I’ve spoken before about building a context layer as a first step in designing NLG. In this post, I’ll discuss the difference between static and dynamic context, and what that means in terms of the automated content you want to build.
In verticals like sports, the context layer can remain somewhat static. The rules of the game rarely change, and what the end customer (the fan) wants from the result (the article) is pretty much the same from article to article.
That’s not to say that the purpose of the automated content can’t be different from one sports article to the next. One series of articles about the NFL may talk about betting lines, another about fantasy, and another may just be an old-fashioned game recap.
But within each of those types of articles, the context layer is rarely going to change. What makes a good sports gambling bet — the betting line, the makeup of the team against the opponent, the location, injuries, etc. — those components don’t change from one game to the next or one year to the next.
The rules are locked in place and for the most part so is the context layer. Even within a vertical as loosely defined as finance, once you get an idea of the requirements of the article, whether it’s a personal finance report or a public company quarterly earnings report or a real estate loan rate report, the context layer can be built in a kind of set-it-and-forget-it style.
But what about something way more dynamic like Business Intelligence? Even the term itself isn’t strongly defined, so let’s get as rigid as we can up front. Business Intelligence is about analyzing known data in order to make an educated guess about future unknown outcomes.
In other words, if we take what we know about the sales of widgets in November — who bought them, when they bought them, how many at a time, through which channels, and so on — we can make some smart assumptions about how many widgets we’ll sell in December.
We also have to consider seasonality and history. If the global widget market is growing overall, we might expect our targets to grow at the same rate, if not more. And we should look back at last December, and maybe also look ahead to the fourth quarter and the year as it comes to a close.
Then we can start making decisions about labor, production, marketing, and so on, for the next 30 days to five years. The more we know, the better our decisions, the more efficiently our business runs, the more revenue and profit we realize.
Now here’s the thing.
Every business unit has its own set of goals and metrics.
In sports, you win by scoring more runs, points, or goals than the other team. In business, you win by hitting the goals you establish as critical to the survival and growth of the company.
Papa John’s, Domino’s, and Pizza Hut all sell pizza. I assure you that they all have dramatically different goals. Furthermore, those goals are segmented along the units of each business. Marketing has different goals than IT, who has different goals than Operations, and so on. And finally, those goals can be wildly different depending on any number of external or internal factors: Time of year (as I stated above), location, market forces, all of that comes into play when “doing” Business Intelligence.
It would be sheer disaster for Papa John’s to analyze and execute to Domino’s goals. It would be apples-to-oranges for Papa John’s marketing to try to live up to Papa John’s operations goals. And it would be a circus of bad assumptions for Papa John’s sales to use NFL in-season data (when they sell the most pizzas) against April goals.
This is why Business Intelligence platforms start and end with the ability to customize. Business Intelligence platforms that equip their users with NLG must do the same.
When Automated Insights took on the Business Intelligence vertical, we applied the some of the same principles we used to build Wordsmith, namely, make Natural Language Generation accessible and customizable for everyone who isn’t already a data scientist.
One of the great debates of automating content over the last seven years has been whether NLG is a job-taker or a job-enhancer. In all cases, the latter has been true. NLG takes the data science out of broader sciences, like journalism, finance, and marketing, and allows the experts in those fields to discover and report their findings faster and more efficiently.
Journalists are freed up from number-crunching to be able to dig deeper into the plain-English insights NLG pulls from those crunched numbers. Financial advisors can spend more time working with their clients on needs and goals. Marketers can execute more quickly in an industry that puts a premium on speed.
Wordsmith for Business Intelligence integrates directly into the top BI platforms, like TIBCO, Tableau, and MicroStrategy, turning dashboards into narratives in real time. This integration allows for a completely dynamic context layer, one that changes not only with the rules of the data, but the goals and metrics as defined by the user.
Further, the context layer can be customized either by the users themselves, or by our team via managed services. In either case, the narratives are driven by the same data as the dashboards. The goals and metrics are unique to the business, and even the lexicon is independent from one instantiation of Business Intelligence NLG article to the next.
With a truly dynamic context layer, truly useful Business Intelligence NLG is possible. It’s like having a data scientist sitting with you, writing plain-english reports with every parameter you tweak and view you change.