Announcement: Introducing iLumOS by Lumenci: Expert-Powered AI Platform for Patent Intelligence

AI Patent Portfolio Review: Why Claim Construction Still Needs Experts 

Take a portfolio of 2,000 patents. A standard review ranks them on citations, family size and claim counts, sends the top 100 to analysts, and ends with a handful of patents charted against products. A patent with few citations that reads directly on a best-selling product can fall out at the ranking step, before anyone has compared its claims with a product. 

AI changes that step. A model can read every claim in the portfolio against product documentation and flag the patents with a likely product match, so experts start from mappings instead of a ranked list. It also means patents rejected in earlier reviews can be re-screened when new products ship. 

The new weak point is the mapping itself. Each flagged match depends on what the claim terms mean, and in a patent case that meaning is fixed through claim construction, which a court decides. A mapping built on the wrong reading of one term is a false lead, however confident the output looks. This article covers where proxy-based ranking loses good patents, how an AI-first review changes where experts start, and what they check in each mapping before a patent moves to claim charts. 

Key Takeaways 

  • Citation counts, family size and claim counts describe the patent document. They do not show whether a product practices the claims, so ranking on them can drop strong patents before anyone compares them with a product. 
  • AI can compare every claim in a portfolio with product documentation, which moves the analyst’s starting point from a ranked list to a set of flagged product mappings. 
  • Every flagged mapping rests on a reading of the claim terms. In litigation, the court decides claim construction, reading each term as a skilled person would in light of the specification and prosecution history. 
  • Before a patent goes to claim charts, experts test each strong mapping against the likely construction, the claim structure and the evidence available. 
  • Patents set aside in an earlier review can be re-screened against current products without re-running the whole review. 

How a Proxy-First Portfolio Review Works

Most portfolio reviews narrow the field in four steps. The team filters out expired, encumbered or out-of-scope patents; ranks the rest on objective indicators such as forward and backward citations, claim count, family size and litigation history; runs a high-medium-low (HML) review on the top 50 to 100; and takes the 5 to 10 strongest into detailed infringement analysis and claim charting. Lumenci’s patent portfolio triage guide covers the scoring criteria and tier matrix for that process.

Is Your Licensing Programme
Growing Fast Enough?

Lumenci’s claim charting and campaign support helped
triple the client’s licensing growth.

Where Proxy Metrics Drop Strong Patents

The ranking step is where a proxy-first review loses patents. Citations, claim length, family size and prosecution history are signals about the patent document. They can show that later applicants built on a patent, or that its owner paid to file it in several countries. None of them shows whether a commercial product practices the claims. 

So a patent can rank low and still read cleanly on a high-volume product. In a proxy-first review nobody finds that out, because claims are only compared with products after the ranking step. 

The same applies to patents set aside in earlier reviews. A patent that mapped onto nothing three years ago may map onto a product launched since. With AI screening, it can be checked against current products without re-opening the whole review. 

What Changes When AI Reads Every Claim Against Products

AI first entered portfolio work as an analyst productivity tool: summarizing patents, pulling out technical concepts, searching documents. The larger change comes from moving it to the front of the process. A model can compare every claim in the portfolio with product documentation and surface the patents with the strongest combination of likely infringement and detectability, meaning the claimed feature can be observed or tested in the product. 

The first question for the analyst changes from “which patents should I read?” to “which of these flagged mappings hold up?” 

blog inline image ai patent portfolio

The main risk moves with it. A proxy-first review risks missing good patents. An AI-first review risks trusting weak mappings, and the main check against that is claim construction. 

Why Claim Construction Still Needs Experts

Claim construction is the process of deciding what the words of a patent claim mean. In U.S. litigation it is the judge’s job. In Markman v. Westview Instruments (1996), the Supreme Court held that construing a patent, including terms of art in its claims, belongs exclusively to the court. Disputed terms are usually settled at a Markman hearing before trial. 

Still Spending Months
Reviewing Patent Portfolios?
Image link

The Markman case shows how much turns on one word. The patent claimed a dry-cleaning system that tracked “inventory.” A jury found infringement, but the trial court construed the claim to require a system that could track articles of clothing, and Westview’s system tracked invoices and cash totals. Under that construction there was no infringement, and the Supreme Court affirmed that the call was the court’s to make. A tool matching the everyday meaning of “inventory” could easily have flagged Westview’s product as a hit. 

The method courts follow comes from Phillips v. AWH Corp., a 2005 en banc Federal Circuit decision. Claim terms take the ordinary and customary meaning a person of ordinary skill in the art would give them at the time of the invention, read in the context of the specification and the prosecution history. That is where AI mappings are most likely to go wrong. A model comparing claim text with product text can match the words while missing a definition in the specification or a narrowing statement the applicant made to get the patent granted. 

What Experts Check in an AI-Flagged Mapping

  • Claim structure. Every claim has a preamble, a transitional phrase and a body of elements. The transition sets scope: under MPEP 2111.03, “comprising” is open-ended and does not exclude unrecited elements, while “consisting of” excludes any element not specified. A product with an extra component can fall outside a “consisting of” claim that a word match would flag. 
  • Term meaning. Does the specification define a key term, or use it more narrowly than its everyday sense? 
  • Prosecution history. Did the applicant narrow a term or give up scope to get past prior art? Those statements can limit the claim later. 
  • Every element. Infringement requires every claim element to be present in the product, literally or by an equivalent. A mapping that matches four of five elements still needs the fifth. 
  • Evidence. Can the matched feature be shown from public documents, testing or reverse engineering? Lumenci’s guide to patent infringement analysis methodology covers when public evidence is enough. 

Patent claim construction at this stage is a forecast. Nobody knows how a court will construe a term until it rules. What an expert can do is identify the terms likely to be disputed and test whether the mapping survives each reasonable reading. A mapping that only works under the broadest reading is a weaker candidate, however strong the product match looks. It is worth settling this before charting, because the positions taken in infringement contentions later shape the construction arguments available at a Markman hearing. 

What Stays With the Patent Analyst

AI moves the patent analyst downstream: less time finding and screening, more time validating, correcting and building evidence for the strongest mappings. Human judgment still decides whether a mapping is technically sound, whether the evidence is enough, whether the likely construction supports it, and whether the patent is worth pursuing commercially. Lumenci’s article on AI in IP and the patent analyst role covers how that role is changing, including hallucination and automation-bias risks. 

Check out our Claim Charts Services
claim chart report

How Lumenci Takes Patents From Flagged Mapping to Construction-Tested Claim Chart

Lumenci works with patent owners, law firms and litigation funders building licensing and litigation programs, and applies the same technical rigor when supporting defendants. 

  • Construction-aware mapping. Lumenci engineers read each key claim term against the specification and prosecution history before a patent is charted, and Lumenci’s technical teams support claim interpretation for Markman hearings through domain expertise, standards analysis and reverse engineering. 
  • Experts through trial. Lumenci’s technical experts have testified through claim construction, expert reports and trial. 
  • Track record. Lumenci has served 200+ clients, analysed 100,000+ patents and supported $3.5B+ in verdicts, settlements and cost savings. 

Talk to Lumenci about taking your portfolio from screening to claim charts. 

Conclusion

AI-first screening closes the coverage gap in proxy-based portfolio review, because every patent gets compared with products. Claim construction is the check that keeps the resulting shortlist reliable: a mapping is only as strong as the reading of its claim terms that a court is likely to adopt. 

For portfolio-scale screening, iLumOS, Lumenci’s expert-powered AI platform, scores each patent on strength, quality, infringement signal and validity resilience, and Lumenci experts take the flagged patents through construction review and claim charts. 

Table of Contents
Let's Start a Conversation
Product Testing to Outcome
Still Spending Months
Reviewing Patent Portfolios?

Identify high-value patents, discover licensing opportunities, and prioritize portfolios with expert-powered AI tool built by Lumenci.

Image link
Image link

Frequently Asked Questions

What is claim construction?

Claim construction is the process of deciding what the words in a patent claim mean. The meaning given to each term sets the claim’s scope and often decides whether a product infringes. 

The judge. Markman v. Westview Instruments (1996) held that construing patent claims belongs exclusively to the court. Disputed terms are usually resolved at a Markman hearing before trial. 

AI can flag terms likely to be disputed and match claim language to product documentation. Construction depends on the specification, the prosecution history and how a skilled person reads each term, and the final call belongs to a court, so experts forecast it and test mappings against it. 

A claim has a preamble, a transitional phrase and a body listing the elements. The transition sets scope: “comprising” is open-ended, while “consisting of” excludes any element not listed. 

The analyst reviews the mappings the AI flags, tests each against the likely claim construction and the available evidence, and builds claim charts for the ones that hold. 

Citations measure how often later patents refer to a patent. They say nothing about whether any product practices its claims, so a rarely cited patent can still read on a high-volume product. 

Claim construction decides what the claims mean. Infringement analysis compares the accused product with the construed claims, element by element. 

Related Posts