The worst thing about patents is the SLOW feedback loop. The field doesn’t evolve quickly, bad policies and inefficiencies linger, and there’s almost no room for real experimentation. You can file something today and not learn whether it mattered until long after everyone involved has changed jobs.

How do you get better at something when the system hardly tells you what worked?

The closest thing to a feedback loop appears in litigation and licensing. Licensing is mostly hidden behind NDAs, but litigation is public and surprisingly rich. At Meta, I had access to LexMachina (a great litigation analytics tool) and finally could close the loop on many lingering questions on what “worked”.

I analyzed ~100 asserted patents from the highest-damages cases. Not perfect - some of the strongest patents never reach trial due to early settlement - but litigated patents are, on average, much sturdier than the millions that never see a courtroom. After all, people are spending a lot of money to fight over them.

A few interesting (to me) takeaways on these 100 asserted patents:

• 50% were original filings, 50% were CONs
  • I expected more CONs due to targeted/broadening CON practices
• Many of the inventors had 5+ patents to their name
  • Treat your prolific inventors well!
• Most patents were organic and not acquired
  • Damages stories resonate more when they trace back to the original team
• ∼150 words is the sweet spot for granted independent claims
  • Thus, filed independent claims should be ~100-120 words
• Avg office actions to allowance was under the overall PTO average
  • Faced less trouble during prosecution, thus foundational ideas are worth more than incremental ones
• Universities were well represented
  • They patented more foundational technology that created entire industries

Useful signals, but still only a hundred data points.

Now, thanks to good-ol AI, we’ve assembled a list of all 21,000+ litigated patents in the last 5 years. We can extract all the properties we’ve ever wanted to know:

• claim count & length
• patent age and case timing
• # OAs to allowance
• # forward citations
• whether the plaintiff is an NPE, operating company, or university
• the relationship between plaintiff and defendant (SEP, vertical relationship, direct competitor, etc.)
• technology category
• whether the patent’s idea emerged early in the industry lifecycle (foundational), mid-curve, or later-stage

What else should we measure? Please let me know. If the data exists or if AI can infer it, we can pull it into the pipeline.

We’ll publish the results to give practitioners a sharper intuition around what makes a patent valuable.