Principles of Modern Analytics
There was a time, not long ago, when enterprise analytics was simple: You bought software from the leading vendor and installed it on a box. If your needs changed, you bought more software from the same vendor, and installed it on a bigger box. Analytics was a niche field populated by specialists, all of whom used the same software they learned in graduate school. People still believed that a single data warehouse could hold everything worth knowing.
The business cadence was, in retrospect, leisurely: If it took two years to implement a predictive model, well, that was just how things worked. Not that long ago, a big bank ran four campaigns per year to promote its credit card; at the time, executives thought that was an accomplishment.
Well, good-bye to all of that. Digital media is here; so are Web 2.0, mobile, cloud, and Big Data. The volume, velocity, and variety of data are exploding; enterprises are abandoning the ideal of the single data warehouse because it is impossible to stay on top of the tsunami. Diversity rules—we have a plethora of sources, an alphabet soup of platforms, and data everywhere: on premises, hosted by third parties, and in the cloud.
The changing landscape of data brings with it sweeping changes to the field of analytics: new business questions, applications, use cases, techniques, tools, and platforms. Techniques now considered mainstream were exotic five years ago. A single vendor once dominated analytic software; today, there are 851 analytic startups listed in Crunchbase, the leading source of information about startups. Open source software continues to eat the software world: two of the four Leaders in The Gartner Group’s most recent Advanced Analytics Magic Quadrant are open source projects, and surveyed analysts prefer open source analytics to the most popular commercial software by more than two to one.
Above all, the cadence of business accelerates exponentially. Yesterday, we ran four campaigns a year; now we can run four campaigns an hour. Nobody can afford to take two years to implement a predictive model; we will be out of business by then.
We can no longer afford the luxury of the blue chip, single-vendor proprietary analytics architecture. In its place, we see enterprises building an open analytics platform based on diverse commercial and open source tools, tied together through open standards. In this new world, each organization must define a unique analytics architecture and roadmap, one that recognizes the complexity of the modern organization and business strategy. This architecture will include many vendors and open source projects because no single vendor can meet all needs.
In this book, we propose an approach based on nine core principles:
- Deliver Business Value and Impact—Building and continuously evolving analytics for high-value business impact
- Focus on the Last Mile—Deploying analytics into production to attain repeatable, ongoing business value
- Leverage Kaizen—Starting small and building on success
- Accelerate Learning and Execution—Doing, learning, adapting, and repeating
- Differentiate Your Analytics—Exploiting analytics to produce new results
- Embed Analytics—Building analytics into business processes
- Establish Modern Analytics Architecture—Leveraging commodity hardware and next generation technology to drive out costs
- Build on Human Factors—Maximizing and grooming talent
- Capitalize on Consumerization—Leveraging choices to innovate
Next, we fully explore each of these principles because they are the foundation upon which Modern Analytics are built.
Deliver Business Value and Impact
Later in the book, we describe how to go about creating a unique analytics roadmap and how to prioritize projects. For now, suffice it to say that one of the principles of Modern Analytics is a focus on analytic projects with potential for game-changing value to your organization. To hold the organization accountable for delivering value, measure your current state to establish a baseline and set initial quantifiable target business objectives and ongoing business objectives. For example, current revenue is $100 million with CAGR 4%. The initial target is to identify 15% net new revenue with an ongoing net new revenue contribution of 10% annually.
Although such a metric can be easy to identify and measure, other metrics can be harder to identify and measure. To discover these potential metrics, identify points where business decisions are typically made. Start by measuring impact at these points. Then work toward establishing metrics that have a direct impact on the business. Whereas in the past, companies typically aspired for either a revenue metric or an operational cost metric but not both, today mature analytic organizations often establish metrics on both sides of the balance sheet. This sends a clear signal to the team that revenue growth has to be accomplished cost efficiently.
Savvy organizations identify potential analytic opportunities by thinking outside the box. Typically, the hardest, most entrenched problems in an industry or company have been around so long that people start to think about them as hard-and-fast constraints for their business. However, often, the barriers that existed in the past that made them impossible to solve no longer exist. Unleashing the bottleneck typically results in massive business value creation. Analytic-driven organizations dare to think outside the box and identify some of the most challenging problems facing their industry or business. When that is done, they work toward identifying how they solve or reduce the problem through innovative data- and technology-driven approaches. This is usually accomplished with a clean sheet brainstorming approach and imagining that all the resources needed to solve the problem exist. After ideas are vetted, the team typically has another brainstorming round to determine how to get everything they need to solve the problem without settling. Instead of using samples or backward validation1 to estimate a solution, the team will identify potential new resources—data, symbiotic partnerships, or technology—that will help them achieve their business objectives.
To realize the business value both initially and over an extended period of time, you need to deploy the analytics into production. Before any analytics can be deployed, the results of the analytic model need to be validated for accuracy. Today, that typically occurs in a “sandbox” with a limited subset of the data and in an artificial, nonproduction environment. It is all too common for an analytic model to meet or exceed business criteria in a sandbox but significantly underperform in a production environment. Be sure to evaluate your analytic models based on the environment that they will be deployed into, not any idealistic environment. Deploy the analytic models into a replicated production environment to fully test the model prior to going live to get a realistic assessment against the target business objectives. Where deploying used to be a “post” process after the model was built, deployment is now part of the full life-cycle analytic process. Once all potential technical deployment barriers are identified, obtain legal and/or procedural process validation before the “go-live” launch into production. After an analytic model is deployed, measure the initial business impact and identify quick ways to continuously improve on the results.