Post by Ted Sichelman
Updated July 3, 2015
In a recent post on this blog, Janet Freilich asserts that “it is exceedingly difficult (if not impossible) to know if one is infringing a patent.” Freilich further contends that “larger numbers of patents exacerbate this problem,” and approvingly quotes Christina Mulligan and Timothy Lee’s claim that “In software, for example, patent clearance by all firms would require many times more hours of legal research than all patent lawyers in the United States can bill in a year.”
These views generally reflect the academic zeitgeist regarding “search costs” in patent law. However, they do not accord with my own experience. In this regard, I have conducted many hours of patent searches—first, when I founded and ran a venture-backed software company for five years, for which I regularly reviewed patents of competitors; and second, as a practicing attorney, which included reading thousands of patents in many different fields (from software to electronics to medical devices) for “freedom to operate” and prior art purposes.
In my view, Mulligan and Lee arrive at their conclusion through faulty assumptions about how patent searching and review works. First, they wrongly assume that “it takes a patent lawyer one hour to determine whether a given widget-related product infringes a given widget patent.” In most freedom-to-operate searches, it usually takes five minutes or less—often a minute or less—to determine with reasonable certainty if a given patent is relevant to a company’s products. Indeed, at a recent conference at the University of San Diego on artificial intelligence & law (a topic I return to below), Google employees estimated that its in-house attorneys spend 5-7 minutes determining whether a patent potentially covers any of the numerous products it offers. Typically—even for software—a very small percentage of patents that turn up in a search are highly relevant. Though these patents may take an hour of attorney’s time, in my experience they never make up more than 10% of all reviewed patents. (Of course, a small number of patents may take more time because of long disclosures or complex claims, but—contrary to popular accounts—these patents have not in my experience made up any more than 1% of all patents in a typical clearance search.) So, at worst, the average amount of time needed to review a patent in a clearance search for a single product is about 12 minutes (6 minutes x 90% + 60 minutes x 9% + 120 minutes x 1%), not an hour.
Second, Mulligan and Lee wrongly assume that a given company must review every patent held by every company in its industry. Intelligent patent searches—even in software—usually reduce the set of relevant patents by a huge percentage, typically by 90-95% or more. Additionally, many companies engage in widespread licensing deals—particularly cross-licensing—which substantially reduces the number of patents that need to be reviewed. Unfortunately, there is no hard data on the prevalence of licensing and cross-licensing, but my best guess is that they reduce the search set by another 25-35%, and even higher for regular patent action defendants (upwards of 50%). Furthermore, many patents are not renewed and thus effectively have expired, and do not need to be reviewed. (Although the face of the patent does not indicate whether a patent has been maintained, this information is now easily obtainable—for example, through Google Patents.)
Third, each company holds many patents. Thus, once it’s determined that one of a given company’s patents need to be licensed, it’s not as important to carefully review all of the other company’s patents, because patents are typically licensed as portfolios (even in settlement of litigation). The patentholder will usually do the hard work of determining which of its other patents may need to be licensed.
Fourth, many patentholders rarely assert their patents. Patent litigation is costly and uncertain, and typically only cash-rich plaintiffs can sustain litigation for a lengthy period of time. As a result, many patents go unenforced. Relatedly, Mulligan and Lee estimate that there are 600,000 companies that produce software. Even if this is correct—which I sincerely doubt—a company that simply writes some code now and then that is used internally or in some minor fashion is extremely unlikely to be sued or asked for a license. The majority of software companies are small consulting shops that create software “on spec.” These consulting shops are almost never sued and will generally have indemnification agreements with the company for which they write software. At most, 10% of these companies (60,000) face any non-trivial exposure to costly patent assertions. Indeed, there are only about 1,000 total defendants sued for infringement of software patents per year. It’s highly unlikely that 60 times that number receive demand letters and actually pay up.
Mulligan and Lee estimate that comprehensive patent search in the software industry would cost $400 billion per year (even assuming 10 minutes per patent for review). That’s simply ridiculous. A better (and still conservative) estimate would be the following: (1) 60,000 companies that on average release a product per year face some potential exposure to software patent threats, (2) 50,000 software patents issue per year (I assume that product releases for most software companies are similar enough to previous ones that companies do not need to extensively search or review prior years’ patents); (3) 10% of software patents that are potentially relevant to each company on average (so 5,000 patents per year)—remember, most of these companies are small and have few products; (4) 12 minutes on average per relevant patent; and (5) 30% of all those patents are already licensed or could easily be licensed (that is, for a reasonable fee and outside of litigation), are part of a portfolio, or are not in-force, essentially obviating the need for review. This results in an estimate of 700 hours per company. At $200 per hour, that’s about $140,000 per year per company, or $8.4 billion for the industry as a whole.
Many companies would still consider $140,000 per year high—and, indeed, given the low odds of being sued or even threatened, it’s still not worth the time for many companies to perform broad-based “freedom to operate” searches on all of their product lines. (However, I will note that the 2008 Berkeley Patent Survey, which I conducted along with Stuart Graham, Robert Merges, and Pamela Samuelson, found that even among software startups, about 25% regularly review the patent literature, especially as part clearance searches prior to product development and launch—further justification for a somewhat high, but generally reasonable, per company annual search cost in the software industry. Outside of software, the percentages jump to 65-75%.)
Finally, I return to the above-mentioned artificial intelligence & law conference, where Google presented on numerous automated techniques it is developing to review patents. Not only were these techniques faster in a head-to-head test against Google’s own lawyers, but they were also more accurate and more comprehensive—four hours spent by the lawyers working individually, compared to minutes spent by the software. Over time, these technologies will spread and will vastly reduce the amount of attorney time required to review patents. My very speculative guess is that within a decade or two, this sort of technology will reduce the costs of clearance searches by 75% or more.
So are the roughly 2.5 to 3 million in-force U.S. patents too many patents to search? Perhaps somewhat at the moment, but not incredibly so, and almost certainly not in the near future.
Categories: