Why patent landscaping is hard for R&D teams, and how to fix each failure point

Patent landscapes break at six points: defining the patent set, counting publications instead of inventions, reading filing volume as momentum, freezing a moving field in a one-off snapshot, missing non-English filings, and stopping at a map with no verdict. GetFocus addresses these by retrieving patent families by meaning, filtering them with a language model, and measuring each technology's improvement rate from the citation network.

A patent landscape should help an R&D leader decide which technical route to fund. In practice it often arrives as a long deck of filing charts that answers who holds what, and leaves the funding question open. This guide walks through where the work goes wrong, with sources for each problem, and what a better method looks like.

What is a patent landscape supposed to tell an R&D team?

WIPO describes patent landscape reports as an overview of patenting activity in a technology field, for a country, a region or the world (WIPO Patent Analytics). The vendor guides follow the same arc. Patsnap's 2025 guide sets five steps: define scope, build the search strategy, collect and normalise the data, analyse patterns, and generate recommendations (Patsnap, updated 24 Nov 2025).

For an R&D team, that last step is the one that matters. The question on the table is usually a choice between competing routes: sodium-ion or LFP, electrolysis or electrified cracking. Each of the six problems below weakens the answer to that choice.

Why is defining the patent set so hard?

Each chart in a landscape inherits the quality of the patent set underneath it. Practitioners measure that set with two numbers. Recall asks how much of the relevant material the search found. Precision asks how much of what it found is relevant. WIPO's Guidelines for Preparing Patent Landscape Reports give recall and precision their own section (WIPO, 2015).

Keyword and Boolean searches struggle on both counts:

• Recall drops when inventors describe the same idea in different words. In a GreyB specialty-chemicals project, a literal chemical-name search "returned a thin result set that clearly understated activity in the space" (GreyB case study).
• Precision drops when the query is widened to recover recall. In the same project, descriptor-based searches "produced too many irrelevant results" (same source).
• Iteration never ends cleanly. Patsnap notes that "initial searches rarely capture perfect scope" and that human expertise stays essential for edge cases (Patsnap guide).

A better method separates the two jobs. GetFocus runs recall and precision as two stages (GetFocus method):

1. Recall by meaning. The technology is written as a concept. Vector search pulls every patent family close to that concept in meaning and casts a deliberately wide net. As the method page puts it, "A Boolean string can only match the words an inventor happened to choose."
2. Precision by reading. A language model reads each retrieved family (title, abstract, claims and description) against one written instruction and returns INCLUDE or EXCLUDE.

The instruction is plain language a domain expert can review and sign off, which makes the set definition auditable. The help center documents how these LLM-Filters work and how long they take (LLM-Filters; What is vector search?).

Should you count patent families or patent publications?

One invention can produce many documents: an application and a grant in several offices, each with its own number. The WIPO Patent Analytics Handbook warns that patent data "involves multiplier effects that often leads to the duplication or republication of the same document" and describes a family as "a stack of documents published in any language anywhere in the world" (WIPO handbook, Counting Patent Data).

Count publications and a company that files in fifteen countries looks fifteen times as inventive as one that files at home. Patsnap's evaluation guide makes the same point: "counting publications instead of invention families can overstate activity" (Patsnap, 22 Jul 2026).

The fix: use the patent family as the unit of analysis. GetFocus keeps one row per invention: "The same invention filed in fifteen countries is one patent family" (GetFocus method). Even then, the WIPO handbook notes that databases list priority numbers inconsistently, so check how your tool groups families before comparing counts across sources (WIPO handbook).

Why does counting patent filings mislead technology forecasts?

A typical landscape chart plots filings per year, and a rising curve reads as momentum. The record shows the opposite pattern often enough to matter.

In 2001, hard drives out-filed solid-state storage six to one: 3,316 patent families against 580. In the same year, solid-state storage was improving at 50% a year against 29% for hard drives (GetFocus backtest: HDD vs SSD). The filing count pointed at the incumbent. The improvement rate pointed at the route that became the standard.

That case is one of seven replayed on patents public at the time. In each, filing volume pointed at the loser while the improvement rate pointed at the winner (GetFocus proof). Other examples on the same page:

• In 1990, LED lighting measured 32.5% a year against incandescent lighting's 8.6% (lighting backtest).
• In 2001, NMC lithium batteries measured 54% a year against hydrogen fuel cells' 21% and the combustion engine's 11% (mobility backtest).

The academic work behind this is published. Benson and Magee found that the number of patents in a domain was not a significant predictor of its improvement rate (r = 0.33, p = 0.085), while citations within three years of publication tracked improvement closely (r = 0.76) (GetFocus summary with full references).

The fix: read the citation network. GetFocus computes two signals per patent family. Cycle time is the gap in years between an invention and the earlier ones it builds on. Knowledge flow is how much later inventions cite it. Together they yield the Technology Improvement Rate (TIR), validated on 150+ domains at R² ≈ 0.7 against measured rates (GetFocus method; What is TIR?). Filing counts still have a use: they show who is active and where. They answer a different question from "which route is getting better fastest". For more on reading patent signals, see How R&D teams can use patent trends to forecast emerging technologies.

Why does a patent landscape go out of date so fast?

Two clocks work against a one-off landscape.

The first is publication lag. "The gap between the filing of an application and its publication is generally 18 months," says the WIPO handbook, which is why recent years in a filing chart appear to fall away. The handbook suggests pulling the date range back two to three years and documenting the choice (WIPO handbook).

The second is the field itself. New entrants publish, a route stalls, a competitor files in an adjacent class. Patsnap describes landscape analysis as iterative (Patsnap guide), and its evaluation framework scores tools on "the effort needed to rerun the landscape" (Patsnap evaluation guide). A snapshot commissioned for a stage-gate review is often stale by the next one.

The fix: keep the dataset live. In GetFocus, Monitoring refreshes a project on a weekly, monthly or quarterly cadence, checks for new patent families, applies the saved filters and LLM-Filters to the new material, updates the TRL × TIR dashboard and reruns scheduled Deep Dive questions (GetFocus Monitoring). See also Automatic technology monitoring.

How much does language coverage matter in a patent landscape?

A great deal, because a large share of new filings is not in English. WIPO counted a record 3.7 million patent applications worldwide in 2024, and China's office accounted for 49.1% of the world total, up from 34.6% in 2014 (WIPO World IP Indicators 2025). A search run only on English full text sees a thin slice of that activity, and an English keyword cannot match a Chinese claim.

Coverage varies by office and by field: bibliographic data, full text, claims, descriptions and English translations each start at different years. Ask any vendor for its per-office table. GetFocus publishes one: bibliographic data from 105+ countries and full text from 59, sourced from IFI Claims (GetFocus data coverage). Retrieval by meaning across the corpus means a family is found by what it describes (GetFocus method).

How do you turn a patent map into an R&D decision?

This is where many landscapes stop. A cluster map shows where activity is dense. Patsnap cautions that a sparse area may reflect "the search strategy, publication delay, terminology" or classification choices, so it should not be read as open territory (Patsnap evaluation guide). Neither density nor white space tells you which route to fund.

A decision needs two readings per route:

• Maturity (TRL): how ready the route is today.
• Improvement rate (TIR): how fast it is getting better.

Plot both and every route lands in one of four quadrants: Fund (high TRL, high TIR), Rising star or monitor closely (low TRL, high TIR), Phase out (high TRL, low TIR) and Ignore (low TRL, low TIR) (GetFocus Monitoring docs).

A live example: the GetFocus radar on low-carbon ethylene scores 14 routes. Seven clear the set's median improvement rate of 20% a year. Electrochemical CO₂/CO measures about 44% a year at TRL 6 (Fund); the steam-cracking benchmark measures 13% a year at TRL 9 (Phase out) (ethylene radar, updated September 2026). For why maturity alone misleads, see our guide to technology readiness levels.

Where patent landscaping breaks, and what fixes it

What are the limits of patent-based forecasting?

A good landscape states the limits of its method. These are the ones GetFocus publishes.

• TIR is a directional signal and carries no guarantee of commercial success (GetFocus Monitoring).
• A reliable TIR needs a dataset of at least 50 patent families (What is TIR?).
• The method is less reliable in fields driven by secrecy, tacit knowledge or weak patenting. GetFocus flags thin coverage during scoping (method FAQ).
• Timing is directional. Exact crossover dates are uncertain (What is TIR?).
• GetFocus is built for choosing which technology to fund. It is not a prior-art, freedom-to-operate or portfolio-valuation tool (GetFocus tools comparison).

To test any method on your own industry before trusting it, replay it on a past technology shift: How to backtest a technology forecasting method.

FAQ

Why is patent landscaping difficult?

Because each step can distort the result: keyword searches miss relevant families and include irrelevant ones, publications inflate counts, recent years are hidden by an 18-month publication lag, and a large share of filings is in Chinese. Even a clean map shows activity, and an R&D decision needs a ranking of routes.

What is the difference between recall and precision in a patent search?

Recall is the share of all relevant patents your search found. Precision is the share of retrieved patents that are relevant. Raising one with keywords usually lowers the other. Searching by meaning for recall and filtering with a language model for precision lets you optimise each separately.

Should a patent landscape count families or publications?

Families. One invention filed in many offices produces many publications, so publication counts overstate activity for applicants who file widely (WIPO handbook).

Can patent filing counts predict which technology will win?

Filing counts are a weak predictor. In 2001 hard drives had 3,316 patent families against 580 for solid-state storage, while solid-state was improving at 50% a year against 29% (GetFocus HDD vs SSD backtest). Improvement rates measured from the citation network called the outcome.

How often should a patent landscape be updated?

Match the cadence to the field. GetFocus Monitoring offers weekly runs for fast-moving areas, monthly for most domains and quarterly for slower ones (Monitoring).

What does GetFocus add to a patent landscape?

A forecast. GetFocus builds the patent set by meaning, filters it with a language model, measures each route's Technology Improvement Rate from the citation network, and places every route on a TRL × TIR radar with a verdict: fund, monitor, phase out or ignore.