Ranking Patent Assignees by Filing Volume: How the Tables Are Built

A ranking of patent assignees looks like the simplest table in the record: count the patents, sort the names, print the result. Almost every step in that sentence hides a decision, and the decisions change the order.

What is actually being counted

The first choice is the unit. A ranking can count granted patents, published applications, or patent families, and the three produce different tables. Applications measure activity, since they include filings that will be abandoned or refused. Grants measure what survived examination. Families collapse the continuations, divisionals and foreign counterparts of a single invention into one entry, which measures inventions rather than documents.

Most published league tables of United States patenting count grants in a calendar year, because grants are public, final and easy to date. That is a reasonable convention, but it is a convention. An organisation that files heavily in continuations will look larger in a grant count than in a family count, and the difference is not a matter of opinion about which is right.

Which date decides the year

The second choice is the date. A patent carries a priority date, a filing date, a publication date and a grant date, and a ranking for “2015” can mean any of them. Grant-year tables are dominated by examination backlogs: a surge in filings shows up in the grant figures two or three years later, and a slowdown at the examining office can shrink every assignee’s total at once without any change in research activity.

Filing-year tables track the underlying work more closely but are never final, since applications filed in recent years are still being examined. A filing-year ranking for the latest years is always an undercount, and it is undercounted unevenly, because some technology areas move through examination faster than others.

The name problem

The third choice is the hardest. The assignee field records whatever name was on the assignment at grant, typed as it was typed. One company can appear under a dozen variants: with and without “Inc”, under a former name, under a subsidiary created for a particular product line, or under a holding entity set up to own intellectual property. Two unrelated organisations can share a name closely enough to be merged by careless matching.

Any ranking therefore depends on a disambiguation step, and the quality of that step sets the quality of the table. Large public datasets such as PatentsView, the United States Patent and Trademark Office’s research data platform, publish disambiguated assignee identifiers precisely because raw name matching is unreliable. Even disambiguated data has to settle questions that are not purely technical, such as whether a subsidiary acquired last year belongs with its new parent for filings made before the acquisition.

This archive faces the same problem on a small scale. The assignee listing groups names on a best-effort basis, and it is explicit that two entries which look distinct may be the same organisation.

A small example

Sorted by record count, the holdings here put the University of Connecticut first with 23 records, followed by Sega with 10, Sun Microsystems with 6, Amazon Technologies with 4 and Apple with 3. Read naively, that is a ranking of patent activity. It is not. It is a ranking of what this archive has compiled, which is a sample, and the order mostly reflects how the sample was assembled.

It also illustrates the name problem in miniature. A university files under one name; a technology group files under several, and a ranking built from raw names spreads the corporate holdings across entries that each look smaller than the whole. Why an academic portfolio behaves this way is covered in the piece on universities as patent holders.

Assignee is not owner

A further caution applies to every assignee ranking, however carefully built. The assignee on the face of a granted patent is the owner at grant. Patents are sold, licensed exclusively, pledged as security and transferred in acquisitions, and none of that changes the printed document. A table of grants by assignee is a table of who held rights at a series of moments, not of who controls a portfolio today.

The distinction between the person who made an invention and the party that held it at grant is set out in more detail in assignee or inventor: who actually holds a patent. For a ranking, the practical consequence is simple: reassignment records are a separate source, and a table that does not consult them is describing the past.

Volume is not value

Counting treats every patent as equal, and no one who reads patents believes they are. A single foundational filing can matter more than a hundred narrow refinements, and organisations differ widely in how readily they file. Some patent nearly everything that might be defensible; others file sparingly and rely on trade secrets or speed to market. A volume ranking rewards the first habit and says nothing about the second.

Analysts sometimes weight counts by citations received, by family size, or by renewal, on the reasoning that an owner who keeps paying maintenance fees thinks the patent is worth something. Each weighting introduces its own assumptions, and none recovers the plain question of which inventions mattered.

How to read a ranking

The same habits apply to any ranked list, not only patent tables. Lists of the best gambling sites Canada readers are shown in newspaper guides are a familiar consumer example: the order only means something once the method behind it is clear, which is why a careful reader of the Ottawa Citizen’s gambling site picks looks first at how the paper explains what they recommend and on what criteria, before looking at who came first. A patent league table deserves the same treatment.

In practice that means asking five questions of any assignee ranking before trusting its order:

• Unit. Grants, applications or families, and whether design and plant patents are included alongside utility patents.
• Date. Which of the document’s dates assigns a record to a year, and whether the most recent years are still incomplete.
• Names. How variants and subsidiaries were grouped, and whether the grouping is published so it can be checked.
• Coverage. Whether the source is the full corpus or a sample, and how the sample was chosen.
• Ownership. Whether reassignments after grant were taken into account.

Where rankings remain useful

None of this makes rankings worthless. Within a single method, held constant across years, a ranking shows direction reliably: an assignee climbing steadily through a consistently built table is filing more, whatever the absolute figures mean. Comparisons across methods are where rankings mislead, because two tables that look alike can disagree for reasons that have nothing to do with the organisations in them.

The most informative use is usually the narrowest. Take one assignee, sort its records by filing date, and read the sequence. That turns a position in a table into a research history, which is the thing a ranking was trying to summarise in the first place.