NLG in BI dashboards: Avoiding the illusion of insight
September 3, 2019 | Greg Williams

Just as when Gartner called Arria “a world leader” in NLG a few weeks ago, the firm also appears to have been correct in 2017 when it predicted that, by now, 90% of dashboards would include some form of NLG.

Today, plenty of NLG vendors claim to be able to narrate your dashboard. Business leaders should keep in mind, however, the possibility that bad NLG—narration that does not go beyond the surface visuals—is worse than no NLG at all. Mere textual decoration of charts and graphs can—in the best case—waste viewers’ time; in the worst case, it can lull viewers into a false sense of confidence that they fully understand the complex data set as a whole. This makes further investigation seem unnecessary, and effectively obscures critical insights that are not available at a glance anyway.

It is as if Gartner predicted that we would all soon commute without cars, and in response—to check that box—some companies bought bicycles, while Arria chose jetpacks. The bicycle might feel like progress, but in reality, its riders are being left behind. To get the most out of dashboard-NLG integrations, and to protect their organizations from the negative effects of the illusion of insight, business leaders need to choose the right NLG platform, and assemble the right subject matter experts and analysts to configure it.

Choosing the right NLG platform

As decision makers consider how best to enhance their organizations’ data sets and BI dashboards with NLG, it is important for them to choose a technology that does more than simply state the obvious about some portion of the surface visuals. Too often, we’ve seen Arria’s competitors offer narration along the lines of, “Sales of Product A increased during the third quarter compared to the second quarter.” Well, that’s not very helpful. You can tell that by glancing at the bar chart. The task is to go beyond the simplest of interpretations in order to uncover actionable insights that are available in a fully contextualized perspective on the data. Some of these insights may be represented in various visuals on the screen, but not always. The insights may also exist only in the underlying data, without representation on the screen at all. What you want to know is how this quarter’s sales of Product A compare to the long-term trends across multiple dimensions, whether geographic, seasonal, managerial, and so on. What changes should you make in order to increase sales of Product A in the future? Or should you begin winding down Product A, and concentrate your efforts on Products B, C, and D?

Similarly, simple high-low statements, such as, “The best-selling product during the third quarter was Product A, which sold $1,000,000 worth, while the worst-selling product was product B, which sold $25,000 worth,” are of little value, and should not “check the box” on adding NLG to the data sets or dashboards.

In our view, the addition of this kind of rudimentary language to dashboards is not a neutral move but a negative one—and not just because you will have invested time and money to produce statements that provide no additional insight. The longer-term danger is that simple statements can act as mortar, turning the surface visuals into a brick wall separating your employees from insights to be found in the relationships between data sets represented visually, or within the underlying data that might not be represented on the screen. Having seen the visuals, and having read the captions, your employees might believe that their understanding is comprehensive, that they’ve taken a full dose of medicine when in fact they’ve swallowed a placebo. Without Arria, you would be better off taking away the false-confidence pill of other vendors’ simple narration, because then at least there’s a chance that someone will look beneath the rudimentary charts and graphs.

Arria’s technology, by contrast, can look at all of the underlying data and describe it in natural human language. Also, critically, Arria’s technology can be deployed entirely on-premise, relying upon no calls to external servers.

Configuring the implementation team

The natural inclination is to view an NLG project as a technological endeavor, and indeed there is nothing wrong with getting technologists involved with your Arria implementation. But as we’ve argued before, since writing is the final and only expression of a Natural Language Generation project, it is also important to involve writers and subject matter experts. For various reasons, these employees should embrace the opportunity to multiply their output and increase their importance to the company. Keep in mind that because Arria NLG Studio is a standalone tool that anyone can use, these traditional writers and analysts will be expanding their technical skills to a technology that is broadly applicable within the firm and highly regarded across a wide range of industries.

Conclusion

Narrating your BI dashboard is not a check-the-box exercise but a real opportunity to wring more power and insights out of the dashboard, and add efficiencies to the organization. Business leaders seeking to add NLG to BI dashboards must avoid creating the illusion of insight, inadvertently discouraging further investigation of the underlying data. For the most effective integration of NLG and BI dashboards, leaders should start with Arria’s NLG platform, and should involve traditional writers, analysts, and subject matter experts to produce prose and insights that reflect the firm’s language, sensibility, and business priorities.

Subscribe to the Arria Blog