Clarity of purpose for data science projects: a case study
By James Taylor, Dec 07, 2016
I work with many clients who are trying to effectively adopt advanced analytics – data mining, predictive analytics, data science. One of the biggest problems these clients face is to how to get everyone – business, IT and data science professionals alike – on the same page. They have seen how much time and money is wasted if the data science team goes off on the wrong track, producing a great model that can’t be used or deployed. They know that clarity about the business problem right at the start is an essential ingredient but they don’t know how to bring that clarity to bear.
One of these clients has been successful over the last year or so using a new technique – decision modeling – to focus their efforts and build a shared understanding with their business clients. The result is a great new leading practices brief: “Bringing Clarity To Data Science Projects With Decision Modeling: A Case Study”. In this organization, as in most, the business people involved in these projects don’t know much about analytics/data science while the data scientists don’t necessarily understand the business context. Data science tools and techniques provide no way to bridge this gap at the beginning of a project – visualization and story-telling tools might help when the analytic is finished but these are not helpful before you build a model which is when you really need the clarity.
Decision modeling takes the business decision that the organization wants to improve – what cross-sell to make to a customer, what discount to apply to an order, which supplier to use to fulfil an order – and models it to see how data science and analytics can be applied to improve it. Rather than beginning with the data that is available or with “cool” analytic techniques/technologies, organizations adopting decision modeling focus on the business problem at hand (making a better decision) and model that. Decisions are decomposed into sub-decisions, the input data and knowledge involved in the decision are identified, and the model is linked to the business results and organizations involved.
The case study describes some great examples of decision modeling bringing real clarity to a project, helping projects regain purpose and keeping teams from being distracted from the core business value they need to deliver. Even very simple decision models have been useful and the case study shows how readily the approach can be adopted – ideally as part of a project methodology like CRISP-DM where decision modeling falls naturally into the Business Understanding step. There’s also more details on how to do this in “Framing Requirements for Predictive Analytic Projects with Decision Modeling” another IIA brief that published in 2015.
About the author
James Taylor is the CEO of Decision Management Solutions. James is the leading expert in Decision Management Systems — systems that are active participants in improving business results. Decision Management Systems applies business rules, predictive analytics and optimization technologies to address the toughest issues facing businesses today, changing the way organizations are doing business. James is passionate helping companies develop agile, analytic and adaptive processes and systems. James has over 20 years developing software and solutions for clients and has led Decision Management efforts for leading companies in insurance, banking, health management and telecommunications.
In addition to strategy and implementation consulting, James is an experienced and highly rated keynote speaker at conferences in the U.S. and around the world. James also regularly runs webinars for clients, and as educational outreach for Decision Management Solutions. James has just completed “Real World Decision Modeling with DMN” with Jan Purchase (Meghan Kiffer, 2016). He previously wrote “Decision Management Systems: A Practical Guide to Business Rules and Predictive Analytics” (IBM Press 2012), “The Microguide to Process and Decision Modeling” with Tom Debevoise, and “Smart (Enough) Systems” (Prentice Hall, 2007) with Neil Raden. He has contributed chapters to several books on business rules and business analytics.