How are NLG applications solving real marketing challenges?
Read the full article below.
As soon as you hear the words “artificial intelligence” or “machine learning” your imagination might run wild.
Can intelligent machines make decisions on their own?
Can they magically solve business problems?
What happens when machines start doing or saying things that we can’t predict?
Misunderstandings about what AI can and can’t do have sometimes fueled unease or opposition to technological applications that are augmenting human capabilities and empowering teams to operate with greater efficiency.
At Phrasee, we have built an advanced artificial intelligence system that writes short-form marketing copy like email subject lines, Facebook ads, display ads, SMS messages and push notifications. It’s an AI-powered solution for a well-defined challenge – writing optimized language at scale and in a brand’s voice that engages an audience and boosts marketing ROI.
Our language technology (a powerful combination of natural language generation and deep learning) generates and optimizes language for a very specific purpose. This is known as “narrow AI”, since it is not attempting to solve the general problem of human intelligence.
Artificial intelligence is not magic. It is a scientific answer to real-world business challenges like stagnant engagement rates and declining revenue.
By getting the facts straight and learning more about AI and its subset technologies, and what they are currently capable of, marketers will benefit from learning more about how AI can be used to improve marketing and maximize business impact.
What Is Natural Language Generation and How Is It Being Applied in Marketing?
Language is complex.
By diving into the 20-volume Second Edition of the Oxford Dictionary, you will find that it has full entries for more than 170,000 English words in current use. In English alone, there are billions of ways to string together words to convey a message, and all languages follow their own complex set of grammar rules.
When launching a marketing campaign, not only do you need to target the right people at the right place and time, but there is another important piece to the puzzle…the right message. What are the chances that you choose the right combination of words and linguistic properties to engage your audience?
Natural language generation (NLG) offers a solution to this problem. It’s a sub-field of natural language processing (NLP), which is software that processes human language in either written or spoken form. NLG gives computers the ability to write English (or any other language) with the intention that will be read by people.
Present-day NLG applications cover a range of use cases in areas such as journalism, business and financial reporting and the writing of product descriptions. For the marketer, the technology makes the writing of short-form marketing copy more effective. In the case of email marketing, for instance, NLG can write subject lines that are tailored to a brand’s audience, and optimized for maximum engagement.
NLG systems can adapt the tone, style and structure of the output language to a brand’s audience, context, and narrative purpose. It is a complicated task that takes into account multiple aspects of language such as grammar, syntax, sentiment, word usage, and perception.
This is what Phrasee’s technology does. Our state-of-the-art NLG system can generate millions of human-sounding, brand-compliant copy variants at the touch of a button. And it’s helping world-leading brands like Virgin Holidays, eBay and Domino’s generate millions of dollars in incremental revenue.
NLG therefore offers brands a practical marketing application and commercial utility by powering optimized customer communications at scale.
Misconceptions About AI and NLG
Marketers have a lot of misconceptions about what AI can and can’t do.
A great example is Microsoft’s infamous Tay bot on Twitter. The bot, which was powered by NLG, was supposed to be a demonstration of how AI could learn to engage with users online. Instead, it ended up saying things that were racist and inflammatory.
AI can’t invent bad language like this on its own. Many AI systems use a technology called “machine learning”, where the system builds models based on existing data. In the case of Tay, the systems was trained on data that already contained the racist and inflammatory language. In other words, Tay learned how to mimic the bad behavior that already exists in a subset of internet users. The designers should have prevented undesirable content from being included in the training data in the first place. If they had, it would have been impossible for the system to learn to behave the way it did.
Marketers, however, might worry that AI-powered machines will fall into the same trap as Tay. In reality, for most NLG-based solutions this is very unlikely. The important thing to remember is that AI systems are constrained by the data they are trained on. For example, Phrasee’s models are not trained on random content scraped from the internet. They are trained on data sets that are carefully curated (by hand) specifically for the problem at hand.
Another misconception is that an AI system using NLG will eventually put everyone who writes for a living out of work. The assumption is that an AI system can do everything a human writer can.
Make no mistake, AI can do a lot. However, modern AI systems excel at narrowly defined tasks. In the case of Phrasee, our system specifically generates short-form marketing copy. In some situations, it is better at this task than humans. However, we can’t flip a switch and use the technology to write a blog post or an ebook.
Artificial intelligence and NLG have evolved and improved dramatically in the last two decades. But, today, most AI tools do just a couple of things well. They’re not general systems that beat humans at everything.
As marketers learn more about AI, they should keep these two facts in mind:
- AI systems based on machine learning reflect the data that they are trained on. This is a double-edged sword. On one hand, they can be powerful since they can access information from vast data sets. On the other hand, bad input data leads to bad output. When AI vendors, such as Phrasee, take care with their training data there is no risk that the systems will behave inappropriately.
- The AI systems you encounter can’t do everything equally well. Rather, each system excels at a few narrow tasks.
By keeping those two facts in mind, marketers will find themselves better equipped to understand—and vet—AI-powered solution providers.