But for those ideologues who are trying to convince themselves, and others, that software-implemented inventions are simply no longer validly patentable, I am afraid I have bad news. I have analysed US Patent and Trademark Office (USPTO) data covering millions of patents issued over the past decade, which shows that patents covering computer-implemented inventions, outside the particular fields of ecommerce, business and finance, not only are not in any form of decline, but have never, at any time over the past decade, shown any significant difference in grant rate from the bulk of patents issued by the USPTO across all fields of technology.
Patent claims classified by the USPTO as relating to ecommerce, business methods and finance, are another matter. The Supreme Court’s Alice decision, issued in June 2014, led to an immediate, dramatic and sustained fall in the number of patents being granted in these fields of endeavour.
Interestingly, the top US Court’s previous pertinent decision, Bilski (issued in June 2010, shortly after I started this blog), had no noticeable impact on the rate of grant of business method patents. This is, perhaps, unsurprising given that Bilski was the result of an appeal against a rejection of the application at issue by the USPTO, which had already been upheld by the Court of Appeal for the Federal Circuit (CAFC). Thus it did not represent, or require, any significant change in policy or practice at the Patent Office which the Supreme Court confirmed had been right to reject Bilski’s application. To the extent that Bilski changed anything with regard to business method patents, it was following the CAFC decision, rather than that of the Supreme Court.
In fact, the last event prior to Alice that had a significant effect on rates of grant of business method patents was the 2009 transition from Jon Dudas to David Kappos as Under Secretary of Commerce for Intellectual Property and Director of the USPTO (via John Doll, who was acting in the role between Dudas’ resignation on 20 January and Kappos’ confirmation on 9 August). Granting of business method patents had been in decline under Dudas, however this trend was soon reversed under Kappos, with issue of business method patents once again tracking rates across all technology areas by the first quarter of 2010.
My analysis also shows that the CAFC’s Bilski decision and the early months of the Kappos era are associated with an increase in the average word-count of independent claims in granted business method patents, and also in patents directed to other forms of computer-implemented invention, along with a significant jump in the average number of claims in issued business method patents. These effects are almost certainly a result of changes in examination practice at the USPTO, likely influenced at least in part by Bilski.
Source DataThe USPTO has a wealth of data available for those of us who wish to analyse patent statistics. The PAIR (Patent Application Information Retrieval) Bulk Data (PBD) set comprises over 9.4 million records containing bibliographic details of patents and applications in the Public PAIR system. The USPTO claims that PBD is complete back to 1981, with some data dating back as far as 1935. (My own review of the data suggests, however, that the data for 2016 is currently incomplete, with only about a third of the total number patents issued during the calendar year currently accounted for. I have therefore not included figures for 2016 in the following analysis.)
The PBD set includes application, publication (where relevant) and issued patent numbers and dates, along with identification of the USPTO Art Unit to which the application was allocated, the primary US Patent Classification assigned, and much more. The entire dataset is available in either JSON or XML format. I downloaded the zipped XML archive, which was 4.9GB in size and took about an hour. It unpacks to a set of XML files (generally one per year of data) varying in size up to about 7GB for the 2011 data set, and comprising about 105GB in total. (Just to jump on a hobby horse for a moment, this is not ‘big data’. It is certainly awkwardly large data, in that some care is required to manage and process it efficiently, but it fits comfortably on the dual 1TB hard drives of my not-too-obsolete HP laptop, and can be processed using conventional techniques.)
Another very interesting data set, which has been made available by the USPTO Office of the Chief Economist, is the Patent Claims Research Dataset. The dataset comprises six data files containing individually-parsed claims, claim-level statistics, and document-level statistics for all US patents granted between 1976 and 2014 and US patent applications published between 2001 and 2014. The claim-level statistics include counts of the number of words and characters in each individual published claim, as well as a flag identifying independent claims. The zipped data files vary in size up to about 4.5GB, and unpack to a total of about 60GB.
Processing the Raw DataTo cut a very long story short, I developed a program (in Python, for those who care) to parse, extract and clean relevant records and fields from the PBD XML files for 1974 onwards, and uploaded the six Patent Claims Research data files and my resulting PAIR data file into corresponding tables in a PostgreSQL database (running under FreeBSD in a VirtualBox VM on my same HP laptop, again for those who care).
I should note, for anybody thinking of trying to replicate this effort, that cleaning up the Patent Claims Research Dataset for my SQL database actually took a few days of work. If you know something about relational databases, you will understand that I wanted the ensure that the combination of patent (or application, or publication) number and claim number, in each record was unique, so that it could be used as a primary key, and a means to relate records in the claims datasets to corresponding records in the PBD set. However, due to errors in OCR, automated claim parsing, and who-knows-what-else on the USPTO side, this is not actually the case in the downloadable dataset. My ‘cleaning’ efforts involved a mix of Unix stream editing on the original CSV files, and identification and elimination of duplicates in SQL, before adding the key constraint.
The end result, however, is a database on which it is necessary only to formulate and execute the right SQL queries in order to extract all sorts of interesting information!
‘Business Method’ and ‘Software’ DefinitionsFor the purposes of this analysis, I have identified the technology field of the invention claimed in each granted patent according to the USPTO Art Unit to which it was assigned for examination.
Specifically, I define a ‘business method’ (‘BM’) patent as one assigned to any of the following Art Units:
- 3621-3629: Electronic Commerce;
- 3680-3689: Business Methods; and
- 3690-3699: Business Methods – Finance.
- 2120-2129: Miscellaneous Computer Applications;
- 2140-2149 and 2170-2179: Graphical User Interface and Document Processing;
- 2150-2169: Data Bases & File Management; and
- 2190-2199: Interprocess Communication & Software Development.
Finally, data obtained for ‘all technologies’ uses the complete available dataset over the relevant period, covering grant of all patents by the USPTO, including BM and SW patents as defined above.
Patent Grants, 2007-2015The chart below shows the total number of patents issued by the USPTO each month between January 2007 and December 2015, relative to 2007 averages. For comparison of absolute numbers, in January 2007 there were 17852 patents granted in total, of which 733 were from SW Art Units, and 271 were from BM Art Units.
What is very clear from the chart is that, for the majority of the period shown, the rate at which SW and BM patents were issued closely tracked the total issue rate across all technologies. There are, however, two notable exceptions to this ‘rule’.
First, having hit a peak in the early part of 2008, the grant of BM patents entered a decline in the latter part of the Dudas era, falling to just 50% of the average 2007 rate.
Second, the Supreme Court’s Alice decision in June 2014 killed about 75% of all BM patent applications stone dead. The remaining 25% of cases that do issue as patents now appear to be tracking the general rate across all technologies, presumably representing the proportion of applications with sufficient technical inventive merit to avoid the Alice exclusion test.
For those who suggest that ‘software patents’ in general are in trouble, this data clearly indicates otherwise. Across my selected SW Art Units, growth in patent grants has consistently and closely tracked that across all technologies over the past decade. Changes in USPTO Directorship, and decisions of appellate courts, have had absolutely no impact in this trend.
Independent Claim Length, 2007-2014Speaking very generally, the number of words in an independent claim may be regarded as a measure of the scope, or breadth, of a patent. More words usually means that there are more limitations set out in the claim, i.e. more features that are specifically defined as being essential to the invention, and which therefore must be present for infringement to occur. An applicant, typically through their attorney, will want to obtain the broadest – and usually, therefore, least verbose – claim possible subject to the legal requirements to define a new and nonobvious invention in sufficiently clear and concrete terms.
You would therefore expect claims to become narrower, and more thus more verbose, over time in any mature field of technology, because as the background of prior art grows larger, the need to define new inventions in more precise and distinct terms also increases.
The chart below shows the average length (in words) of all independent claims of patents issued by the USPTO each month between January 2007 and December 2014.
Across all technologies the trend is generally as expected – a slow but steady increase in average independent claim length over time.
Two things are notable about SW and BM patents, however. Firstly, independent claims in these technology areas are universally longer than average, with BM patents in particular having consistently the wordiest claims. Secondly, there is a distinct, and sustained, increase in average independent claim length for both SW and BM patents over the Dudas-Kappos transition in 2009. Interestingly, since that time claim lengths in both areas seem to have levelled out.
The most favourable hypothesis to explain the SW and BM claim length acceleration is improved examination quality. If examiners began to find more pertinent prior art, and issue better-quality rejections, applicants could be expected to respond by filing amendments to more-clearly distinguish their claims from the prior art.
The subsequent levelling-out of claim lengths, however, suggests that some of the additional verbiage may have been ‘pre-emptive’. That is, applicants may have begun to file claims with more words in an effort to placate examiners, even if the additional language did not, in practice, have a significant impact on the scope of the patent. This would make the step-change in claim length somewhat artificial, allowing such pre-emptive verbosity to be substituted for meaningful distinguishing limitations as time goes by. My own recollection of my experience during this period is that there would certainly be an element of this at play. Indeed, there is some evidence in the above chart that, for SW patents at least, average claims lengths may be back on track to a more ‘normal’ increasing trend, as seen prior to 2009.
Notably, there is no evidence in this data that either Bilski or Alice had any significant impact on issued BM or SW claim length. I thought there may be something in this, which is why I ran the analysis in the first place.
Total Number of Claims, 2007-2014There are many reasons why applicants may file larger numbers of claims in a single application. Sometimes it is because there are many possible variations (‘embodiments’) that fall within the scope of the more general inventive concept. Other times it may be to provide a range of ‘fall-back’ positions – e.g. narrower, more specific, dependent claims – as a defence against rejection (or post-grant invalidation) of broader claims. In still other cases it may be to target specific known or anticipated infringing activities. And sometimes it is simply because the applicant wants to see every single feature of their precious invention reflected in at least one claim, even though many of those features may not be sufficiently innovative to add anything of value to the patent.
Given this wide range of reasons for filing more claims, it is difficult to draw confident conclusions from statistics on the number of claims in issued patents. Having said that, however, the results in the chart below – which shows the average number of claims in patents issued by the USPTO each month between January 2007 and December 2014 – are striking.
I expect that the general decline in the average number of claims across all technologies is largely cost-related. The USPTO charges additional fees for each independent claim in excess of three, and for each total claim in excess of 20. There is, therefore, a cost incentive for applicants to keep below these numbers. The average is pulled up, however, by a relatively small number of patents with very high numbers of claims (e.g. in excess of 100). Quite possibly, the applicants in such cases have become more cost-conscious over time, particularly as fees have increased.
It does not surprise me that SW patents typically have more claims than the overall average. Computer-based inventions tend to have more possible embodiments and variations that inventions in fields such as, say, mechanics and chemistry. This results in larger numbers of dependent claims. Furthermore, it is common practice to claim computer-implemented inventions in multiple forms, e.g. as a method, a processor-based apparatus, and a computer-program product, which increases the claim count.
As for the striking jump in the number of claims in BM patents during the Dudas-Kappos transition, my hypothesis is that this marks a response to changes in USPTO examination practice following the CAFC decision in Bilski, which came down on 30 October 2008. The court emphasised its so-called ‘machine or transformation test’ (subsequently overruled as ‘too rigid’ by the Supreme Court) as a requirement for patent-eligibility. This would have caused BM patent applicants to start drafting (or amending) ‘pure’ business process claims as ‘computer-implemented business method’ claims, resulting in claiming strategies more similar to SW than other technologies.
ConclusionThe data presented above demonstrates that in established fields of software technology, covered by associated Art Units in USPTO Technology Center 2100, neither US court decisions nor changes in management have resulted in any identifiable deviation in US patent grant rates, despite the consistent gleeful claims of opponents that ‘software patents are dead’.
Software patents are not dead. They are here, they have been here for many years, and they are here to stay.
Computer-implemented business methods, including ecommerce and finance applications, on the other hand, are a different matter. The data clearly shows that the USPTO under Kappos was more friendly to this subject matter than under his predecessor, although the CAFC decision in Bilski appears to have forced applicants to claim machine-implementation more explicitly. However, the US Supreme Court decision in Alice looks to have eliminated about 75% of new business method patents. This implies that a similar proportion of such patents issued at least since the start of the Kappos era are invalid, which accords with the impact we have seen at the CAFC since Alice.
The Dudas era was widely regarded a dark time for patent applicants – remember, for example, the claims and continuation rules that were ultimately struck down as exceeding the USPTO’s authority? Ironically, however, with the benefit of hindsight, it appears that many fewer patentees would now be left holding expensive, but worthless, patents had the approach to business methods under Dudas been maintained.