Highly differentiated startups raise 117% more funding

At best, it's a case of confirmation bias. The most highly-differentiated startups don't raise anything and no one ever hears about them. Crunchbase doesn't capture the majority of these, so the data used here isn't representative.

Then they use NLP to determine how different the products are.

Those of us who get pitched by SaaS vendors all the time know that two companies can do the exact same thing and have totally different product descriptions. Salesforce started talking about IoT for a while, and they still bill themselves more as an ERP when most people still use them as a CRM.

There's also the issue of those who pivoted. Most startups pivot, and the author admits to ignoring pivots altogether!

> "2. Found the websites of all these companies, in the year each startup was launched (using WaybackMachine, a historical archive of web pages)."

> "3. Scraped the text on these websites and ran them through a natural language processing (NLP) machine learning algorithm (doc2vec)."

> "4. Measured how different the startups’ value propositions were to those of competitors (e.g. focused on a niche product feature vs others that talked about price)."

I mean this makes sense, although I worry this is much too broad of a definition to glean data off of, especially when looking in hindsight when it's not as obvious how "differentiated" the company was when they first pitched compared to the rest of the market

I avoid using this kind of terminology in my writing because I'm never completely sure that my readers will share the same exact definition of how those numbers should be interpreted.