Rising Above the Noise — Why We Doubled Down on Anyword
By: Dror Berman, Rick Scanlon, and Aravind “Avi” Bharadwaj
Writing is hard. For centuries, people have struggled to find the perfect phrasing to achieve all sorts of objectives — be it to enthrall the masses or to sell knick-knacks.
Yet, effective writing has never been as crucial as it is now. As content turned digital and computers became ubiquitous, the cost of generating and distributing content became negligible. On any day, most of us are bombarded with orders of magnitude more content than we were merely a decade ago. As the noise increases, writers have to do a better job to capture our attention.
However, in today’s world of digital communication, we are all writers. We all create content, whether an email to a customer, a blog post, or a description for a new product. Most importantly, we need our content to rise above the noise and drive action.
That is where technology comes in.
Over the last few years, transformer-based language models have emerged with the ability to generate human-like conditional synthetic text of unprecedented quality. For instance, Google’s BERT and OpenAI’s GPT-3 have achieved strong performance on many datasets in tasks such as translation, question-answering, and novel content creation. However, these models are minimally-tuned on purpose to support a wide array of downstream tasks. As such, while these models create human-like content, they are not optimized for specific objectives.
For instance, below are two headlines about a kid’s scooter. Which do you think drives more engagement on a website targeting moms in the 25–40 age bracket:
- These Scooters Have Been Designed With Safety In Mind
- A Kid’s Toy You Can Trust Your Child To Ride Safe
While most experts think #2 will perform better at conversion, #1 performed 70% better in the real world. Up until now, marketing experts wrote well-thought out content based on their understanding of their audiences. But as we can see in the example above, this is an area where AI can learn from vast amounts of data and surpass experts’ capabilities.
Wouldn’t it be great if you could use AI to auto-generate content and predict its performance without having to A/B test 1000s of copies? Or better yet, automatically generate thousands of copies, each perfectly aligned to drive action with the context of a specific product and audience that reads it?
That is what Anyword does. Anyword is a natural language generation platform that generates predictive copy, fine-tuned to a specific objective. The platform leverages an ensemble of language models and learnings from content generated from over $250 million of ad spend. It provides predictive insights on the copy generated, optimizing content by target audience segments and channels (e.g., blog, ads, web page). Further, Anyword also offers additional features to make the process more efficient, including adherence to brand guidelines and workflow capabilities.
Anyword launched the content creation offering earlier this year and has rapidly scaled to become a world leader in this emerging field. Within a short amount of time, Anyword has attracted thousands of paying customers, including Fortune 100 companies, marketers, publishers, and small businesses — all using Anyword to create product descriptions, ad copy, and blog posts at scale.
Anyword is leading a revolution. In just a few years, AI generated content will be the new norm. We couldn’t be more excited to double down in leading this round and to deepen our partnership with this fantastic team.
