How NLG Supports Marketing

4 min read
How NLG Supports Marketing

Marketing tools based on NLG (Natural Language Generator) are pushing the boundaries of what’s possible in the field of content creation. The goal of modern marketing is to provide recipients with personalized, useful content for their specific needs. NLG algorithms are in turn an engine for the automation of marketing processes, streamlining work on repetitive tasks. Here’s how they’re changing the way we communicate with consumers.

First - Content

Content is a key element of communication with customers online and the basis for building marketing automation. Well-prepared content will not only improve customer relations and help in the construction of brand identification, but will also step in when technology fails.

Annoyed when an error page pops up? A clever text might take the edge off for customers. Using the NLG network in this case will help you create different page views of the error page, depending on which stage the client is in.

Eror 404 page example

Also, thanks to systems for marketing automation based on artificial intelligence, we can learn more quickly what combination of content will best suit a specific segment of customers and automatically display the right content for them.

However, the greatest support for marketing automation is the use of Natural Language Generation (NLG), which helps in generating reusable content based on a structured data set. This means that with the right data, NLG can automatically convert numbers from a spreadsheet into a data-driven narrative.

Second - Time


Despite the many advantages of using natural language, such as automatically generating editorial texts, personalizing unique content for customers so that they feel like an individual, serious challenges remain. The main one is the investment of time and money in the NLG implementation process in the company.

All data must be identified and cataloged so that the algorithms interpret it correctly. Therefore, before starting activities, you have to ask yourself three questions:

  1. Does your text focus around similar subjects and types of messages? – press releases, product descriptions, coupon text, etc.
  2. What format is your data in? It needs to be unified in a single format, preferably CSV
  3. Do you have the necessary financial and talent resources to prepare and support the implementation of NLG in your business?

And most importantly, artificial intelligence must work together with intelligent and creative people, since only such cooperation will bring good results. Companies will continue to improve texts and tailor them to their needs. Because, as Paul Roetzer, the founder of the Marketing Artificial Intelligence Institute, says, you need a human to use NLG systems:

Natural language generation (NLG) has very narrow applications in marketing today, and, in most cases, it’s enhancing what marketers are able to do, not replacing them.

Third - Custom Made Solutions

If you already know that NLG is for you, it’s time to plan the process of implementing the alogirthm in your business. Start by analyzing how long it currently takes to create reports, descriptions, articles and any other content made by the marketing department.

This will be the starting point to optimize the creation time of NLG.

The Arria NLG Engine

Once you have chosen the right system and prepared the correct data for it, it's time to start the NLG content generation process. It goes through several stages:

  • Data Analysis – on the basis of the presented data, the process of searching for the most important key facts begins.
  • Data Interpretation – on the basis of the content analysis, the program searches for repeated patterns and draws concusions about their meaning. The interpretation of the data will be faster if focused on a single category of test data, like product descriptions from one broad category.
  • Document Orchestration – at this stage, the narrative for the content and structure of the document is analyzed and created.
  • Microplanning – time to create preliminary sentences and choose the right words and descriptions for the document.
  • Surface Realization – the last stage focuses on the accuracy of the created content in terms of grammar and punctuation and is generally consistent with standard usage.

Content that passes through all five stages should be ready to go live, but can always use a final look over. 

Popular NLG Programs

NLG programs do more than create content. They can generate financial reports, convert speech to text and vice versa and much more. The most popular and most interesting algorithms include:

  • Creative stories - Quill is a program developed by Narrative Science to write stories. Based on automatically selected data, he starts writing content in the form of a short story. Interestingly, the content is different depending on the region where the recipient is.
  • Azure Cognitive Services - Microsoft provides this service, based on artificial intelligence, which includes several solutions, such as recognizing emotions, face search in pictures, changing words to text and Custom Recognition Intelligent Service - that is, advanced recognition of speaking and vocabulary.
  • Some large companies may decide to build their own NLG engines. For example, The Washington Post created Heliograf and used it to write stories about the Olympic Games and more than 500 press releases during the 2016 election in the United States.


Marketing automation effectively manages repetitive processes like sending emails, notifications on store websites and product recommendations. The next step in facilitating work by AI is the implementation of NLG algorithms that increase the efficiency and productivity of marketers by accelerating the content creation process.

However, it is worth remembering that a good weapon is not enough—you also need a soldier. In this case, that means a good marketer. This is because a good generator will not perform miracles and create quality content if you do not receive valuable data. Regardless of the sophistication of your artificial intelligence algorithm, you still need a competent person overseeing the process.