Software Patent Allowance Rates: What Examiner Records Show
Imagine two applications for similar machine-learning inventions. Both are routed to the same art unit at the USPTO. One is assigned to an examiner whose public record shows an 88% allowance rate. The other goes to an examiner whose record shows 20%. Neither applicant is told this. Most never find out.
We compiled the public prosecution records of 5,651 examiners across the three USPTO technology centers that examine software, artificial intelligence, and networking inventions — about 2.4 million decided applications in all. This article covers what that record shows, and what it does and doesn't mean for a company considering a filing.
How applications get assigned
When a patent application is filed, the USPTO classifies it by subject matter and routes it to an art unit — a group of examiners who handle a particular technical area. Within the art unit, the application is docketed to an individual examiner. The applicant doesn't choose the art unit or the examiner.
Outcomes vary considerably depending on where an application lands. Some of that variation reflects differences in the applications themselves — a unit that receives harder cases will show lower allowance rates. Some of it reflects differences in how individual examiners approach the same questions. The public record doesn't separate the two. What it does show is the size of the variation.
Across roughly 250 software art units, the average allowance rate — the share of decided applications that were allowed rather than abandoned — runs from 36% at the low end to 94% at the high end. Four out of five units land somewhere between 55% and 82%. Forty-two units average 80% or better; twenty-two average below 55%.
Within a single art unit, the range between examiners is often wider still.
What the record shows for AI and machine learning
According to their published subject-matter descriptions, the USPTO concentrates neural-network and machine-learning applications in art units 2121 through 2129. Across those nine units, the median allowance rate is 62%, compared with 70% for software art units as a whole. Three of them — 2123, 2128, and 2122 — average 51%, 53%, and 55%, meaning roughly half of decided applications in those units were abandoned.
The art-unit average is only part of the picture.
In art unit 2122, the nineteen examiners with at least 100 decided applications in that unit show allowance rates from 20% to 88%. In art unit 2128, the range is 27% to 90%. In every one of the nine AI units, the gap between the highest and lowest examiner is at least 35 percentage points.
Here's the simplest way to see the difference. Counting each examiner once, in the art unit where they've decided the most applications, and limiting to examiners with at least 100 decided applications:
The typical software examiner allows about 4 in 5 applications. The typical AI examiner allows about 7 in 10.
| Allowance rate | Software examiners (3,452) | AI-unit examiners (124) |
|---|---|---|
| 90% or higher | 17% | 6% |
| 80–89% | 35% | 23% |
| 70–79% | 23% | 21% |
| 60–69% | 13% | 26% |
| 50–59% | 8% | 14% |
| Below 50% | 5% | 10% |
Or, out of every ten examiners: in software, about five allow 80% or more of what they decide and fewer than one allows less than half. In the AI units, about three allow 80% or more, and one in ten allows less than half.
Two other areas stand out. Information retrieval and database structures (art unit 2169) averages 47%. The AI and neural-network units in the 2140s (2142 and 2147) average 44–45%, lower than the 2120s. Classification, which is based on how the invention is described in the application, determines which of these units receives it.
How long it has taken
Across software art units in our data, the typical unit averages about 23 months — just under two years — from filing to a first office action, the examiner's first substantive response. Final disposition takes about 39 months, a little over three years. The slowest units average more than four years.
The AI units are slower on both counts: roughly two and a half years to first action and three and a half years to disposition.
| First office action | Final disposition | |
|---|---|---|
| Communications (TC 2600) | 22 months | 34 months |
| Networking & security (TC 2400) | 23 months | 39 months |
| Software & architecture (TC 2100) | 26 months | 42 months |
| AI / machine learning units | 28–29 months | 42 months |
These are historical averages over decided applications. The USPTO publishes its own current pendency statistics, which can differ from the figures here for a given period. What the historical record consistently shows is that software applications have taken years, not months, to resolve — and that the first office action is commonly a rejection rather than a final outcome. What happens after that first action accounts for much of the difference in final results.
How practitioners use this information
Applicants can't choose their examiner. Here is what the process does allow, and how experienced practitioners tend to use it.
Classification is set at drafting. Because the USPTO classifies an application based on how the invention is described in the claims and specification, the drafting stage is where the art unit is effectively determined. Practitioners familiar with the software art units generally consider classification as part of preparing the application.
A first rejection is usually the beginning of prosecution, not the end. The allowance rates above reflect final outcomes — after responses, amendments, and interviews. The same first rejection from an examiner with a 75% allowance rate and from one with a 30% rate can call for different approaches, which is one reason practitioners often look at the examiner's record before deciding how to respond.
Examiner interviews are available. The USPTO permits applicants to request an interview with the examiner, and practitioners commonly use them to understand the examiner's position before responding in writing.
The examiner's record is public. Whether an examiner tends to allow after a written response, tends to require an interview, or has a particularly low or high allowance rate is visible in the public record. Our examiner directory collects that record in one place.
Continuation practice is common. With multi-year pendency, a single application rarely covers everything a growing company develops. Filing continuations — keeping a family member pending so later developments can be claimed — is widely used by software companies to keep options open.
Prioritized examination exists. The USPTO's Track One program advances an application for an additional fee, with a stated goal of final disposition within twelve months. Some applicants use it when timing matters to a financing, acquisition, or licensing discussion.
What these numbers are, and aren't
Every figure here is pooled from the public USPTO record. Allowance rate is allowed applications divided by decided applications (allowed plus abandoned), with pending applications excluded. The rates are not adjusted for differences in the applications each examiner received, and they pool multiple years of activity. They describe what has happened across many past cases. They do not predict the outcome of any particular application.
What the record does is make visible a variable most applicants never see. The USPTO doesn't tell an applicant their examiner's allowance rate, or that allowance rates vary this widely between art units. That information has been public for years, spread across millions of individual records. We put it in one place because the people making filing decisions should be able to see it.
Have a pending application? Look up your examiner to see their public record. Considering a filing? Contact us to discuss your application.
Sean Lynch is a registered patent attorney and partner at Lynch LLP in San Diego. The firm's examiner directory includes more than 13,000 examiner profiles across all eight USPTO technology centers, built from the public record.
This article is general information about publicly available USPTO data and the patent examination process. It is not legal advice, does not address any particular application or situation, and does not create an attorney-client relationship. Historical allowance rates and timing figures are not predictions of any outcome. Anyone considering a patent filing should consult a registered patent attorney about their specific circumstances.