Solve Intelligence

Patent work has a shape that suits this technology unusually well. It is document-heavy, highly structured, governed by strict formal conventions, and enormously repetitive at the level of language while being entirely bespoke at the level of substance.

It also has a property that makes automation genuinely risky: a drafting error can be discovered years later, in litigation, when it is unfixable and expensive. Claim language that seemed fine at filing can turn out to have given away the invention.

Solve Intelligence sits precisely on that tension. This review covers what the drafting tools actually handle, why examiner statistics change prosecution strategy, what the security posture means for a firm's professional obligations, and where the attorney's responsibility remains untouched.

What It Does For A Patent Practice

• Solve Intelligence is an AI platform built for patent professionals, covering the work from drafting an application through prosecution, litigation support and portfolio management.
• Drafting is the flagship, including figure generation from your files, chemical structure and Markush claim handling, and biological sequence listings — the specialised parts, not just the prose.
• Prosecution tooling is equally deep: office action analysis and response drafting, amendments and arguments, with examiner statistics folded in.
• Over 700 in-house IP teams and law firms are cited as users, including Intel, Siemens, DLA Piper, Finnegan and Quinn Emanuel.
• The assurance posture is unusually strong — SOC 2 Type II, ISO 27001, ISO 42001, stated no training on user data and zero-retention options with model providers.

Drafting Is Where The Specialised Work Is

General legal AI such as Harvey can write patent-shaped prose. What separates a serious platform in this field is the material that is not prose at all.

Figure generation from provided files addresses a genuinely tedious cost. Patent drawings follow strict formal conventions — numbered reference elements, consistent labelling across views, specific line conventions — and are usually outsourced to draughtspeople, adding days and expense to every filing.

Chemical structures and Markush claims are a specialism inside a specialism. A Markush claim covers a family of related compounds through a generic structure rather than listing each one, and getting its scope right is among the harder problems in chemical patent drafting. A general tool has no concept of it.

Biological sequence analysis and listings are a compliance requirement in biotech filings, with prescribed formats and their own submission standards. It is exacting, mechanical work that consumes qualified time.

Handling these three signals a product built with patent attorneys rather than adapted for them. They are also, notably, the parts of a filing where automation carries the least judgement risk, because correctness is largely formal rather than strategic.

The prose itself — specification, background, detailed description — is the part where a first draft saves real hours and where the attorney's revision is doing the actual work.

The prose in a patent is the part a model drafts well. The claims are the part that decides what you own, and that is not a drafting problem.

Examiner Statistics Change How You Prosecute

Of everything on the feature list, examiner statistics integration is the one most likely to change a decision rather than save time.

Patent prosecution is not uniform. Applications are assigned to individual examiners who have their own patterns — allowance rates, how many office actions they typically issue before allowance, how they respond to interviews, which rejection types they lean on.

Experienced practitioners have always known this informally, and firms with long histories at a particular art unit carry it as institutional knowledge. Making it explicit and queryable levels that considerably.

The practical effect is strategic. Whether to request an interview, whether to amend narrowly or argue, whether to file a continuation or appeal — these decisions have different expected outcomes depending on who is examining. Knowing that an examiner rarely allows without an interview is worth more than any drafting speedup.

The caution is the ordinary one about base rates: examiner statistics describe historical patterns across other people's applications, not a prediction about yours. They are an input to judgement, and the risk is treating them as a forecast.

The related tooling — invalidity charts, freedom-to-operate reviews, claim-overlap monitoring, portfolio pruning recommendations — follows the same logic. It is systematising work that was previously done by expensive people reading documents.

Security Is Not Optional Here, And They Have Treated It That Way

Most software reviews treat a certifications list as procurement filler. In this field it is closer to a threshold requirement, and Solve Intelligence's is unusually complete.

The reason is specific. An unfiled patent application is one of the most sensitive documents a company produces. Public disclosure before filing can destroy patentability outright in most jurisdictions. Attorneys also carry professional confidentiality obligations to clients that are not negotiable and do not bend for a software vendor's convenience.

So no training on user data and zero data retention options with the model providers are not features here, they are entry conditions. A platform that sent draft claims to an API that retained them would be unusable by any competent practice regardless of output quality.

The certification set is broad — SOC 2 Type II, ISO 27001, and ISO 42001, which is the newer AI management systems standard and still relatively uncommon to hold. GDPR and CCPA compliance, AES-256 and TLS 1.3, and enterprise AWS infrastructure round it out.

Contract-focused tools like LegalOn make similar assurances for a different practice area. The Thomson Reuters partnership is worth noting alongside this. Established legal information providers are conservative about who they associate with, and that association functions as a form of diligence in a market where a great deal is asserted.

1. Take a granted patent you drafted yourself and have the platform draft the same application from the disclosure.
2. Compare the claims specifically — that is where value or damage lives, not in the specification.
3. Run a real office action you have already responded to and compare its analysis with yours.
4. Check examiner statistics against an art unit where you have your own experience to test them.
5. Have your own counsel review the data processing terms before any live client matter touches it.

How A Firm Pays For This

This site does not publish plan prices and neither does the vendor. The structure is what a practice can plan around.

The model is contact-sales, with demo requests as the entry point rather than a signup form. In legal technology this almost always means seat-based licensing with firm-level agreements and annual terms.

The variables that usually drive the quote are the number of fee earners with access, which modules are included — drafting, prosecution, charting and portfolio analysis are distinct bodies of work — and whether the deployment needs specific data handling arrangements.

For a firm, the arithmetic is unusually legible compared with most software. Patent drafting is billed work with a known hourly value, so a time saving converts directly into either capacity or margin. That is a rarer situation than it sounds and it is why this category can sustain enterprise pricing.

The harder budgeting question is the one nobody puts on a page: if drafting time falls substantially, what happens to fixed-fee arrangements and to client expectations about cost? That is a business model question for the practice rather than a procurement one, and it is worth thinking about before the efficiency arrives.

The Responsibility That Does Not Transfer

This deserves stating plainly, because it is the one thing no platform in this category can change.

Claim drafting is where the value sits and where judgement is irreducible. Claims define the legal boundary of what you own. Too narrow and competitors design around them; too broad and they are invalid over prior art. That balance depends on commercial strategy, the specific prior art landscape, and what the client actually intends to sell — none of which is fully in the disclosure document.

The attorney signs it. Professional responsibility for what is filed does not move to a vendor. Duties of candour to the patent office, competence obligations, and malpractice exposure all remain exactly where they were.

Search platforms like Cypris sit upstream of this and face the same constraint. Prior art coverage is never complete. Applications stay unpublished for around eighteen months, so the most relevant prior art may not be visible to any system at the time you draft.

The realistic and still substantial value is compression of the mechanical hours — the specification prose, the figures, the sequence listings, the first pass at an office action response. That is a large share of the time and a small share of the judgement, which is exactly the right thing to automate.

• Claim scope remains a strategic judgement, not a drafting task
• Professional responsibility and malpractice exposure stay with the attorney
• Recent applications are unpublished for around eighteen months and invisible to everyone
• Examiner statistics describe history, not a prediction about your application
• Client confidentiality obligations bind you regardless of vendor assurances

Patent Vocabulary Used Above

Five terms that this listing assumes and that decide what the product is actually doing.

Office action

The patent office's written objection to an application. Responding to it — with amendments, arguments or both — is the bulk of prosecution work.

Claims

The numbered statements at the end of a patent that define the legal scope of what is owned. Everything else in the document supports them.

Invalidity chart

A table mapping each element of a patent claim to prior art disclosing it, used to argue a patent should not have been granted.

Markush claim

A chemistry claim covering a family of compounds through a generic structure rather than an explicit list. Notoriously difficult to scope correctly.

Prosecution

The back-and-forth with the patent office between filing and grant. Distinct from litigation, despite the name.

Coverage, Customers And Assurance