Natural Language Generation (NLG) is a hot topic and seems to be popping up in increasing areas of our lives. Alexa, Siri, customer service chatbots, and automated text messages provide information about products and services, and more of our interactions with businesses involve some level of NLG.

Despite these well-known NLG success stories, using NLG for business purposes is still in its infancy, both in terms of the technology available and the proven use cases NLG has been successfully applied to.

Depending on how consumer-facing, customer service driven your organization is, as well as whether your organization is more bleeding edge or more wait-and-see when it comes to new business methods, it may be time to consider how you might leverage NLG.

In this post, we’ll start by defining what NLG is and isn’t and consider some of the main use cases for NLG out there. Hopefully, with a better understanding of both, you can make a more informed decision about using NLG.

What is NLG?

NLG is a subset of Natural Language Processing, which includes both Natural Language Understanding (NLU) and NLG. NLU is all about extracting data from inputs, whether through voice recognition (think Alexa as well as customer service lines where you speak responses), handprint recognition (for extracting data off handwritten correspondence), and intelligent capture (for extracting data off machine-printed forms).

NLG is all about output — automated communication with humans. These outputs can be simple, such as chatbots, or complex, like Alexa and other digital assistants.

On the simple end, NLG is a set of pre-programmed responses to if/then decisions (e.g., if the human says “1,” then read out the following text). On the more complex end, NLG is a set of algorithms that decide what output is most relevant based on inputs.

For example, an input can be an open-ended question posed to Alexa or, in the most extreme examples, a Jeopardy question that requires a response. In terms of AI, it’s a computer that can hold a human conversation on the fly.

From a business standpoint, most NLG in production today (other than digital assistants) is on the simpler end. The majority of NLG customer service interactions consumers have today fall into this category, i.e., when a chat window pops up on a website homepage asking if you’d like to chat.

Typically, the NLG ends where the conversation gets more complex, like once the first- or second-level of decisions gets made and fuzzier logic begins: “I want help restructuring payments on my existing mortgage,” or “I want to change an existing hotel reservation,” etc.

What NLG isn’t (yet) is a technology that can respond to user inputs in a meaningful way without significant levels of programming. AI promises to be able to do so at some point in the future, but for now, this is aspirational.

Key Current NLG Use Cases

Most of the NLG use cases in play today are basic customer service applications, such as chatbots. For instance, a person comes to a website and interacts with a chatbot to get basic information or direction on how to solve a straightforward if/then problem, e.g., “How do I reset my password?” or “When is my next payment due?”

NLG is not yet in wide production for more complex questions, e.g., “What is the right investment mix for me?” or “What is the right level of insurance given my risk profile?” or “What are my mortgage options given the house I live in?”

For these more complex questions, NLG would be used at the top level for triage: What big bucket topic area is this person interested in? Once NLG determines the answer to that simple if/then question, it can hand the person off to a live customer service representative to address their detailed inquiry. Eventually, NLG (particularly as enabled by AI) will be able to handle more and more of these queries without human input.

Beyond these up-the-middle use cases, there are more leading-edge uses that involve content authoring, such as fantasy sports. For example, after you complete a game, an NLG engine uses your game data to author an email with a play-by-play of your results, offering suggestions (including a heavy dose of sarcasm).

The Associated Press is also using NLG in certain cases to author content that is revised by an editor prior to publication to ensure readability and accuracy. Some corporations are using NLG to author site content, particularly as it relates to search marketing. (You can read more about these use cases here.)

However, these use cases are far more the exception now than the rule. For the majority of business applications, NLG is very much in its early stages.

The Net Net

The proven, battle-tested uses of NLG are too limited and the tech is far too immature for unbridled optimism. So, don’t get lured in by the hype and think that your organization can just up and roll out NLG to cut back significantly on human employees while supercharging the customer experience in the near term. All of this will likely be a reality in five to ten years, and will almost certainly be table stakes for how businesses interact with customers. For now, proceed cautiously and test the waters.

Joe Shepley is a VP at Doculabs and Practice Lead, focusing on developing’ InfoSec practice and its applications in a wide range of industries.