Building Better Insights for Analytic Applications

November 27, 2018
Miroslav Sova's picture
Sr. Technical Product Marketing Manager
Mirek is Senior Technical Product Marketing Manager at GoodData. He has 12+ years experience in the tech industry with B2B cloud-based products. He started as a .NET software developer and over time entered into product management and more recently into product marketing. Throughout his career, he has contributed to launching an array of products - cloud security systems, software development tools and an enterprise document digitization platform. Mirek holds a BSc in Computer Science from University of Wollongong, Australia and a Master of IT from the University of Sydney, Australia.

Part Four of Four

So far in this blog series, I’ve talked about integrating embedded analytics into data products to help achieve greater usability, and how semantic layers, logical data models, and metrics are good ways to achieve that for large-scale analytics applications. From all of this work going on behind the scenes on the product side, the end result—from the user’s perspective—is an actionable insight that helps them make better business decisions or, better yet, suggests the next best action to take.  

What is an insight in an analytic application?

Forrester analyst Brian Hopkins has defined digital insights as the recommendations and possible actions supported by data in a given data product.” We additionally refine that definition of insights as collections or visualizations of metrics that reveal patterns or previously unknown trends -- for example showing change over time, or a combination of metrics that together show a correlation. Insights are a powerful way to help your business grow by illuminating areas of opportunity. By uncovering trends and relationships, and delivering recommended actions to the user, the application can inspire users to think of creative ways to solve a problem and make better decisions.

Where do insights fit in the semantic layer?

In slightly more technical terms, insights combine one or more metrics and slice or filter them by selected dimensions to create relevant and actionable information. They are often delivered in the form of data visualizations—things like charts and graphs—and they can incorporate information gleaned from machine learning and artificial intelligence, not just from raw analytical data. Because of their value, companies often strive to use insights to support their business decisions, preferably when those insights are embedded in a user’s daily workflow.

Best practices for insights in large-scale analytics apps

In order to be meaningful to the business, insights must be designed to support decisions that will further the organization’s goals. This means validating key objectives and desired outcomes early in the process and identifying places in the workflow where additional information or a recommendation can make a positive impact. Insights should be tested under a variety of conditions and edge cases to ensure that the implied actions and recommendations are accurate.

Don't overwhelm your users with too many insights or too much data at once. Instead, walk through the workflow and identify key decision points. Provide insights that support the most important decision points within the business process you are trying to improve. Understand your user’s needs and workflow, and think about revealing information only when it is needed, in a stepwise fashion or by allowing drill-downs.

Keep in mind that the majority of your users will not be data analysts. Utilize good user-centered design principles to provide users with a logical, consistent and intuitive layout that keeps the focus on the most important metrics. If you do decide to allow for self-service exploration, make sure that you choose a user-friendly platform that not only lets users easily combine measures to build insights, but also gives automated recommendations to modify, customize, and break down an insight.

Finally, remember that insights that are integrated into day-to-day (especially front-line) decision making are on the critical path, so performance, accuracy, and reliability cannot be taken lightly.

Semantic layers are crucial for large-scale embedded analytics

To sum up this blog series, the semantic layer of your application documents the way your business uses data to measure itself, and translates that data into business-friendly terms. It creates a single-source-of-truth for everyone from developers to business users. And it allows for agility and flexibility by providing an abstraction between your application and all of the original sources of data. Remember to validate and support the business goals of your organization, and keep the focus on your users’ critical decision points while you’re designing. And as your business goals and procedures change, your semantic layer will need to change too, so build up your metrics in a way that allows for future extensibility. By doing so, you’re more likely to build an application that continues to deliver value to end users for years to come.