Professionalizing Advanced Analytics: Mobilizing for Value (Part Three)

The promises of advanced analytics are inherently appealing to businesses: make real-time, data-informed decisions, optimize and automate processes, and monetize data in new ways. In the final installment of a three-part blog series, we’ll examine how development and deployment for business value round out the core elements of a successful advanced analytics capability.

In our last blog, we addressed organizational culture and technology and their respective roles in driving embedded advanced analytics capabilities. We will conclude our series by discussing the importance of the fourth and fifth success factors:

  1. Ensure analytics teams have strong leadership and alignment to business value
  2. Develop an analytic organization and culture through proper role identification, career development paths, and recruiting 
  3. Establish modern tools that fit into existing technical infrastructure
  4. Have a well-defined development methodology to lead from idea to prototype to production deployment
  5. Operationalize models for proper business value

Success factor 4: Have a well-defined development methodology to lead from idea to prototype to production deployment

There are two core principles to apply when configuring development methodology for successful advanced analytics capabilities:

  • Develop a sound process for development teams
  • Create a flexible format for analysis work

1. Develop a sound process for development teams

In the last blog we discussed the tools and technology available to help build analytics capabilities. A sound process for development teams is the key enabler of these tools. In our work, we typically approach projects with an agile-first mindset and have developed a methodology that blends successful agile concepts with the unpredictable nature of advanced analytics work. For example, we have found that out-of-the box Scrum methodology often puts developers in an uncomfortable position: how can tasks be time-boxed into user-stories, when discoveries over the course of a sprint may reveal that a method or dataset chosen for analysis is inappropriate for the task at hand? Keeping focus and driving progress forward is important, but data science is creative work that requires adjustments as the team gets comfortable with the business goals, evolving tools and data being used.

2. Create a flexible format for analysis work

Project planning frequently needs a flexible format for analysis work. Without pre-analysis, it may be difficult to articulate potential business value from an analytics project. The unfortunate reality is that many data science projects fail to deliver because of poor data quality, information density, or other factors that aren’t what was hoped. For example, text fields that a human can translate into actionable information may baffle a machine learning tool because the data volume may be too low, or because of highly jargon-based text and acronyms, or because of inconsistent writing style. The most effective solution is to rapidly prototype against “close enough” requirements and data. This can shed light on whether a project will be successful when scaled up to larger and more diverse information. It can also help analysts have confidence that their chosen methodology will work, enabling the delivery of prototypes at a faster rate. 

Success factor 5: Operationalize for business value

Now, how do you operationalize a promising prototyped model inside a production environment? There are a few critical hurdles to clear before being ready: 

  1. Test across the range of data in your business. Testing across the entire range of data of your business is another key step, and one that can take much longer than the initial prototype phase. Teams must be prepared for this step and the challenges it presents and start developing testing scenarios in prototyping.
  2. Integrate advanced analytics into DevOps practice. Before you can scale up, you must first have established pathways into the IT promotion process, solving for components such as production infrastructure, migration processes, and time on other teams’ calendars. Determining how to integrate into your DevOps practice will make your pushes to production easier and more effective.  
  3. Define metrics to gauge performance. Business needs also change over time, so your models cannot remain stagnant. Extending the DevOps practice into model performance monitoring, rather than just focusing on functional aspects, ensures that your model remains useful for the organization and continues to deliver on promised value.  And let’s not forget, this also quantifies the value advanced analytics brings to the bottom line—helpful tidbits during budgeting cycles.
  4. Integrate with the business. And finally, as a capability grows, the emphasis on repeatability and team profitability grows. We often see tighter integration with PMO capabilities and the creation of more senior analytical roles such as the Chief Data and Analytics Officers. These roles are to strategically shepherd the analytics organization and give analytics a seat at the table alongside traditional IT and core business skillsets. 

Bringing it all together

Developing a stand-out advanced analytics organization takes time and requires the balancing of a variety of factors, which can be confusing to know where to start. If you’re able, start by bringing on that charismatic leader than can bridge business and tech and build the team from there. If you must show progress up front to justify the cause, separately find an enthusiastic business partner who wants to explore use cases, and someone in your organization that can develop statistically significant data science.  Put them together with oversight and monitor the results. If advanced analytics has a place in your organization (and it probably does) either of these approaches will yield results, and further you down the path to becoming an analytics-driven organization.