Patent Portfolio Quality Metrics In-House Teams Actually Track

Filing count is not a quality signal

Most IP teams report up the org with one headline number: filings per year. That number is easy to track and easy to compare. It is also a poor quality signal, because most patents never get tested.

In Mark A. Lemley, Rational Ignorance at the Patent Office, 95 Nw. U. L. Rev. 1495 (2001), Lemley estimated that only a small percentage of issued U.S. patents are ever litigated and that fewer still are litigated to judgment. The patent portfolio is judged in shadow. That makes internal quality signals the only durable performance signal counsel actually have.

When patents do get tested, the litigation outcomes do not flatter portfolios optimized for volume. John R. Allison, Mark A. Lemley and David L. Schwartz, in Understanding the Realities of Modern Patent Litigation, 92 Tex. L. Rev. 1769 (2014), studied every patent case filed in 2008 and 2009 and found that patentees won decisively on the merits in only a minority of fully-adjudicated cases, with software and business-method patents performing worst.

The board-level question is not “how many patents did we file.” It is “what work do those patents actually do.” A portfolio-quality framework has to answer that question.

The system the metrics sit inside

The USPTO is operating at scale. The agency’s Patents Dashboard and recent Performance and Accountability Reports show the unexamined application backlog growing past 800,000 in FY 2024, among the largest in the agency’s history, against annual utility application filings in the high six figures, first-action pendency in the high teens to roughly 20 months, and total pendency including continuations and RCEs in the mid-twenties of months. Aggregate allowance rates published by USPTO and tracked by independent observers like Patently-O sit in the mid-70s to roughly 80% on a “first-pass” basis once continuations and RCEs are included, and substantially lower on a strict “applications-as-filed” basis.

Those system-wide numbers are the denominator. Portfolio-quality metrics measure where a specific portfolio sits inside that distribution.

Two structural facts make the measurement worth doing. The examiner is the binding constraint: Michael D. Frakes and Melissa F. Wasserman, in Irrational Ignorance at the Patent Office, 72 Vand. L. Rev. 975 (2019), estimated examiners spend on average roughly 18 hours per application across the entire prosecution. That is a hard ceiling on per-application attention. Counsel who deliver well-bounded, technically specific disclosures get disproportionately better outcomes from that fixed budget.

The second is variance. Empirical work, including Mark A. Lemley & Bhaven N. Sampat, Examiner Characteristics and Patent Office Outcomes, 94 Rev. Econ. & Stat. 817 (2012), and recurring Patently-O analyses of USPTO data, has consistently shown that examiner-level allowance rates inside a single art unit vary widely, with the most lenient examiners granting at multiples of the rate of the strictest. Lemley and Sampat further showed that examiner experience and seniority significantly affect grant rates. Examiner draw is a measurable variable, not noise.

A portfolio-quality framework has to let counsel see both: where the portfolio sits versus the system baseline, and how much of the variance is attributable to inputs the team can control.

Seven metrics that survive board scrutiny

The metrics below are not novel. They are the metrics that hold up when a CFO or an IP committee asks how the portfolio is doing.

• Allowance rate, sliced. Aggregate allowance rate is a baseline. The portfolio-quality slices are by outside firm, by art unit, by internal author or business unit, and by examiner. Outperforming the art-unit baseline by five points after normalizing for examiner is a real signal. The data lives in PAIR; modern docketing systems will surface most of it.
• Office actions to grant. A portfolio that grants in two actions or fewer is consistently better-bounded at intake. Three or more actions on average, particularly in stable software art units, is a drafting or disclosure problem upstream.
• Independent claim count. The Allison and Lemley empirical work suggests that more than five independents per case rarely improves litigation outcomes and reliably increases excess-claim fees. Two to four independents covers most defensible scope.
• Independent claim length. Alan C. Marco, Joshua D. Sarnoff and Charles A. deGrazia, in Patent Claims and Patent Scope, USPTO Office of the Chief Economist Working Paper 2016-04, documented that shorter independent claims correlate with broader scope. Independents over roughly 200 words are a yellow flag for narrow effective scope.
• Forward-citation impact. Bronwyn H. Hall, Adam B. Jaffe and Manuel Trajtenberg, in Market Value and Patent Citations, 36 RAND J. Econ. 16 (2005), established forward citations as a market-validated proxy for patent value. Citations are read against the tech-center average. Top-quartile citations relative to peers is the meaningful target.
• Examiner and art-unit variance. Track which examiners draw which families, what their published allowance rate is, and what their average actions-to-grant looks like. Strategy should adapt to the draw: different traverse and interview strategy for a 30%-allowance examiner versus a 90%-allowance examiner.
• Forced abandonment rate. Strategic abandonment is healthy portfolio pruning. Forced abandonment, the team gave up because they could not get acceptable claims allowed, should sit below ten percent of dispositions. Higher rates suggest a drafting or disclosure mismatch with the prior-art landscape.

A portfolio dashboard that reports those seven, sliced, will tell counsel and the board more in five minutes than a year of filing-count updates.

The metric to deprioritize

Raw filing count, and its near relative “patents per engineer,” should sit below the fold.

The case for deprioritizing them is empirical, not aesthetic. Lemley (2001) shows that the overwhelming majority of issued patents are never tested. Allison, Lemley and Schwartz (2014) show that fully-adjudicated patents lose decisively on the merits more often than they win. A portfolio measured purely on volume optimizes for examiner throughput, not for claims that survive litigation, defend product launches, or support licensing.

This is not an argument against filing. It is an argument against treating the count as the goal. The count is the byproduct of doing the job correctly upstream.

Reading the metrics together

Individual metrics are diagnostic, not dispositive. The pattern across metrics is what tells the story.

• Low allowance rate paired with high office-action cycles. The portfolio is consistently failing to anticipate the prior-art landscape. The intervention is upstream: better disclosure capture and stronger pre-filing search.
• High allowance rate paired with low forward-citation impact. Claims are being granted, but they are not being read by anyone, including competitors. Scope is too narrow. The intervention is broader independent claims and better scope decisions at drafting.
• High variance by examiner inside the same art unit. Strategy is not adapting to the draw. The intervention is examiner-level data plumbed into the prosecution playbook.
• Forced abandonment concentrated in one tech area. Disclosures in that area are arriving too thin, or the team is filing in a tech-center crowd it is not equipped to clear. The intervention is either upstream disclosure quality or strategic deprioritization.

The patterns are visible only when the metrics are collected together. They are invisible when filing count is the only number reported.

Where the data lives

Most of these numbers are in public USPTO sources. The USPTO Patents Dashboard publishes pendency and filings. The Office of the Chief Economist publishes the Patent Examination Research Dataset (PatEx) for application-level history. PatentsView publishes citation and assignee data. Modern docketing systems can roll the numbers up by family, firm, examiner, and business unit without a separate analytics platform.

The harder data is internal: disclosure-to-filing pendency, the rate of non-disclosed inventions caught only at retrospective review, the share of applications drafted from thin IDFs. These come from the intake pipeline, not the prosecution record. They are the numbers that explain why the prosecution metrics look the way they do.

Quality is downstream of intake

The seven metrics in the framework above are outputs. The inputs are upstream of prosecution.

Specifications drafted from thin disclosures produce more office actions, narrower independents, and higher forced-abandonment rates. Specifications drafted from rich, technically grounded disclosures produce the opposite. The leverage point for portfolio quality is the intake layer, not the drafting bench.

This framing is not new. The roughly 18-hour examiner budget Frakes and Wasserman documented sets a hard limit on what an examiner can do with a poorly bounded specification. The Berkheimer fact-finding posture, after Berkheimer v. HP Inc., 881 F.3d 1360 (Fed. Cir. 2018), increased the evidentiary load on the spec further. Both shifts reward upstream investment.

Counsel who measure portfolio quality eventually run out of ways to improve it without addressing the intake layer that feeds it.

What to read next

The companion to this metrics framework is the intake question it points back to. For the operational version of that question, see our note on attorney-led invention discovery, which describes the intake model that produces the disclosure inputs the seven metrics reward.

ALID surfaces invention candidates directly from engineering artifacts and packages each candidate with the technical specifics, alternatives, and source citations that produce defensible specifications and cleaner prosecution records. To see how the flow operates, read how ALID works, or request access to run a first discovery against your own engineering data.