There are ~18M patents in force worldwide. With entries dating as far back as the 19th century and patent applications cresting at 4M annually, this raw IP dataset is arguably one of the world’s oldest, uninterrupted, active record of humankind's discovery and innovation – as well as a reflection of the economics and mores across the centuries.
Today’s story: Amazon IP
Taken individually, a patent (family) encapsulates an invention. Each patent is tagged with a collection of CPC codes that position an invention inside a rich classification system. The CPC (Cooperative Patent Classification) system is designed to categorize every possible innovation from mousetraps to quantum computers and it’s organized into sections, which in turn are sub-divided into classes, sub-classes, groups and sub-groups. There are approximately 250 000 classification entries.
Patent sentiment: (versus Twitter, etc.) In my last post, A Big Bang in the Patent Data Universe, I had hinted at how a deeper analysis of patent applications whose CPC codes included both machine learning (ML) and healthcare informatics revealed a rapid expansion of inventions taking on materially higher risk medical use cases that far outpaced the already significant increase in ML patents more generally. This kind of “sentiment” would probably be relevant to regulators, legislatures, investors, etc.
IP Strategy: Combining CPC codes with a patent's Assignee (patent owner) opens a new dimension of inquiry. Patents are not static – patents evolve - even after a grant. One of the most interesting examples of a post-grant update is reassignment (change of ownership). While there are many potential root causes behind a patent reassignment, one of the most common (and interesting) is M&A (mergers and acquisitions).
I’m going to focus on Amazon here for three simple reasons; AMZN are prolific patent authors, active in the M&A space, and I have no past or current business with them (being an Amazon Prime member doesn’t count, does it?). Let’s see what the public patent record can tell us about AMZN.
AMZN has submitted 6,873 patent family applications. Those families reference 159,733 unique CPC codes that fall into 249 CPC groups/sub-groups. Straight away we can see that each patent is likely relevant to multiple CPC categories (averaging 23 CPC codes per application) but as a group, relatively focused (a 27:1 patent to CPC group ratio).
Post-grant patent reassignments: AMZN is the current owner of 38 patent families where the original assignee was NOT AMZN. These acquired patent families reference 1,451 CPC codes that fall into 15 CPC groups. A quick observation might be that AMZN does indeed acquire patented IP. One might think this too obvious to call out, but I have heard more than one venture capitalist question the conventional wisdom that patents add value. Clearly, in some cases, it most likely did.
Taking the analysis to the next level, are their material differences between the natively authored patent dataset versus the acquired dataset? What can those differences tell us about AMZN competencies, M&A strategy, and AMZN’s overall corporate strategy?
The chart above shows the relative distribution of patent applications over the past 3 ½ years by CPC group.
Observation 1: given the massive gap in volume between Amazon originated patents versus acquired, I think it is remarkable that nearly 20% of acquired patents were in entirely novel CPC groups for AMZN (note, patents that were allowed to expire due to lack of maintenance payment are excluded from this comparison as one can assume AMZN didn’t see those patents as relevant to their business).
Observation 2: The 2nd greatest concentration of Amazon IP patents inside a single CPC group (H04L) represents 23% of its total patents (its top group sits a little higher at 28%) but that same group represents a whopping 73% of its acquired patents.
Clearly there is an area of embedded M&A interest. What more can we see?
H04L is defined as “TRANSMISSION OF DIGITAL INFORMATION” – that’s pretty general to say the least. We would then investigate the specifics of the CPC codes that are referenced in the acquired patents.
There are too many for this short post, but here are two representative examples worth calling out.
Could the data points and behavioral trends help to predict AMZN’s M&A strategy?
Or factor into valuation negotiations during negotiations or when raising capital?
Are there vulnerabilities in AMZN’s stated strategies that may be inferred through their behavior and/or gaps in their coverage?
What’s the point of all of this?
I have already written about the “buried gold” inside reservoirs of data sitting unmarked in organizations and businesses just waiting to be mined and refined for Machine Learning applications yet to be conceived.
Raw patent data may well be the motherload. …but there's a real issue. When held up to ML data cleansing and curation standards, raw patent data is very very raw indeed.
There are eighteen million stories in the naked IP. This has been one of them.
Again, with gratitude and apologies to the classic TV series, NAKED CITY.