As behavioral economics teaches, people operating under conditions of scarcity rarely make optimal decisions that lead to their greatest benefit. Instead, seemingly rational people can be clouded by conventional wisdom, emotions, and the influence of peers who are likewise subject to their own biases and imperfections. In the academic world, Richard Thaler established that people are predictably irrational. For the rest of us non-economists, Michael Lewis recounted that lesson in his book Moneyball: The Art of Winning an Unfair Game, which walks us through the foibles of professional baseball scouts before Billy Beane and the Oakland Athletics came along and changed everything.
It might be anecdotal, but adoption of data-driven analytics appears to happen quicker in environments of direct competition like athletics, while until recently the application of data-driven analytics by in-house IP departments seemed slow. Since the late 1990s, the conventional wisdom for many corporate patent teams has been to grow their patent portfolios, perhaps even measuring their success against growth of yearly filing rates or upward movement in overall portfolio size rankings. Patent benchmarking across industries often relied on portfolio size as an indicator of relative quality - falling into the classic behavioral economics trap of substituting an easy question - like, how big is your portfolio? - for a hard one, like: how effective is your portfolio for achieving desired outcomes?
More recently, however, parts of the IT sector appear to be questioning conventional wisdom. Some very large, highly-innovative and well-resourced companies are either growing their portfolios at slower rates or actively working to reduce their overall portfolio size. Perhaps this change can be attributed to changes in the law or financial hardship - external forces at play to work against conventional wisdom without being a direct assault on the premise that portfolio growth is good. At Uber, we believe this trend is not due to external forces, but instead marks an early-stage rejection of oversized portfolios and the beginning of an emphasis on portfolio quality, knowing the purpose of each asset and striving for high portfolio efficiency.
This article gives a glimpse into a portion of our thinking that helps us: (i) avoid the bloat that comes from the endowment effect (an invention disclosure gaining heightened importance merely upon receipt of the disclosure); and (ii) overvaluing one asset without taking account of the many factors contributing to its potential value (using slower analytical thinking instead of faster intuition based thinking that often proves inaccurate for complex multivariable considerations like building a patent portfolio).
Factors for right sizing a patent portfolio
Uber employs some of the brightest and most innovative minds in the tech world. As a result, the number of ideas that flow from this talent pool is sizeable. How does Uber decide what makes the cut? The focus is on building a high-quality portfolio that is effective and efficient in achieving Uber’s business goals and protecting our ecosystem. There are several factors that feed into a high-quality portfolio: relevance of the assets to Uber’s business, the assertability of the assets, and the efficiency of the portfolio.
Our team prioritises filings on products and features that are of strategic importance to Uber. Our goal is to build robust portfolios for business areas that are high-value propositions for the company. As a platform service, we want to not only file on concepts covering our core business, but also strategically file on forward-thinking inventions that will clear the patent-path for Uber and our industry partners. Having a strong portfolio is a strong value proposition we can deliver to our partners. From a defensive standpoint, we often file on consumer-facing concepts that are product differentiators which are susceptible to copying.
Our team also places an emphasis on the level of difficulty to assert any resulting patent. The invention needs to be detectable in both our own as well as others’ products. If a feature is undetectable and there are no plans from the engineers to publish details about the feature, our team leans toward trade secret protection. We also work closely with our engineering teams to determine whether the feature is integral to Uber’s products or the competitive landscape. An additional consideration is whether we believe the invention could be used across different technology areas and/or by companies outside of our technology area. Assets that fall into this category could prove to be useful in future cross-licensing negotiations against companies with broad technology footprints.
Without proper grooming, the continual growth of a portfolio becomes unwieldy from increasing carrying costs. Our team actively avoids growing our portfolio to a size that would become burdensome to our budget and affect our ability to freely file on inventions with a higher likelihood of utilisation and relevance to Uber. As part of our mission to maintain a high-quality portfolio, we aim to identify strategic purposes for assets in our portfolio. We have been working on building out a system to track the utilisation of our assets from the invention disclosure stage to expiration of the asset, so we can appropriately invest in and prune patent assets.
Using analytics to right-size patent portfolio development
We believe that analytics are a powerful tool to help develop and maintain a right-sized patent portfolio. Analytics can provide real-time projected measurements of the performance of the portfolio, guide investments to the most impactful parts of the portfolio, and help prevent wasted costs associated with patents having costs that exceed their potential value to Uber.
Although we look at each patent on a case by case ad hoc basis, it’s helpful to think about the impact of the portfolio at an abstract level. The impact of the portfolio is the sum of the economic coverage of its patents relative to each business-relevant product. The economic coverage is measured by the probability-weighted royalty. Mathematically, impact can be defined as follows:
where denotes the number of features of product; denotes the portfolio’s economic coverage on the feature of the product/revenue stream; denotes the probabilistic coverage of the portfolio against the feature; denotes a royalty rate of the feature; denotes the magnitude of the product/revenue stream; denotes a base royalty rate; and denotes an average feature significance, which serves as an appropriation scaling factor of the base royalty rate to determine the effective royalty rate of the feature.
The number of features and the feature significance can be selected based on knowledge of the product. Insignificant features from a patent enforcement perspective should be disregarded from the outset as the corresponding feature significance will be small and thus any patents covering such features will negligibly contribute to .
The probabilistic coverage can be defined by the likelihood that at least one patent of the portfolio reads onto the feature and is valid. This approach helps eliminate royalty stacking of similar patents. Formulaically, this measure can be expressed in terms of relevant patents as
where random variable denotes the number of patents of the portfolio that cover the feature; and denotes the likelihood that the relevant patent would fail in enforcement.
Analytics can be used to determine the likelihoods . Below are a number of illustrative analytical metrics that can contribute to quantifying the likelihoods without having the benefit of full blown litigation and discovery to obtain case specific facts:
Breadth – This measures the breadth of claims of the relevant patent. As an example, historical litigation outcomes can contribute to correlating the measured claim breadth with likelihood of litigation success.
Examination rigour – This suggests how the prosecution history of the relevant patent may affect likelihood of litigation or inter partes review success. One factor is the actual examiner’s allowance rate compared to the average for the art unit. Time in prosecution (eg, number of office actions) and the number of cited or relied-upon references can be used as other factors.
Innovation timeline – This considers how the priority date compares with the dates of similar patents or literature.
Relevance – This indicates how well the subject matter covered by the patent substantively matches the accused product. For a particular technology (such as one of the N features), machine-learned models can be developed to measure a patent’s relevance to the accused technology.
We’ve worked with David Andrews of Legal Analytics to develop analytic indicators along the lines described above.
Impact modelling can be used to inform portfolio management decisions. For instance, Equation 1 can be used to determine whether a portfolio has sufficient coverage to protect the company’s business or deter threats. The portfolio sufficiently covers a product when the corresponding score meets a threshold condition. When a product is covered, additional patents in that area can be abandoned, divested or ignored because owning such assets is not financially justified. Instead, resources should be focused on obtaining patents that improve scores for technology areas that fail the threshold condition.
For products lacking a sufficient level of coverage, Equations 1 and 2 inform which features of the product to focus on. As previously stated, patents that only map to features with more speculative significance may be deprioritised to invest in areas with higher marginal value relevant to higher significance features not adequately covered by the portfolio.
In fact, Equations 1 and 2 help achieve resource optimisation. The optimal choice is dynamic as the portfolio changes. When the number of relevant patents are about equal among the N features, adding a new patent relevant to the highest feature significance value will have the largest impact. However, as shown in Equation 2, adding a new patent to a feature group will have decreasing gains for each subsequent addition. At a certain point, adding a patent to the next highest feature significance will result in the biggest change in .
These analytics can be calculated continuously to monitor the state of the portfolio. Continuous monitoring can be used to make real-time prosecution decisions. For instance, examination statistics can be used to estimate carrying costs to obtain a particular patent. If the carrying cost is greater than the added impact of the patent, then the patent should be abandoned. These analytics empower the portfolio manager to make rationally-justified, disciplined decisions.
We’re excited about the potential that data analytics holds for building an impactful portfolio in this new era of data-driven decision making without breaking the bank on a voluminous patent portfolio. Although we’ve made progress, there’s much more work ahead.