Imagine you’re the head of business intelligence (BI) for a large company, or maybe you already are. You have dozens of dashboards, with countless variations, being used by thousands of users across every business unit and across multiple systems. The recurring piece of user feedback you receive is: “it would be great if we had a story explaining this dashboard to me!”
How would you feel if your only option was to have analysts and business leaders manually create templates? Manual templates need to consider every possible scenario that may occur—not only in the dashboard itself but in the underlying data as well. Would you trust the potentially large team of people tasked with building these templates to create a configuration that tells an accurate story for 100 percent of your dashboards? Could this manual process miss important scenarios or avoid introducing manual errors in the template? When dashboards need to be updated to account for new sources of data or include a new type of visualization, how can you be sure that every story template is updated correctly?
These are the types of questions and concerns I discuss with clients daily at Narrative Science. Most individuals who use BI dashboards accept, at a high level, that a dashboard with plain-English stories is better for most users than a dashboard without them. However, the concept of embedding natural language to explain dashboards is a new thing to most people, so my team and I find ourselves spending a lot of time educating and helping clients ‘see around the corner’ when contemplating adding stories to dashboards.
Thankfully for our clients, Narrative Science allows you to embed a natural language that requires no manual templates and allows for a consistent methodology to be applied in every scenario. Now, don’t get me wrong, there is a time and place for highly structured, customized content. Our intelligent automation platform, Quill, provides those exact capabilities. What I’m referring to here, is really understanding how a BI leader or analyst can create and maintain insights within hundreds or thousands of visualizations.
Here are a few examples of the dangers clients of ours have run into when attempting to create narrative templates to explain their dashboards.
1. Humans make mistakes
Say a particular dashboard’s main purpose is summarizing analysis of numerous metrics over the last 10 years. The template builder wants to communicate correlation analysis to users. What if the builder inserts an incorrect calculation for correlation? Or, inserts a hard-coded calculation that only looks at five years of data every time, but when the user filters on the last two years, the calculation does not update? Moreover, there are multiple ways of calculating correlation—how does the user know which method is being applied in the dashboard they are looking at?
2. Humans can’t scale like machines
Over time, updating and maintaining dashboards with narrative templates becomes extremely difficult. If dashboards, on average, are updated once a quarter, with 500 dashboards across your business, that’s 2,000 updates a year. This means, not only do you have to update the 2,000 visuals but also the 2,000 templates to get the accompanying story. If each update took five hours on average, that’s 10,000 hours annually just updating templates. Some things to consider:
- How many dashboards do you manage?
- How much time will it take to maintain these dashboards in your organization?
- Will you be able to make the updates correctly?
- How do you create a scalable quality assurance (QA) process to validate quality?
3. Foreseeing every scenario in a dashboard is improbable
My favorite example of this situation was when I was working with a Tableau sales consultant for a leading global systems integrator on a local government project. Our extension was able to pinpoint eight suspicious payments in a Tableau scatterplot, which contained thousands of data points, out-of-the-box and without any configuration. The consultant could not visually identify these suspicious payments, but it alerted him to investigate further. When applying a templated solution to the same dashboard, it was able to find the same risks, but ONLY after hard coding the template to find those payments. This begs the question: how many possible insights would you need to account for in a template that our system would just find automatically?
4. Configuration is complex
Many users are concerned that extensions for their BI tools are intimidating. Since one of the key tenants of modern BI tools is eliminating the requirement for programming, no one wants to reintroduce a feature that requires coding. Configuring a template fundamentally contradicts this concept and these solutions often require some programming language to configure. Additionally, this configuration takes place outside of the dashboard. Make sure you are considering where and how the set-up occurs and how this may disrupt your workflow. With our out of the box extension, users can be confident that they are gaining value without programming or special administration required.
5. Natural Language Generation methodology is inconsistent across BI platforms
Most companies use numerous platforms, including custom-built dashboard tools. For example, a large software company has tens of thousands of users across Qlik, Tableau, PowerBI, and Sisense, plus a homegrown system. Sometimes, the dashboards are recreated across each system and designed to analyze the exact same underlying data. Using templates, it is very hard—maybe impossible—to ensure consistency in how narratives are created across systems. In this scenario, having one vendor apply the same methodology, regardless of the BI platform, makes it easier for BI leaders to continue supporting multiple systems for their business partners.
6. New features are nearly impossible to deploy
Say 30 percent of users request a new feature or capability and you’d like to respond by delivering that new function quickly. If your approach to natural language uses a template, you would have to manually update each template to include the new feature. With our powerful analytics features, these updates can be made in minutes per dashboard, instead of weeks. For example, users now wish to see a compound annual growth rate (CAGR) in the story. In a template, you’d have to copy and paste the CAGR calculation in every template—and run tests to make sure it’s working correctly. Would it be better if your story already had a CAGR feature that could be quickly pushed to users? Of course.
7. Enterprise security and user governance is messy
Many of you may be wondering how all of this applies to on-premises deployments. This is another common question we address. As a head of BI for a large enterprise organization, you want to ensure that all users are understanding and regularly using your dashboards, so you introduce plain-English stories to help. In some instances, desktop users are looking for help understanding dashboards independently, utilizing templates and tools not built for the enterprise. How do you ensure they are using natural language tools appropriately? By providing desktop users access to the exact same stories used in the central server, organizations can be confident the information is accurate and secure.
Start Understanding Your Dashboards
Now that we’ve painted the picture of all the risks of using templates in dashboards, let’s summarize how Narrative Science addresses these points for our target market.
You invest in BI platforms to put insights into the hands of everyone — our extensions make sure this information is fast, accurate, powerful, and secure. We designed extensions to instantly provide users with the key takeaways, making your team superheroes in the organization. They will receive new insights instantly and have more time for proactive analysis. Best of all, viewers will have a better understanding of their dashboards, so they are more likely to actually use them! Our extensions for business intelligence illuminate dashboards for everyone, helping you drive better outcomes.