Skip to main content
Behind the dataData & Methodology

From public records to
partner intelligence.

FlowIP connects applicants and representatives across patent and trademark filings to reveal patterns of international activity. Here is how those patterns are built, and what they can tell you.

Follow the records
What linked records revealIllustrative example
KRRecord A

Korean filing

Applicant: Company A
Representative: Firm A

Priority · publication number · filing date

USRecord B

US filing

Applicant: Company A
Representative: Firm B

Priority · publication number · filing date

Linked by priority / family
KRFirm A
USFirm B
1 family

Both firms appear in the linked records.
This is evidence of participation, not proof of a referral or an instruction between them.

Network scope
  • KoreaKR
  • United StatesUS
  • EuropeEP / EM
  • JapanJP
  • ChinaCN

Patent and trademark records use their own linking and interpretation rules.

01 / The process

From records to metrics

Select a stage to explore

Start with published records and specialist data feeds

We collect patent and trademark bibliographic and representative records from specialist data suppliers and public patent-office sources. Fields and publication schedules vary by source; not every filing includes the same information.

Sources include Lighthouse IP and public patent-office data. Official registers and firm websites help establish identities and profile information.

  • Filing records
  • Bibliographic data
  • Representatives

Source coverage varies by field, office and period.

02 / Reading the data

Three layers of information

Observed records

What is in the records?

Filings, applicants, representatives and priority details supported by source records and their links.

Example: two representatives appearing in the same family

Calculated metrics

What patterns emerge?

Counts and distributions calculated from the linked records within a defined scope and period. Patent counts include estimated links.

Example: outbound and inbound applications for a corridor

Estimated signals

What do we infer from the records?

Links reasonably inferred from the records, and signals that prioritize candidates using observed activity and fit. A recommendation does not predict or guarantee a future referral.

Example: the Possible tier of a patent count, partner fit or estimated influence on counsel selection

Applications by default, families for older periods

Filings, families and partner connections count different things.

Patent counts on firm profiles and lists are foreign patent applications filed in the selected period. A firm counts as outbound when it handled the origin filing and inbound when it handled the destination filing. Public pages without full access, periods that reach back before the latest application data, and some screens not yet moved to applications use families instead. A patent family groups related filings for the same invention. Yearly family counts count each family once per firm, filing office and year, then add the selected years. Partner connections count each family separately for each partner firm and year. A family linked to two partner firms in the same year counts as two connections.

  • Korean application
  • US publication
  • US grant
1
family

Three documents do not mean three independent cases.

How should I read Outbound and inbound?

Outbound means the firm handled the origin filing of a linked application; inbound means it handled the destination filing. PCT filings are linked as well. Read direction relative to the selected firm, jurisdiction and period. Patent application counts use filing records; where a screen offers filing and applicant perspectives, compare like with like. These counts include estimated links, so they show likely participation, not proof of who instructed foreign counsel.

03 / Coverage & quality

Accuracy is more than a single percentage.

Source completeness, organization matching and relationship evidence answer different questions. Read the reporting period, coverage and evidence alongside each metric.

What to keep in mind
Source coverageAvailable records

Some documents have no representative field. For example, our commercial patent feed provides US attorney data on granted documents, while US pre-grant publications in that feed omit it. The feed alone can therefore understate US-side relationships. Japan-to-US inbound counts are the clearest example. Follow-on filings a firm sends abroad are not yet fully credited as outbound, so outbound counts for US firms can also read low.

Organization identityMatching evidence

Spelling variants, name changes and the distinction between individuals and firms can introduce matching errors. Registers, addresses and aliases help resolve them, but not every name can be identified conclusively.

Time & locationReference points

Publication and processing introduce delays. Map locations also vary in precision: a verified address, an approximate point and a regional location do not make the same claim. A marker is not always an exact office address.

Zero is not the same as unavailable.

An observed zero, an unresolved identity and a metric that has not been calculated are different states. A missing value does not establish a lack of activity.

Compare the same scope.

Use the same period, jurisdiction and counting unit. A decline in the most recent period may reflect publication delays rather than a change in activity.

Use patterns to build a shortlist.

Start with repeated participation and technology fit, then review individual records and current organization details before approaching a potential partner.

04 / Common questions

A few useful distinctions

Do patent counts include estimates?

Yes. Public records cannot confirm every relationship, so FlowIP also shows flows that can be reasonably estimated from priority, PCT and document-agent records. A patent count is the sum of its Confirmed, Likely and Possible tiers. These tiers grade the link evidence. Trademark evidence levels grade organization matching, which is a separate scale. The counts show likely participation, not a referral or an instruction.

How are inferred trademark activities reviewed?

Same-mark inference and common applicant / owner activity link filing records using applicant names, marks, classes and representatives. They do not prove referrals or instructions and do not change sent, received or family counts. An independent review of 11,668 sample items from the 2025 batch found source evidence in about 90% and clear conflicts in about 2–3%. For the earlier historical-link update, we reviewed 584 sample items. After adjusting for the sample composition, about 92% had source evidence and about 0.5% had clear conflicts. The independent review sample for the links added in this update contains 882 items. After adjusting for sample composition, about 94% had source evidence and about 4% had clear conflicts. Links with confirmed conflicts are excluded. Links with insufficient evidence are labeled as inferred. These sample findings do not measure accuracy or error rates across all published records.

Does a connection prove a referral or an instruction?

No. It shows representatives participating in linked international filings. For example, Firm A may handle the domestic filing, Firm B the PCT filing and Firm C the foreign filing. An A–C link can appear even if B arranged the foreign work. A Madrid basic filing and international registration can also name different representatives. A recorded role does not establish who selected or instructed foreign counsel; the applicant may have appointed counsel directly.

Is a representative’s current firm also their firm at the filing date?

Not necessarily. When a person is mapped to a firm using a current register, that mapping does not prove their historical affiliation. Read current profile information separately from the representative information on the filing record.

How should I interpret technology fields and client size?

Activity derived from patent IPC/CPC classifications and expertise declared on an official firm website are different kinds of evidence. A declared practice area alone does not establish prosecution experience. An unidentified trademark client size remains unknown; it should not be treated as an SME.

Are the figures a real-time view of the market?

No. Source publication, supplier delivery, extraction, validation and metric updates introduce delays. Unpublished filings are not visible. Source corrections and improvements to organization matching can change counts. A metric’s reporting period and a profile’s verification date may also differ.

Use the evidence to find your next partner.

Explore firms and their international activity, then review the records behind a promising connection.