Natural Language Generation and Its Business Applications

Natural Language Generation (NLG)

As a continued exploration of AI Authors and Robot-Generated news, it is worthwhile to explore some of the technology driving these algorithms. AI designed to generate documents that read like a human wrote them rely on Natural Language Generation (NLG) algorithms. NLG algorithms are meant to automatically generate text from structured data that reads as though the generated text was written by a human author. Structured data are documents such as:

  • Product reports describing the features of a new product.
  • Survey results from an online customer satisfaction survey.
  • Financial reports (like those a robot author would use).
  • Personalized emails.

In other terms, NLG can be defined as collecting raw text data and turning them into a human narrative. NLG is related to Natural Language Processing (NLP), but works in the opposite direction. NLP collects data from text; NLG generates text from data. Or, NLP reads what NLG writes. The general order of operations for a NLG algorithm looks like:

  1. Data Collection: finding the right structured data to train on and choosing the right content to convey in the NLG output.
  2. Deciding Content: deciding what main topics will be conveyed and how to convey them.
  3. Document Structuring: outlining the content in the most coherent and “natural” way.
  4. Content Aggregation: creating sentences, putting similar sentences together, and adding references.

What must NLG decide?

As it moves through the steps listed above, a NLG algorithm must make two decisions: what is meant by “natural language” and how to create a “human narrative” using this natural language. To a NLG algorithm, natural language “looks like”:

  • Large, separate chunks of text regarding a single subject.
  • Complex sentences with consistent syntax.
  • Comprehensive structure with logical information flow.
  • References and analogies.

Next, a NLG algorithm must then decide how to compile its output narrative, or execute Step 4 from the previous section. Based on the context of the natural language input, the algorithm must understand how to best format its output to read like a human voice. It must decide:

  • The core idea of the text to generate.
  • The best structure, narrowed down over iterations, of the text narrative.
  • The sentence flow and word choices.
  • The right idioms / references / expressions.
  • The right syntax and voice.

NLG in business applications

The NLG “human narrative” output has many applications to modern business practices. This article will briefly describe two of them: reporting and marketing.


As discusses in the “Rise of AI Authorship” post, NLG together with NLP can be used to quickly compile easy-to-read reports for marketers based on complicated data. For example, NLG can provide a concise sales report for the previous week(s). NLG can also use personal email as a data source to then quickly and efficiently create human-sounding emails, saving users time spent formatting and wording outbound email messages.


In addition to quick reporting on data and formatting messages, the general consensus among users is that the main payoff when using NLG comes in the field of marketing. The applications of NLG specific to the field of marketing including:

  • Reducing blanket “one-size-fits-all” marketing materials and increasing customer-specific marketing materials automatically.
  • Quickly generating unique marketing content targeting a unique customer.
  • Quickly reaching a broader customer base by rapidly generating and distributing targeted marketing materials.
  • Increasing a customer’s loyalty by creating exciting content related to how that unique customer uses the company’s products and/or services.
  • Understanding metadata about a unique customer and tailoring product information to that customer.

In summary, NLG provides the other side to NLP. NLG AI takes structured data as input and outputs a text document. However, what makes NLG special is the way it outputs text such that the text seem human-authored. Many nuances exist in correctly operating NLG, and using NLG the “right” way isn’t always easy. However, the payoff NLG can have for a business is worth the risk. Modern-day businesses and marketers will have to understand NLG and how to harness it for their benefit to be successful in the digital age.

topic modeling for product managers

AI-enabled research management system for market-intelligence.