Skip to main content
Product updateSignalsPatents

Patents are a buying signal. We rebuilt ours from the string up.

A granted patent tells you what a company is actually building, who built it, which firm they trust, and how fast their R&D is spinning up. The hard part was never finding patents — it was turning QUALCOMM INCORPORATED into a domain you can route and a person you can email.

·8 min read

Article Content

Most GTM teams treat patents as legal trivia. They are not. Raw patent data has always been unusable for go-to-market: the bulk grant file hands you a legal-entity string and an inventor name — not a company you can assign to a rep, not a contact you can reach. Turning that into a clean, routable signal is the whole job, and it is exactly what this relaunch fixes.

How many signals is this, really?

Our last weekly delivery was 24,213 patent signals — 11,223 company and 12,990 contact, split into two buckets on a universal schema. Patent grants and their inventor contacts do the volume; law-firm edges, velocity snapshots, and challenges layer on top. We expect this to climb toward 40,000 a week as coverage scales.

24,213
patent signals in the last weekly delivery
▲ scaling to ~40K / week
as coverage grows
5,647
Patent Grant — companies
12,990
Patent Grant — inventor contacts
3,108
Uses Law Firm
2,457
Portfolio Velocity
11
Patent Challenge

Delivered as two buckets — 11,223 company + 12,990 contact.

These are real delivered counts, not projections. Coverage widens as the filing history deepens, and none of the growth comes at the cost of a resolution shortcut — which is the part that actually matters.

Four subtypes, four go-to-market questions

One “new patent” event becomes four distinct signals — each routable on its own, or composed into a full account picture. Open any one to see the full signal envelope: every field, exactly as it ships.

Grab a sample of the feed

Six real signal envelopes — all four subtypes, company and contact, including a resolved inventor and an honest-blank — as newline-delimited JSON, the same format we deliver over GCS.

Download sample.jsonl

Patent signals

See the patent feed run against your target accounts

We will pull a live week of grants for your ICP — companies, inventor contacts, law-firm edges, and challenges — resolved and ready to route.

Why you can trust the resolution engine

A signal is only as good as the entity behind it. The bulk grant file gives you a legal-entity string — QUALCOMM INCORPORATED, GLOBAL HOLDINGS LLC — and a list of inventor names. Turning those into a company you can route and a person you can reach is the entire problem, and it is where most patent data quietly breaks. Here is how we do it, and why it holds up.

Raw grant string → resolved entity
QUALCOMM INCORPORATED
assignee string
qualcomm.com
verified domain
Priya Raghavan (IBM)
named inventor
/in/priya-raghavan-ml
verified profile
GLOBAL HOLDINGS LLC
ambiguous entity
null + logged
honest blank

CPC codes are how we know what a company actually does

Every grant carries Cooperative Patent Classification (CPC) codes — the patent office’s own taxonomy of what the invention is. We decompose them: the primary subclass maps to a technology area, and the mix of sections across a company’s portfolio tells us whether it is fundamentally a software, hardware, chemistry, or biotech shop. That derived picture is not just delivered as a field — it is fed back into resolution as the disambiguation hint, which is what separates a GLOBAL HOLDINGS LLC that files neural-network patents from the identically-named holding company that files nothing.

Primary CPC
G06F40/35
Subclass
G06F — digital data processing
Tech area
AI / machine learning
what_they_do (fed to the resolver) → “Builds LLM-based conversational-AI systems.”

Accepted only above a bar — and with a margin

We hand the assignee string, a one-line description distilled from the company’s own patent titles, and the CPC-derived technology area to our entity-resolution service. It returns a candidate domain, a confidence, and the margin over the runner-up. We accept only when the confidence clears our bar and the winner beats the next candidate by a margin — a near-tie is rejected even when the top score looks high, because that is exactly the case that produces confident-but-wrong matches. A junk-domain blocklist strips government sites, aggregators, free-email, and encyclopedia pages before anything ships.

How the resolver decides

accept ≥ bar · + margin

Every candidate match gets a confidence from our entity-resolution service. We accept a domain — or a LinkedIn URL — only when it clears the bar and beats the runner-up by a margin. Below that, the field is null and the reason is logged. The signal still ships; we just refuse to guess the identity.

Abstain → null, logged
Accept → verified
12345678910
QUALCOMM INCORPORATED9/10
qualcomm.com
P. Raghavan (IBM, AI/ML)9/10
/in/priya-raghavan-ml
GLOBAL HOLDINGS LLC6/10
null · logged
clears bar + margin → verifiedbelow → null (logged)every decision audited

Inventors run through the same service — resolved against the name, the employer we just resolved, the geography, and the invention itself. When a match clears the bar we deliver a LinkedIn URL; when it doesn’t, we ship the inventor name-only with a null profile. We never drop the person, and we never invent one.

The philosophy

Honest blanks over confident guesses.

When a match doesn’t clear the bar, we don’t drop the signal — we ship it with a null domain (or a name-only contact) and log exactly why. You get the event; you never get a fabricated identity. Every decision — accept, abstain, the margin, the number of attempts — is written to an audit trail, and every signal_id is a deterministic hash of the patent’s natural key, so re-running a week is a verified no-op instead of a pile of duplicates.

What you can read off a patent portfolio

Once the entities are clean, the filings themselves become a rich behavioral feed. The reads GTM and data teams pull from it:

  • What an account is building — the invention in plain language, tagged to a technology area and dated to the grant. Months ahead of the press release.
  • Where R&D is accelerating — trailing 30/60/90/365-day grant counts and a trend classification, so you can time outreach to the upswing rather than chase it.
  • What kind of company it really is — the CPC section mix separates a semiconductor shop from a fintech from a biotech, regardless of what the website says.
  • Who’s in a fight — PTAB challenges flag six-figure competitive disputes, with both the challenger and the defender resolved to real companies.
  • Which firms are inside the account — the outside counsel on the front page is a warm-intro path and a partner-channel target.

Rigor your data team trusts, plays your GTM team runs

The point of all of this: one feed your data team can defend line by line, and a set of plays your GTM team can run tomorrow.

For the data team

  • Gated resolution. Company → verified domain, inventor → LinkedIn — accepted only above a confidence bar with a margin check. Below it, the field is null.
  • Honest nulls, never fabrication. Fill-only against enrichment truth data; a junk-domain blocklist strips government, aggregator, and free-email matches; every decision is logged.
  • Deterministic IDs, clean upserts. Each signal_id is a UUIDv5 over the patent’s natural key — re-running a window is a verified no-op.
  • Split delivery. Company and contact records on a universal schema, JSONL or parquet, via GCS push or flat file.

For the GTM team

  • Route on real R&D. A grant tells you what an account is actually building, before the press release.
  • Time the reach-out. Velocity flags accounts whose filing pace is bending up.
  • Personalize with proof. The invention in plain language and the inventor by name.
  • Find the warm path. The outside counsel already inside the account.

Get the feed

Wire the patent feed into your stack

Public, structured, forward-looking — and almost nobody operationalizes it well. Clean entity resolution plus four complementary subtypes turn patents into routable, personalizable pipeline.

24,213 signals last delivery·4 subtypes·verified entities, or an honest blank