The Patent Signal: What Filings Quietly Reveal

Patents are a lousy way to learn about products. They're obfuscated on purpose, written in language designed for lawyers, and a third of them never become anything real. As a foresight signal, though, they're one of the best we have.

Here's why, and how to read them without drowning.

Why patents lead

A company files a patent because it expects the underlying technology to be commercially valuable in the next 2–4 years. The filing date isn't when the product ships; it's when the legal team thinks the science is solid enough to defend. Patents lead products by roughly the time between those two milestones.

This makes patent volume a near-leading indicator of where companies are betting. By the time a product hits the market, the patent was filed years ago, and follow-on patents in the same family are already disclosing the next generation.

What to count

Three signals are worth tracking, even informally.

Filing volume per CPC class. Patent offices classify filings into hierarchical categories. Watching a specific CPC class grow tells you which technical area is heating up. The class for "neural networks applied to image generation" rose 4x between 2019 and 2023.

Assignee concentration. Is the growth coming from many companies (broad industry interest) or one or two (a corporate race)? Both matter, but they mean different things for strategy.

Geography. When filings in a specific class start coming from a country that wasn't previously active, the field is being commercialized somewhere new. China, Korea, and Israel often show up early in emerging hardware classes.

What patents hide

Three big caveats:

• Defensive filings. Many patents are filed to block competitors, not to ship a product. They look like investment but aren't.
• Open source moves first. In software, the most important innovations now often happen in open-source repos before any patent exists. Patents in software lag rather than lead.
• Geography distorts the signal. Different countries patent at different rates for cultural and policy reasons. Don't naively compare absolute counts.

The pairing that works

Patents alone are noisy. Patents + academic papers + GitHub activity, taken together, are surprisingly informative. Each source has a different bias. Triangulating across them gives you a cleaner signal than any one source alone.

For most strategy work, you don't need to read patents. You need to count them, classify them, and watch the curves. That's a weekly cron job, not a literature review.