed business users to find what information was critical for them to make their decisions – applying the critical social science discipline. They learned how to educate users on the technology solution, and how to earn their trust and approval – applying the all-important managerial science discipline.
What they found this time around was surprising: They realized that predicting the exact price for the coming weeks was not important, contrary to prior assumptions. Instead, they found it was sufficient to know whether the price was likely to increase or decrease – a prediction of the direction of change.
Since the organization held physical inventory, this insight could help them make the decision of whether to sell or hold inventory for a few more days. With these inputs from business analysis and social science approaches, the team revisited the data science techniques. The new set of machine learning techniques delivered a slightly lower accuracy of 88 percent on this alternate approach. However, the recommendations on whether to sell or hold inventory were solid and actionable.
Finally, the team set in motion the interventions needed from the managerial science perspective. They incorporated the algorithm recommendations into the business workflow and worked with the stakeholders to influence the change needed for adoption.
With data insights driving the business decisions, the team monitored progress over the subsequent months to measure the savings and return on investment. Within the first quarter of deployment in production, the revised solution developed with decision intelligence delivered a 3.2 percent increase in revenue directly attributable to the pricing.
Data science alone wasn’t enough. Decision intelligence helped come up with a comprehensive approach to deliver the desired business outcomes.
Decision intelligence to promote data-driven decision-making
As part of your digital transformation efforts, before you outlay your budget for data and analytics, find out how it will lead to better decisions.
Are your teams paying attention to the users and their social behavior, rather than just the insights? Are your teams working to help the user act upon the recommendations and manage adoption in the long run?
You need much more than data and analytics to make better business decisions. You need decision intelligence.
[ Get exercises and approaches that make disparate teams stronger. Read the digital transformation ebook: Transformation Takes Practice. ]