Focus on the Last Mile
Today, very few organizations get to the promised land of deploying analytics into production environments to drive game-changing business value for their organization. To get to this end goal, start with the end in mind and work backward. Understand day-to-day issues at every level in the organization by speaking with frontline workers—from strategy through to execution. These domain experts are acutely aware of the issues, problems, and constraints impeding their success. Clearly understand what it will take to achieve success—not how success can be attained. With this understanding, set quantifiable and ambitious goals for your analytics. For example:
What is the target business value to be obtained?
A 3% lift in revenue?
Inventory saving of $10 million annually?
Total cost savings of $100 million in the first year of deployment?
What is the expected service level agreement (SLA) for the business?
Reevaluated credit scores nightly?
Portfolio evaluation within 5 minutes?
What is the operational model?
How does the model get moved into production?
Does this analytic model need to integrate with other business systems? If so, how do the operational processes and decisions change?
Is this analytic model triggered from another business system?
Is this analytic model deployed in one location or multiple locations?
Are there multinational or localized requirements?
What is the frequency of updates to the model?
What are the key success factors that measure the business impact?
How is success measured?
What constitutes failure?
How long does the team have to achieve success?
What is the model accuracy?
Is the accuracy “good enough” to realize immediate business value?
How much should the model be improved in what period of time?
Traditionally, one team—quants, statisticians, or data miners—has been responsible for the model creation while a second team—typically IT—has been responsible for the production deployment. Because this often crosses organizational boundaries, there can be long lags and disconnects between the model creation and the model deployment or scoring. The teams must function as if they are one team even if organizational boundaries exist and will persist. A full life-cycle methodology can serve to bring these two teams into alignment if the analytics methodology goes beyond just creating and assessing the initial analytic model to encompass the actual production deployment and ongoing reassessment of the analytic model to achieve the business objectives.
With Modern Analytics, teams focus on delivering results quickly rather than waiting to build the “perfect” analytic model. They do so by starting with Proof-of-Concepts (POCs) or prototypes that may be limited in scope, but help the organization increment toward realizing business value. They quickly mature and harden the POCs or prototypes into a production deployment where the rewards can be systematically reaped.