Corporate Communication

1. Foundation

Communication Goal

Ambition

World-leading Innovation Intelligence.

We define the market & customers we serve (high-stakes R&D)

the products standard we ship(TRUST-honest, traceable, expert, decision-ready)

and where we're headed by 2035.

We lead the category — and we set the bar for what qualifies as R&D-grade AI（AI-powered R&D engine ,purpose-built for high-stakes R&D）

AI-Native

Our product, our organization, and our business model are all AI-native by design, built to compound as AI itself compounds.

Advantage

Sharp differentiation from both legacy R&D tools and general-purpose LLMs. Our moat: proprietary data, deep fluency in real R&D workflows, and our Expert + FDE network.

Top 5 Brand Keywords

When customers search "Patsnap," these are what we want them to see. When they think of these concepts, we want Patsnap to be top of mind.

1. Innovation intelligence (solutions)

2. AI-driven R&D decisions

3. AI-powered R&D engine (purpose-built for high-stakes R&D)

4. Agentic innovation platform

5. IP Agent (Intellectual Property Agent)/ R&D agent

Narrative Architecture & Brand pyramid

"Yes" (Global leader in innovation intelligence) & "Defense" (Won't be replaced by LLMs; differentiated from both legacy competitors and new entrants; a Singapore-born global company )

Key message house

Roof — World-leading Innovation Intelligence

• Global leading innovation intelligence service provider (no footnote required — use anywhere)

• The world's largest provider of innovation intelligence solutions (always with footnote: by revenue in 2023, per CIC)

• oof —

3 pillar

World-leading Innovation Intelligence — powering smarter R&D decisions at every stage, from strategy to commercialization. （not only IP, strength R&D）

World-leading Innovation Intelligence — AI that R&D experts can trust./ where every answer is built to be defended /AI that lives in your domain（OUR AI value： TRUST-honest, traceable, expert, decision-ready.）

World-leading Innovation Intelligence for high-stakes R&D — shaped by 19 years inside the field./ rooted in the field since 2007.（18 years of R&D know-how. We are not a random pop-up AI company.）

World-leading Innovation Intelligence — powering smarter R&D decisions at every stage, from strategy to commercialization. （not only IP, strength R&D）

World-leading Innovation Intelligence — AI that R&D experts can trust.（OUR AI value： TRUST-honest, traceable, expert, decision-ready.）

World-leading Innovation Intelligence for high-stakes R&D — shaped by 19 years inside the field.（18 years of R&D know-how. We are not a random pop-up AI company.）

Pillar #1 — Agentic innovation platform for the R&D value chain

Golden quote: World-leading Innovation Intelligence — powering smarter R&D decisions at every stage, from strategy to commercialization. （not only IP, strength R&D）

we create an Agentic innovation platform that provide a AI-driven decision-ready solutions for our customers — three vertically integrated layers (data foundation → innovation ontology layer → agent layer) designed as one system.

Our 13 productized agents span strategy & scouting, ideation & feasibility, experimentation & simulation, IP protection & management, and commercialization & GTM validation. We're shifting from "software-as-a-destination" to "infrastructure that powers every R&D decision" — priced per decision, not per seat.

*"Concretely: your data foundation reads 2B patents. Your innovation ontology reasons across them — understanding the structure of claims, the causal logic of prior art, the taxonomy of technologies. Your agent layer answers 'should I launch this drug?' instead of giving you 10 links to go read. One system, one answer, one job done."*

Evidence: 6 product apps (Patsnap, Discovery, WorkSpace, Bio, Chemical, Citation) + 4 vertical domains + 13 productized agents; outcome-based pricing pilots underway

Pillar #2 — AI that R&D experts can trust

Golden quote: World-leading Innovation Intelligence — AI that R&D experts can trust. / where every answer is built to be defended / AI that lives in your domain

General LLMs are plausible but unverifiable — they summarize, suggest, and sometimes fabricate with equal confidence. For low-stakes tasks, that's tolerable. For R&D, one wrong call ends a strategy. A hallucinated prior art reference can sink a patent filing; a missed freedom-to-operate risk can derail a product launch. Trust in high-stakes R&D is not a feature — it's the foundation.

We don't claim trust as a marketing word. We engineer it as a four-part discipline:

Honest — 知之为知之，不知为不知

Our AI is built to admit boundaries rather than fabricate. When it doesn't know, it says so — because a wrong answer in R&D is worse than no answer. We apply an Honest Model Protocol: every output must be either grounded in evidence or explicitly flagged as uncertain. We'd rather give a professional user "I don't have enough data to answer this" than a plausible hallucination that leads to a $50M mistake.

Traceable — 每个结论都能追到专业数据与文献

Every output is grounded in source documents via RAG (Retrieval-Augmented Generation) and RAT (Retrieval-Augmented Thinking) — our agents retrieve real patents, papers, and technical literature, then rank and interpret the results. Chain-of-thought reasoning is visible before the agent acts. Every claim, every risk flag, every citation can be traced back to a specific document. This isn't a chatbot that gives you an answer — it's an analyst that shows you its work.

Expert — 深耕专利、技术与行业知识，懂专业工作流

Our model — PatsnapGPT — is trained on patent and technical literature, not general internet text. It understands that a claim is not a paragraph but a set of constraints, dependencies, and protection scopes. It knows the difference between a technical feature and a legal boundary. It reasons across 2 billion structured data points spanning 174 jurisdictions, embedded in the real workflows of 18,000+ R&D and IP teams. Frontier models can read a single patent well; we reason across the entire innovation landscape — understanding claim structures, citation networks, legal status chains, and technology evolution paths.

Decision-ready — 不是聊天，是可执行的判断与建议

Our output is not a conversation — it's a structured deliverable: claim charts, FTO reports, risk analyses, technology comparisons, whitespace maps. These are designed to enter business workflows directly, not to inspire further brainstorming. Guard rails restrict data and actions an agent can access, and require user approval for sensitive steps. The goal: from "here are some interesting links" to "here is your decision, with evidence, and here's why."

Evidence: PatsnapGPT (vertical LLM) + RAG/RAT architecture + Guard rails + Chain-of-thought auditability; 18,000+ R&D and IP teams in daily workflows; 92% reduction in compound search time (University of Edinburgh); 40x improvement in competitive technology identification (Goodyear); 94–95% reduction in patent review time (ArentFox Schiff)

Golden quote: "In R&D, one wrong call ends a strategy. Our AI is built to be honest about what it doesn't know, traceable in what it does, expert in the domain, and decision-ready in its output. Trust isn't a feature — it's the foundation."

Pillar #3 — MOAT: Depth in the R&D industry

Golden quote: World-leading Innovation Intelligence for high-stakes R&D — shaped by 19 years inside the field. / rooted in the field since 2007.

Patents are public the way tax returns are public — technically accessible, but structurally unreadable without 18 years of engineering behind you. A frontier model can read a single patent and summarize it. But it cannot tell you which claims block your product, how a patent family evolves across jurisdictions, where the whitespace is in a technology landscape, or why a competitor's filing pattern signals their next move. That requires reading public data to its full depth — and that depth is our moat.

Our moat operates on three levels, each progressively harder to replicate:

Level 1 — Data depth (not data access)

Having 2 billion patents and papers is not the moat — a startup with $100M and 24 months could assemble a similar corpus. The moat is what we've done to that data over 18 years: claim-level structuring (every claim decomposed into technical features, dependencies, and protection scopes), cross-modal alignment (chemical structures, gene sequences, and patent text mapped into a unified semantic space so they can be searched and compared), and temporal knowledge graph (technology evolution tracked as a time-series, not a snapshot). General LLMs process text; we compute over structured innovation objects. This engineering pipeline cannot be replicated quickly — it is a physical-time asset.

Level 2 — Proprietary assets that frontier models structurally cannot access

FDE（not ready yet） : A network of 200+ patent experts who curate comprehensive patent coverage — reading, annotating, and structuring the most complex and consequential patents within 24 hours of publication. This is human expertise compounded over 18 years into a scalable data asset. No algorithm alone can replicate expert judgment on claim scope, inventive step, and prosecution history.

Level 3 — Workflow embedding (the deepest moat)

18,000+ R&D and IP teams use Patsnap inside their daily workflows — not as a reference tool, but as the system where FTO assessments are conducted, claim charts are built, technology decisions are recorded, and compliance audit trails are maintained. When 5 different roles (scientist, IP counsel, IP manager, compliance, executive) collaborate on the same platform daily, the switching cost is not "replace a SaaS account" — it's "rebuild your entire R&D decision evidence chain." Frontier models, no matter how capable, cannot displace organizational process, compliance obligations, and multi-stakeholder lock-in. This is a socio-technical entanglement, not a technology problem.

The frontier test: If a startup with the latest frontier model + $100M + 24 months could replicate it, it's not a moat. Our Level 2 and Level 3 assets pass this test — they require physical time, exclusive relationships, and regulatory trust that money and models alone cannot buy.

Evidence: 18-year curation pipeline; FDE network of 200+ patent experts; Negative Knowledge Graph in development; cross-modal alignment (chemical structures, gene sequences, patent text); 18,000+ R&D and IP teams embedded in daily workflows; ISO 27001:2022 & SOC 2 Type 1 certified

Golden quote: "Data isn't the moat. The ability to read data to its full depth is — and depth is a function of time, expertise, and workflow embedding that no model can shortcut."

Key Defense" (Won't be replaced by LLMs; differentiated from both legacy competitors and new entrants)

Defense 1: Patsnap vs LLM: "Being Replaced" Is the Wrong Frame

Golden Quote：Frontier models reason. Patsnap decides.

"They have the smartest general reasoning engine on the planet. We have an engine that understands how innovation works. The frontier model is a brilliant junior scientist. Our middle layer is the 30-year veteran who knows why that experiment is going to fail before you run it. Then we put agents on top of that ontology. That's not competition. That's co-evolution."

Core Defence Logic

1. The Wrong Frame: Patsnap is not an NLP tool, and customers don't "use" it — their decisions run on it.

LLMs are the language layer. Patsnap is the execution layer. Customers don't "open" Patsnap and use it; they run a hundred-million-dollar R&D decision on top of Patsnap. That distinction is fundamental — we have moved from being a tool to being a substrate.

Analogy: You don't "use" the electrical grid; your devices run on it. You don't "use" the Bloomberg Terminal; your trading decisions run on Bloomberg's data network.

2. Falling inference costs are a structural tailwind for Patsnap, not a headwind.

We are buyers on the LLM-consumption side, not competitors. Every reduction in inference cost lifts our gross margin. Frontier models drop their cost by an order of magnitude every 18–24 months — every drop is a margin uplift for us.

Lightweight companies — those wrapping a single layer on top of an LLM — will be killed by LLM progress. Companies that own a complete Data + Ontology + Agent system see the opposite effect: the stronger the LLM, the stronger our system's output, the higher the value we deliver.

3. Frontier models are not our competition — they are a distribution channel.

The APIs from OpenAI, Anthropic, and Google are not competing with us; they are one of our distribution channels. When a customer asks ChatGPT a patent question, ChatGPT can call Patsnap's MCP interface to answer it. The stronger the frontier model, the more customers it reaches, and the more questions it surfaces that it cannot answer alone and must route to Patsnap. OpenAI and Patsnap are not in a substitution relationship — they are in a feed-in relationship.

4. Our commercial model is correlated with, not against, frontier-model progress.

We do not charge license fees. We charge three things:

Decision fee — charged per R&D decision executed

Token fee — usage-based pricing on MCP interface calls

Sovereign deployment fee — annual fee for private, on-premise deployments

Frontier model prices fall → more customers use AI → more decisions run on Patsnap → Patsnap revenue rises. The direction is the same, not opposite.

5. Customers don't pick sides. They use both.

Over 18,000 customer R&D teams use ChatGPT (for email drafting and brainstorming) and Patsnap (for freedom-to-operate, decision audit trails) every day, in parallel. The two tools do entirely different things. Customers don't need to close one to open the other.

Our ultimate goal is not to make customers "open Patsnap" — it is to embed Patsnap into the tools they already open every day: Microsoft Word, CAD, PLM systems, Anaqua, Notion, SharePoint. The user uses Patsnap without knowing they are using Patsnap. Distribution is king — not product features.

Defence 2: We vs other AI company

Golden Quote：They built a feature. We built the field.

"Every few quarters, a new IP agent startup raises a round and a headline. They are right that AI is rewriting how patent work gets done — we agreed with that thesis 19 years ago, when they were not yet companies. The difference is this: they are building a product on top of a market that we built. Drafting an office action faster is a feature. Owning the decision substrate that 18,000 R&D teams have committed three years of audit trails to — that is a field. Features get acquired. Fields get listed."

Core Defence Logic

1. Wrong frame: This is not "new AI vs. old incumbent." This is "feature company vs. field company."

The recent wave of "IP agent" startups — Solve Intelligence (US/UK, $55M raised), IPRally (Helsinki, $13.4M), and analogous players — are building point solutions on top of a market category. Patsnap, in contrast, is the market category.

A feature company makes one task faster. A field company holds the substrate on which entire R&D operations run. A new entrant cannot raise their way into being a field — they can only raise their way into being acquired into one.

The frame "old vs. new" is the frame they want you to adopt. The accurate frame is "single workflow vs. full innovation stack."

2. The "AI-native" narrative is younger than our product cycle.

The label "AI-native" became a credible market position roughly when reasoning models reached enterprise readiness — approximately 19 months ago. Patsnap has been embedded in R&D workflows since September 2007. That is not a marketing gap. That is a generational gap.

Every new IP-agent startup making an "AI-native" claim is, by definition, less than two years old as a credible enterprise vendor. Less than two years is not enough time to:

Accumulate failure data from real customer projects

Develop a Causal Mechanism Graph validated across thousands of R&D decisions

Embed Field Domain Experts inside customer workflows long enough to extract methodology knowledge

Generate three years of decision audit trails that customers cannot afford to lose

"AI-native" is a starting line. We are at kilometre nineteen.

3. Drafting speed is a feature. Decision integrity is a field.

The most-funded new entrants concentrate on patent drafting, office action response, and claim charting — three workflows where generative AI shows fast wins because the output is a document.

Patsnap does not compete on document velocity. Patsnap operates on a different layer:

Freedom-to-operate analysis that must defend against future litigation

Patent validity assessment that feeds investment-grade decisions

Molecule advancement and competitive technology landscape that drives multi-year R&D portfolios

Decision audit trails that withstand regulatory and board-level scrutiny

A document is delivered and forgotten. A decision is defended for years. The two are not the same product category, regardless of surface similarity.

4. Single-workflow tools concede the integration battle by design.

New IP agent startups optimise for one workflow because that is what a small team can build, and what a fresh round can finance. They cannot ship simultaneously across drafting, prosecution, FTO, validity, landscape, scouting, and commercialisation — because each layer requires years of domain calibration.

The customer reality: R&D decisions cross workflows. An invention disclosure flows into drafting, which flows into prosecution, which flows into landscape comparison, which flows into FTO, which flows into commercialisation. Single-workflow tools force customers to stitch tools together — meaning the audit trail is fragmented across vendors, and the customer carries the integration burden.

Patsnap is the layer where the stitching is already done.

5. Velocity is not a moat. Distribution and embedment are.

A well-funded startup with a strong AI team can build a faster drafting tool. They cannot, in the same timeframe, build:

18,000+ embedded customer R&D workflows

217 Field Domain Experts inside customer environments

Three years of decision audit trails per major customer

Patented Causal Mechanism Graph and Innovation Ontology Layer

A distribution presence inside Microsoft Word, CAD, PLM, Anaqua, Notion, and SharePoint where customers already work

Velocity makes a product visible. Embedment makes a product irreplaceable. New entrants compete on visibility. Patsnap competes on irreplaceability.

6. The new entrants validate our market — they do not displace it.

Solve Intelligence describes the patent industry as a $200B+ market. Their existence — and their funding — confirms what Patsnap has been compounding against for 19 years. Every dollar a new entrant raises is a dollar that publicly underwrites the value of the category Patsnap defined.

A rising tide does not displace the deepest harbour. It justifies it.

Evidence

Time Inside the Field

Patsnap: founded September 2007 — 19 years of R&D workflow embedment

Solve Intelligence: founded 2023 — two years

IPRally: founded 2018 — eight years

Most new "IP agent" entrants: less than 19 months as credible enterprise vendors

Defence 3: We are a Chinese company pretend to be a Singapore company

Golden Line: A Singapore-headquartered neutral platform in a polarized world

Patsnap is a global company.

Innovation has gone geopolitical. Every side — US, China, EU, India, the Middle East — needs innovation intelligence infrastructure that is trusted and neutral. A US-domiciled frontier model company cannot serve Chinese SOEs. A China-domiciled company cannot serve the US DoD. Patsnap, headquartered in Singapore with complete data isolation by design — US customer data is fully walled off from Chinese customer data; the two cannot see each other — uniquely serves every geography. Our revenue mix is structurally diversified. This isn't a liability; it's the single most defensible structural advantage we have, and one no US or China competitor can replicate.

Evidence: Revenue split — Greater China 48.6%, Americas & EMEA 34.1%, APAC ex-China 17.3%; ISO 27001:2022 & SOC 2 Type 1; data isolation by design; Singapore HQ

How we work

Dokki

2. Legal Guidance & Compliance

Timeline: Tailored for Kinabalu Process

Kinabalu Publicity Compliance Guideline:

Key Training Reminders (One-Liner Takeaways)

When in doubt, remain silent — any question must be referred to Legal

"Ordinary course communications" do not mean "publish at will" — frequency, media, and tone must remain consistent with historical practice

Forward-looking statements are the high-voltage line — words such as "sustained growth," "increased investment," and "All-in" all require prior approval

Statements by senior officers cannot be ring-fenced — statements by senior officers in any setting will be treated as statements of the Company

Websites and social media are high-risk areas — all updates must go through the approval process

The U.S. dimension cannot be omitted — even though the principal jurisdiction is Hong Kong, U.S. rules apply mandatorily and concurrently

Standard response to media inquiries — do not answer, do not comment, refer to the designated spokesperson

Maintain recordings on the record — before any media interview, inform the counterparty of the recording and obtain their consent

If something is said in a public interview that is not in the public domain, it must, by definition, not be sufficiently important

With respect to forecasts and matters that have not yet occurred: avoid quantitative figures and specific strategies, and confine statements to macro-level direction

PR-Legal Mechanism

Dokki + Auto-examination

3. LT PR - SPOKESPERSON

CEO PR/ CTO PR - Sofie

Ray PR- Elizabeth/Kate

4. PR crisis Management 

Mechanism

Mechanism I: Risk Mapping

Owner: PR — Risk Identification Workstream

Historical Audit: Conduct a systematic review of all of the CEO's public statements, commercial actions, and social affiliations, performed once as a comprehensive sweep, then maintained on a rolling basis. The audit is conducted from an adversarial perspective — examining the record as a hostile researcher, journalist, or short-seller would. The first execution is a full sweep; from then on, the workstream shifts to scenario planning and continuous maintenance.

The seven categories of exposure to be mapped:

1. Statements — public remarks, interviews, social-media activity, leaked internal communications

2. Behaviour — private life, health, family, photographs from public or semi-public settings

3. Decisions — workforce reductions, acquisitions, product sunsets, pricing changes, partnership terminations

4. Historical record — resurfaced historical statements, early commercial conduct, verification of educational and professional credentials

5. Affiliations — CEO's family, friends, investment positions, and board memberships that could create exposure through association

6. Values — positioning on political, religious, geopolitical, gender, and racial issues

7. Operational capability — earnings disclosures, product incidents, security events, litigation

Mechanism II: Intelligence System

Owner: PR — Intelligence Workstream

Guiding principle: The KPI of the intelligence system is not "zero negatives" — it is "zero surprises." Surprises are what kill. A negative event that the company sees coming 48 hours in advance is manageable. A negative event the company learns about from the front page is a crisis.

• Media and social monitoring: 24/7 coverage across major languages, primary platforms, and vertical communities — including Reddit, Hacker News, X, Zhihu, Xiaohongshu, Maimai, Glassdoor, and Blind. Anonymous internal communities are frequently the first scene of a crisis. We recommend either procuring a dedicated monitoring system or engaging a specialised monitoring agency.

• External information sources: Maintain a long-term, low-visibility relationship with three to five top-tier PR firms, industry journalists, and government-relations advisors. These external sources typically hear signals before internal channels do.

Mechanism III: Quick response- Crisis management

Owner: Crisis Core

1. Establish a permanent Crisis Core of three to five members. Recommended composition for Patsnap: CEO, Legal, PR, HR VP (the natural channel for employee-sourced intelligence), and the senior finance lead (the natural channel for capital-market intelligence).

2. Allocate a standing budget. Crisis response is not a free activity. Costs include monitoring systems, external legal counsel, paid placement, media engagement, and external communications counsel. Budget must be ring-fenced and pre-approved so that response is not delayed by procurement cycles.

3. Establish a tiered response framework — pre-defined trigger thresholds that determine which Crisis Core members convene, on what timeline, and with what decision authority.

Mechanism IV: Scenario Playbooks

Owner: PR + Legal

Pre-build scenario scripts, response messaging, and decision authority matrices for the highest-probability crisis types. Maintain these as living documents — reviewed monthly, updated after every material event. Reference: Patsnap Risk Response Playbook (separate document).

Mechanism V: Continuous Drills

Co-owned: PR + Legal

• Executive training: The CEO, senior executives, and corporate-account spokespersons participate in regular compliance and media-handling briefings led by PR, Legal, and Compliance.

• Build narrative buffers into the system: deliberate mechanisms that absorb risk before it reaches the CEO. Examples include delayed-release protocols for sensitive announcements and designated spokespersons to handle sensitive topics — so that risk is metabolised at the spokesperson layer before it touches the CEO's direct exposure to the public.

• Quarterly tabletop exercises: Each quarter, the Crisis Core convenes for a structured simulation covering risk identification, response coordination, and sensitive-issue messaging. The point of a tabletop exercise is not "what do we do if X happens" — it is "what do we do if X, Y, and Z happen simultaneously." Real crises never arrive alone. Adversaries never fire a single round.

• Post-event review: Within 72 hours of every material incident, the Crisis Core conducts a structured review — what was detected, when, by whom; what was decided, when, by whom; what worked, what did not, and what changes to the framework are required.

Current Highlight

Glassdoor

US DOD

Structural Risk

Discussion & Next steps：

1. How we work:Dokki Tutorial meeting

2. Company Branding Pyramid/ Product naming system/ Visual

3. Media List and Media plan alignment

4. Content control : PR-Legal / AI generated content

5. Social Media map out and self-check

6. Risk Mapping out - and Follow up on Current risk we found out