How is Jasper.ai Worth Billions, When The Underlying Tech Isn't Proprietary

Jasper.ai raised $125 Million in the summer of 2022, which valued the two-year-old startup at $1.5 Billion. This newly minted unicorn uses GPT-3, a neural network machine learning model created by OpenAI, to help content creators write blogs, product descriptions, and ad copy. There are dozens of companies using GPT-3 for similar purposes (copy.ai, wordhero, and rytr just to name a few), which makes Jasper.ai somewhat of a commodity. How is Jasper.ai worth billions, when the underlying tech (GPT-3) isn't proprietary?

Essentially anyone with an OpenAI API key and a fundamental understanding of HTML, CSS, and Javascript could spin up a Jasper copycat product. To prove this isn't hyperbole, I built a tool that you can test drive. It doesn't have the slick UI and packaging that Jasper.ai has but it is generative ai, capable of writing blogs on prompt exactly like Jasper.

The steel man counter argument is that commoditization and a lack of differentiation are not the same things. Take bottled water for example. Water is arguably the most commoditized natural resource in the world, but that didn't stop Mike Cessario from building Liquid Death, a canned-water company, into a billion-dollar brand through near-perfect market positioning and branding. Bottled water is a commodity, but Liquid Death is a highly differentiated brand with a ton of enterprise value. The same could be the case for Jasper.ai. Let's discuss their differentiating factors.

What makes Jasper.ai unique?

There are many applications built on GPT-3 that serve a variety of use cases. To customize the model to a specific use case, you have to fine tune it, or train it on a particular data set. Take KeeperTax for example. KeeperTax fine tuned GPT-3 to classify bank transactions and identify write-offs for gig workers and freelancers, making it easier for them to file their tax returns. Fine tuning is the first layer of differentiation for applications built on GPT-3.

Differentiation Through Fine Tuning GPT-3

According to Jasper.ai, they fine tuned the model by making it read 10% of the internet (presumably 10% of all blogs, ad copy, and marketing related content available on the web). If that is indeed the case, this would be a great example of how product differentiation could be achieved through fine tuning a pre-trained model. However, I have two objections.

Copy.ai and Jasper.ai (and my sandbox application) produce very similar outputs, so even if Jasper was fine tuned on 10% of the internet, it is not abundantly clear that the fine tuning created a competitive advantage.

The second objection is that OpenAI, the makers of the GPT-3 protocol that serves as the base layer for all of these generative AI applications, continues to train the model. In a recent interview on the AI in Business podcast, Peter Welinder, VP of Partnerships at OpenAI said this when referring to a Swedish company that uses GPT-3 to train employees, (paraphrase) "They didn't need to customize our latest models. It works great right out of the box." In other words, the latest model improved upon the 3rd-party fine tuning -cannibalizing away any advantage gained through customization. GPT-3 is already a great copywriter. Even if Jasper.ai improved upon the model by fine tuning it with more data and marketing content examples, the gains will be short-lived as GPT-3 continues to improve.

Differentiation Through Prompt Engineering

Prompts are inputs that guide GPT-3 to produce a useful output. "Write a blog outline about rent control in New York City" is an example of a prompt. Given this input (prompt), GPT-3 knows exactly what you want. The keywords "blog outline" and the qualifier "New York" give GPT-3 enough information to return a thorough outline. Now imagine if you prompted GPT-3 with simply, "rent control." It won't know what you were asking for. Do you want a headline about rent control, a definition, or a random fact? As a rule of thumb, the better the prompt the better the result.

The goal of prompt engineering is to design a string of inputs that tells the model exactly what you want. Companies like Jasper use prompt engineering to create templates that users can deploy to get results quickly. A template example might be to write a product description for XYZ product on LinkedIn aimed at entrepreneurs. XYZ product helps you automatically categorize your expenses and find tax deductions. Jasper.ai users can simply swap out the variables to generate personalized ad copy for the LinkedIn platform.

The byproduct of prompt engineering is a good template. Script templates however, by nature of their shareability and replicability, are not strong enough differentiators for a billion dollar AI company. Once a winning prompt template emerges, it will be quickly copied - cannibalizing away any advantage gained through prompt engineering.

Differentiation Through UX and UI

The Jasper.ai user interface is superb. It is well designed, intuitive, and subscribers can go from notices to pro-users after watching a few short tutorials. The templates help quick start users with prompt recipes to speed up the content creation workflow. Jasper.ai also integrates with plagiarism checkers and writing enhancement tools like Grammerly to ensure quality. And lastly, Jasper.ai has a thriving community of 50,000+ members who share best-practices, pro-tips, and templates on the forums. All of these factors coalesce into a wonderful user experience on Jasper.ai.

The Jasper.ai UX and UI combination is a formidable differentiator. If you couple that with Jasper.ai's brand positioning, it is easy to see how Jasper.ai can become an indispensable tool for content creators.

There are several examples of multi-billion dollar companies that use nonproprietary technology. Uber is a classic example. Uber relies heavily on Google's mapping technology. They pay millions of dollars a year for access to the Google Maps API.

According to a WSJ article published in Oct 2022, Jasper.ai has approximately 80,000 paying subscribers and generated $35 Million in annual revenue last year. Jasper.ai expects to end 2022 with $75 Million in revenue, according to CEO Dave Rogenmoser. That means that investors paid a multiple somewhere between 20-45 times revenue to buy shares in Jasper.ai, which by all accounts is a massive premium. To understand why Jasper.ai is worth so much, we discussed Jasper.ai's differentiating factors and then opined on the sustainability of each one. My position, in conclusion, is that the only differentiating factors that Jasper.ai has are its UX and UI design and its strong brand. I don't see how Jasper.ai lives up to the $1.5 Billion valuation from a unit economics standpoint. Investors overpaid for this one.