Your advisers screen dozens of deep-tech applicants and have to form a first view on each company’s IP before anyone with patent expertise is involved. IP Impact Index scores a whole cohort on public patent data and shows which portfolios deserve a specialist’s time — and what the reasoning rests on.
Built on public data from the EPO Open Patent Services API. Decision support, not legal advice.
EP 4 912 044 B1
Solid-oxide electrolyser stack with pressure-balanced ceramic sealing
Halvard Elektrolyse AS
Illustrative sample — fictional patent and applicant
The problem
Every cohort brings the same bottleneck. Escalate every applicant to an IP specialist and you exhaust a scarce, expensive resource on portfolios that did not need it. Escalate selectively on generalist instinct and the calls drift between advisers — and the reasoning is hard to reconstruct when someone asks, three months later, why this company was prioritised and that one was not.
This is a prioritisation problem, not a valuation problem. We are not trying to tell you what a patent is worth. We are trying to tell you where to look first.
How it works
The person running this is a generalist adviser. The output is built to be read by one.
A list of patent or application numbers, or an applicant name. Public data only — nothing confidential leaves your side.
Every patent is scored across technological importance, market potential, legal robustness and sustainability impact. Each dimension carries a grade and a confidence value, so thin data reads as thin data rather than as a low score.
A ranked cohort, a tier and a recommended action per patent, and an explanation that names the evidence it rests on.
What we measure
Each dimension is scored 0–100 with an A–F grade and a confidence value. The overall score weights them 30 / 30 / 20 / 20.
Forward and backward citations, how quickly the first citation arrived, non-patent-literature references, claim count, CPC/IPC class spread, and generality and originality indices.
9 weighted inputs
Geographic reach across the key markets (EP, US, CN, JP, KR), patent family size, EPC validation states, years maintained since grant, procedural stage, and sector value-added growth and R&D intensity through an IPC-to-NACE concordance.
7 weighted inputs
Grant status, prosecution health under Art. 94, claim-scope amendments during examination, opposition and appeal outcome, dependent-to-independent claim ratio, independent-claim word count, and X/Y prior-art citation risk.
9 weighted inputs
Y02 climate-mitigation classification, green IPC mapping, EU Taxonomy objectives, UN SDG alignment, and two external datasets — renewable-energy share in the patent's jurisdictions and greenhouse-gas intensity of its sector.
8 weighted inputs
Indicators are normalised on a calibrated scale with reference anchors drawn from typical patent cohorts. They are not live percentile ranks against a peer set, and we do not describe them as such.
Interactive demo
Five patents, scored once. Switch the audience lens and watch the ranking change — the pillar weights change, so what counts as a priority changes with them. Nothing is recomputed and no patent data moves.
Read this cohort as a…
Weighs commercial deployability and legal defensibility — can this be built on, and does it hold up?
Cleantech patents
All other patents
Switching lens re-weights the same four pillar scores. Nothing is recomputed and no patent data changes — only what the cohort is being ranked for.
| Patent | Investor Score | Tier | Open assessment | |||
|---|---|---|---|---|---|---|
| 1 | 70.6 | 78.7 | ATier A, Strong | Investment-ready: high conviction asset | ||
| 2 | 75.1 | 74.6 | BTier B, Solid | Investable with diligence on weakest pillar | ||
| 3 | 70.4 | 70.7 | CTier C, Mixed | Investigate before investing | ||
| 4 | 59.1 | 56.3 | DTier D, Weak | Not investment-grade; revisit if pivoted | ||
| 5 | 50.2 | 54.2 | ETier E, Dormant | Not investable in current form |
Ranked for the Investor lens. 1. Meridian Silicon Labs Ltd, tier A. 2. Halvard Elektrolyse AS, tier B. 3. Nordkap Energilager AB, tier C. 4. Rivelle Biocircular SAS, tier D. 5. Kestrel Navigation Systems GmbH, tier E
Select a patent to see its full assessment, the inputs behind each score, and the evidence the rules selected.
Tiers are ranks within this cohort, not absolute quality bands — the top fifth of any five-patent set is tier A. Assessments are decision support, not legal advice or a patent-attorney opinion.
Why it’s traceable
Language models are good at writing a sentence and bad at being accountable for one. So we split the job in two.
A deterministic rules engine decides what an assessment is allowed to draw on. 23 rules evaluate hard patent attributes — CPC classification, EP validation count, family geography, grant and opposition outcome, applicant type, sector — and select the policy and economic context that applies to this patent and this audience. Rules are sorted by priority with a fixed tie-break, so the same patent selects the same evidence every time, in the same order.
Only then does the language model write. Its input is the pre-selected evidence and the scores. Its job is four to six sentences a non-specialist can act on.
That division is the whole point. The part that decides what is claimed is inspectable and repeatable. The part that decides how it reads is not load-bearing. When someone asks why a portfolio was prioritised, you can show them the rules that fired and the attributes that fired them.
Deterministic layer
Language layer
Every conditional policymaker rule, evaluated against Sparse-attention inference scheduler for heterogeneous edge accelerators — a patent in the demo below. 5 of 7 fired; the 4-rule cap held one back. Rules that did not fire are shown too, because the ones that stayed silent are part of the record.
| Rule | Condition | Source | Result |
|---|---|---|---|
pol-tr-draghi-sectorpriority 5 | Classification maps to at least one strategic sector | Draghi 2024 | selected |
pol-tr-low-eppriority 5 | Fewer than 3 EPC validation states | Letta 2024, Innovation | no match |
pol-tr-4irpriority 4 | Classification falls inside the fourth-industrial-revolution boundary | EPO 4IR 2020 | selected |
pol-tr-ipr-intensivepriority 4 | Applicant's NACE sector is classified as IPR-intensive | EPO-EUIPO 2022 | selected |
pol-tr-y02-eupriority 4 | Y02 climate classification with an EU primary jurisdiction | EEA 2024 | no match |
pol-tr-eis-crosspriority 3 | Family spans member states in 2 or more innovation-performance groups | EIS 2024 | selected |
pol-tr-smepriority 3 | Applicant classified as an SME | EC SME Review 2024 | held back |
The scoring layer gives reproducibility and traceability. The AI layer turns evidence into a decision a non-specialist can act on. What we do not claim: that the language layer’s output is verified against the sources automatically. Evidence selection is deterministic today; automated citation checking is not built yet.
Who it’s for
The buyer is a programme director or head of innovation. The user is a generalist adviser, not a patent specialist. The trigger is a new cohort entering screening with a fixed amount of specialist attention to spend.
National and regional bodies assessing deep-tech applicants against public funding criteria.
Programmes selecting cohorts where IP is one of several signals and none of them get much time.
Advisory networks giving first-line IP guidance to firms without in-house expertise.
Smaller teams with wide remits and no dedicated patent specialist on staff.
Stated as goals, because that is what they are. We have a working system, not yet measured outcomes.
Early-stage investors, technology transfer offices and corporate innovation teams all have a version of this problem. They are adjacent markets, and we would rather be useful to one audience than plausible to four.
How we’re different
Professional patent-intelligence and prior-art search platforms are built for people who already know what they are looking for. They are excellent at that, and we are not trying to replace them.
We answer the question that comes before: out of everything in front of you right now, what deserves that expensive search in the first place? A generalist can read our output in about a minute and decide where to spend a specialist’s day. When they do open a professional platform, they will know what they are looking for.
Where they start
A specific question about a specific patent
Freedom to operate, prior art, claim charting, landscaping. Deep work, done by someone qualified to do it, on a target that has already been chosen.
Where we start
A cohort, and a limited amount of expert attention
Thirty companies, one adviser, and a decision due this week about which three are worth a specialist’s time. We narrow the field and say why.
We do not replace patent attorneys and we do not give legal opinions.
Origin
We met through our work at the European Patent Office and built the first version of IP Impact Index during EPO CodeFest 2026. The prototype scored real patents against live public data, and the response to it is why this is now a company.
The European Patent Office is not a partner, sponsor, customer or endorser of IP Impact Index. We use the EPO Open Patent Services API, which is a public interface available to anyone.
Get in touch
We are looking for a small number of innovation programmes to run a screening cohort with us. If you screen deep-tech companies and IP is part of the assessment, we would like to hear how you do it today.