A Culture of Analytics: Why Amazon & Netflix Succeed While Others Fail
Blog: The Tibco Blog
When it comes to advanced analytics projects, it often seems like success stories are the exception rather than the rule. Many organizations would like to emulate the data-driven culture displayed by leaders like Amazon, Google, and Netflix, yet have trouble with operationalizing analytics. And there’s broad agreement that analytics and ‘big data’ projects fail at very high rates — more than half the time.
If I had to pick one fundamental reason why, I’d say that it’s because analytics projects are irreducibly complex and multi-faceted: they typically have many moving parts, more than a software project for example. So there are many opportunities to stumble along the way. Projects often fail right at the start, when data is off limits or hard to find. And project goals may shift rapidly: often it’s not known whether the data will confirm hypotheses at all, or yield actionable insights, or support predictive models. There may be problems of scalability, of integrating with operational systems, of model accuracy, and so on. And, in the end, there is the stubborn problem of how to deploy complex workflows and models, often requiring a tedious manual conversion from data scientist code into real-time scoring.
As a result, analytics projects can quickly become overwhelmed with technical details. I believe that an agile, business-driven approach can help ensure success.
Analytics initiatives are often viewed solely through a technical lens, resulting in situations where organizations get lost in the weeds agonizing over statistical models, database structures and platform infrastructure. Amazon and Netflix are able to use data in innovative ways not just because they are technically advanced, but also because they’ve created a “culture of analytics” that pervades every aspect of their business.
Here are four key considerations for executing an analytic strategy that puts the needs of the business first:
- Have a purpose. While this might seem obvious, the reality is that many companies lose their way with analytics because they focus first on technical specifications and not enough on a tangible business objective. In effect, they put the technical cart ahead of the business horse.
Let’s step away from analytics for the moment and imagine another type of project — constructing a building. Would you purchase the materials and hire general contractors, plumbers, and electricians before you had a clear understanding of the building’s purpose? Of course not.
Yet all too often, this is exactly how organizations proceed with analytics deployments. They start building infrastructure and hiring data scientists and evaluating technologies (this or that database? Hive or Spark SQL?) long before they’ve specifically defined the business problems and opportunities that can be addressed using analytics. A technology deployment, no matter how seamless or advanced is not going to magically transform your organization into a data-driven powerhouse like Amazon, especially if its purpose is vague and poorly defined.
To avoid costly mistakes, business end users must be part of the analytics strategy from day one. These are the stakeholders who can weigh in on the use cases that have the most to gain from big data and provide a realistic picture of how analytics would/could fit into the applications and processes that they rely on every day to do their jobs.
And start with just one analytics project, with high business potential, and readily available data. Then figure out what technology you need.
- Link “insight” to action. What does it really mean to solve a business problem? In the analytics world, it’s become common to assume that the best way to solve a problem is to provide more information, or “insight,” about it. But that’s the equivalent of having a meeting to address a specific issue and “resolving” it by scheduling another meeting. Insight, to be of any value, must be predictive in nature and – this is the important part – drive action quickly, seamlessly and automatically.
It’s this last step that trips up many organizations. Imagine a company that wants to optimize its Q2 sales. The organization might start by attempting to use analytics to better predict sales volume. But telling the sales team that volume is expected to drop in Q2 and even providing insight into why it might do so is not enough. Predictions should launch tangible actions that sales and marketing can take to turn things around.
Amazon, Google, and Netflix are masters at transforming insight into data-driven action. For example, Amazon uses big data to automatically customize the browsing experience for its customers based on their past purchases, and optimize sales. Netflix seeks to directly impact customer behavior with data-fueled recommendations and, more recently has used data to spawn the successful creation of original content that they’re confident audiences will like.
Connecting analytics to actual results demands “high resolution” data and predictive analysis that prompts actions within purchasing, sales, lead generation – whatever the business objective may be. But one final piece is needed to make this all simple, and seamless: integration.
- Push analytics to business end-points. Most business and customer-facing stakeholders within an organization don’t know Hadoop from NoSQL from Apache Hive. Nor should they. Analytics infrastructure is highly technical and not easy for non-engineering types to interact with and use.
Successful analytic cultures find ways to push outputs to and integrate seamlessly with, the go-to applications that real-world sales, marketing, procurement and other business-level decision-makers use on a regular basis. These include tools like Salesforce.com, Marketo, and Zendesk. A “touchpoint” approach to analytics concentrates on how and where analytics will realistically be utilized and ensures that lessons learned are integrated into these applications — a critical part of connecting insight to action.
- Create feedback loops. Finally, an analytics strategy is never “done.” This is another reason to avoid getting stuck in the weeds by laboriously perfecting statistical models over many months. The analytic climate is always changing for most organizations as new data sources become available, and as business opportunities, challenges and priorities evolve.
Organizations should, therefore, focus on a flexible analytics infrastructure that is not only able to meet a number of diverse requirements across the enterprise, but also one nimble enough to adapt quickly to the needs of the business. This is where the “agile” movement in software development can serve as a useful example: instead of aiming for a “perfect” solution, focus on rapid development of fresh models that can be pushed quickly into production. Then, monitor performance with A/B testing and use “feedback loops” to continually test, refine and improve analytic applications and begin the cycle again. In this way, organizations can rapidly and continuously get actionable data out to the users who need it, even as needs evolve.
The most advanced analytics-driven companies in the world give their employees remarkably free access to their codebases. They encourage safe experimentation with open access to source-code, rigorous code reviews, clearly-defined metrics of success, and instantaneous feedback loops.
Many companies want to be able to emulate data darlings like Amazon, Google, and Netflix. But those organizations have succeeded because they have wisely integrated analytics into the very fabric of their business. Before jumping into the deep end with highly complex technologies and advanced algorithms, companies just getting started should first address low-hanging fruit and build analytic applications that are valuable to actual business users. Don’t let the perfect database or latest and greatest statistical model get in the way of achievable results.
(This blog entry is based on an article that originally ran in insideBigData.)