Stilta growth playbook

Build /benchmarks as the single source of truth. Give each study (PTAB invalidity recall, Polhem examiner-citation recall, the KUKA infringement re-run, Maersk product mapping, portfolio-scale claim-chart cost) its own card with the question, the ground truth, the sample, the result, the date, the limitations and a link to the full write-up. Pick one definition per number: for example, "71% recall (95% CI 64-77%) vs. 40% for the best search tool", or say outright that 35% is an average. Then use exactly that figure everywhere and link every homepage stat to its card. Publish the promised PTAB dataset (IPR identifiers and challenged patents) as a download. Add a "Run this benchmark on your own patents" form, which is already the study's CTA, as the page's main action.

For a product whose promise is defensible, traceable evidence, consistent and cited numbers are the pitch itself. An IP partner who finds a mismatch on the homepage will doubt the claim charts too. With a reproducible dataset, Stilta becomes the vendor that others have to measure against, while Patlytics publishes case-study ROI and no benchmarks.

Number | Homepage today | Source today | Pick one
PTAB recall | 71% vs ~35% tools, 18% LLMs (no link) | 71%; +31pp tools, +49pp LLMs | 71% vs __% (best tool / average)
Runtime | "30 min" and "about 20 minutes" | "21m" header, 17 min median | __ min (median, input to first draft)
Scale/cost | "8-13x faster and 200x more scalable" | 8-13x speed, ~200x cost-efficient | "8-13x faster, ~200x cheaper per comparison"
Patents indexed | 200M+ | Polhem: 170M+ documents | define patents vs. documents
[ ] PTAB dataset CSV published
[ ] Every stat links to /benchmarks
[ ] /benchmarks and /news/introducing-polhem in sitemap

• Restating 35% as 40% looks like a correction

  Show both figures with their definitions (best tool vs. average), so nothing is quietly changed.

• Competitors use the dataset to tune their tools

  That is the point of an open benchmark; keep a held-out set for each new edition.

• Your websiteHomepage numbers: a "30 min" runtime "From input to first draft"; the FAQ says Stilta "replaces the search in about 20 minutes"; "8–13x faster and 200x more scalable"; "We index 200M+ patents across 110+ jurisdictions"; and "In our public PTAB benchmark across 40 IPR institution decisions, Stilta surfaced 71% of cited prior art, versus around 35% for other patent AI tools and 18% for general LLMs", with no link to the study.Source
• Your websitePTAB study (February 17, 2026): 40 IPR institution decisions from February 2023 to January 2026; "71% petition recall across 40 cases, with a median search time of 17 minutes per patent", while the header shows "21m" and callouts of "+31pp" vs. search tools and "+49pp" vs. LLMs; the five tools and three LLMs compared are not named; Stilta also surfaced 161 references not cited in the petitions; IPR prior-art search typically costs $15,000–50,000 and takes 4–8 weeks; the authors write "We are publishing the full benchmark dataset - including IPR identifiers and challenged patents", but the page has no download link; CTA "Want to run this benchmark on your own patents?"Source
• Your websitePortfolio-scale model (October 15, 2026): Stilta post-trained an NVIDIA model on litigation claim charts and EPO examiner decisions; 5,000 claim charts per hour on one GPU, "8–13x faster than open reasoning models"; $0.65 per 1,000 comparisons versus $132 for a leading frontier model, "about 200x more cost-efficient"; a 10,000-family portfolio analysed every month for about $1,600 a year instead of $320,000; early access is limited to a small number of enterprises and firms.Source
• Your websitePolhem 1.0 (November 5, 2026): Stilta's own 345M-parameter search model over more than 170 million patent documents, trained on examiner citations; on 3,969 held-out applications, 72.6% of examiner-cited art is in the top 1,000 results, 11.5 points above the strongest open model tested (over 20 times as many parameters); first on the public DAPFAM and perf200 benchmarks.Source
• Your websiteSitemap (23 URLs, checked November 9, 2026): no lastmod dates; 2 product pages, 13 news posts and legal pages; missing the three newest posts linked from /news (/news/introducing-polhem, /news/product-update-november-26, /news/product-update-september-26) and /webinar; no pricing, benchmark, workflow, practice-area or comparison pages.Source
• Competitor research&AI's own "&AI vs Stilta" page calls Stilta "a newer entrant (founded 2026)", says its court usage should be confirmed, suggests it may fit US IPR and district court work less well because it is not US-headquartered, and offers a 7-day free trial of &AI.See the research Source
• Competitor researchPatlytics homepage: "The Premier AI-Powered Patent Platform"; 160M+ patents and 300M+ literature items; a $40M Series B; ROI case studies such as an Am Law 100 firm finishing a 120-patent project in 20 hours instead of 100 for about $38,000 in added margin; SOC 2 Type 2, ISO 27001 and ISO 42001; no published benchmarks on the homepage.See the research Source

Turn the KUKA format into the standard first meeting. The buyer submits a patent number from a matter they've already closed, keeps their conclusion to themselves, and gets a blind Stilta run plus a 30-minute call comparing the two. Add a short form (patent number, type of question: invalidity, infringement, FTO or SEP, and the region) on the homepage, both product pages and /benchmarks. Write up results that customers agree to share as short benchmark notes.

A lawyer who already knows the right answer can judge the output in minutes, with no data-sharing worries, since only public patent numbers are involved. It also turns each demo into a possible case study, which is what KUKA and Maersk became.

• A blind run misses the buyer's answer

  Show what was searched and let the buyer steer a second pass on the call, which is the product's real use.

• Too many low-value requests

  Ask for a company email and the matter type, and prioritise in-house teams and firms.

• Your websiteKUKA re-run (September 24, 2026): KUKA's IP team gave Stilta a completed infringement case with only the patent ID and public evidence; one autonomous run of about 20 minutes scanned 500+ references, reached the same no-infringement verdict and found the same decisive gaps, in 85–90% less time than the manual assessment.Source
• Your websiteMaersk benchmark (September 10, 2026): from patent IDs only, Stilta mapped patents to Maersk products with 88% accuracy (143 of 163) and 97% (30 of 31) where public evidence taught every limitation, across 500+ patents, 150+ invention families and 2,000+ claim limitations; competitor products are the next stage.Source
• Your websiteDeep Analysis page: five workflows on one page (invalidity, infringement, freedom to operate, patentability, standard essentiality) reached by on-page anchors; a table contrasting "Days to weeks per search" with "Results in minutes" and "High cost per search" with "About a third of the cost"; no testimonials; a "Book a demo" button in every section.Source
• Your websitePTAB study (February 17, 2026): 40 IPR institution decisions from February 2023 to January 2026; "71% petition recall across 40 cases, with a median search time of 17 minutes per patent", while the header shows "21m" and callouts of "+31pp" vs. search tools and "+49pp" vs. LLMs; the five tools and three LLMs compared are not named; Stilta also surfaced 161 references not cited in the petitions; IPR prior-art search typically costs $15,000–50,000 and takes 4–8 weeks; the authors write "We are publishing the full benchmark dataset - including IPR identifiers and challenged patents", but the page has no download link; CTA "Want to run this benchmark on your own patents?"Source

Generate sitemap.xml from the same content source that powers /news, with a lastmod for every URL. Add the missing posts and any new pages (/benchmarks, workflow pages) automatically. Submit it in Search Console and Bing Webmaster Tools, and request indexing for the Polhem post.

Search engines and AI assistants find new research faster when the sitemap is complete and dated, and Polhem is the post most likely to earn links from researchers.

• Old slugs and new slugs both listed

  List only canonical URLs and redirect any duplicates.

• Your websiteSitemap (23 URLs, checked November 9, 2026): no lastmod dates; 2 product pages, 13 news posts and legal pages; missing the three newest posts linked from /news (/news/introducing-polhem, /news/product-update-november-26, /news/product-update-september-26) and /webinar; no pricing, benchmark, workflow, practice-area or comparison pages.Source
• Your websitePolhem 1.0 (November 5, 2026): Stilta's own 345M-parameter search model over more than 170 million patent documents, trained on examiner citations; on 3,969 held-out applications, 72.6% of examiner-cited art is in the top 1,000 results, 11.5 points above the strongest open model tested (over 20 times as many parameters); first on the public DAPFAM and perf200 benchmarks.Source
• Your websiteNews page: 15 posts from January to November 2026, including product updates in June, July, August, September and November, the KUKA and Maersk benchmarks, the LexisNexis and NVIDIA partnerships, the funding round and the October 13 rebrand.Source

Replace the closed notice with the recording, cut into chapters (invalidity walkthrough, the new Home view, Analyst, Portfolio analysis, Q&A). Put a short email gate on the full video, or leave it open with a booking link under it, and offer the blind re-run (idea 2) as the next step. Clip the best two minutes for LinkedIn and the homepage.

A recorded walkthrough lets an attorney see the product without booking a call, and lets the champion forward it to partners, which matters with a four-person team that can't attend every first meeting.

• The demo shows customer data

  Blur or re-record any screen with non-public matters.

• Your websiteWebinar page: "See the new Stilta, live." on 2 and 3 November, 30 minutes (a 20-minute walkthrough of deep analysis and 10 minutes of Q&A); it now says "Registration for this webinar is now closed" and offers no recording.Source
• Your websiteNovember product update (November 4, 2026): a new interface with a Home view of items needing attention; search "up to 10–50x faster"; Analyst (beta) over prosecution and litigation history; Portfolio analysis (beta) with full release later in November; Boards (coming soon) to monitor "patents, assignees, and topics"; a live walkthrough at stilta.com/webinar.Source
• Your websiteSitemap (23 URLs, checked November 9, 2026): no lastmod dates; 2 product pages, 13 news posts and legal pages; missing the three newest posts linked from /news (/news/introducing-polhem, /news/product-update-november-26, /news/product-update-september-26) and /webinar; no pricing, benchmark, workflow, practice-area or comparison pages.Source

Build /pricing to explain the two models, what counts as a "project", what a site licence includes (users, matters, portfolio analysis, data residency, SSO), and a worked comparison: one invalidity search with Stilta vs. a search firm at $15,000-50,000. Give a "from" price or a range per project if the founders are comfortable; if not, say "typically a fraction of an outside search, quoted per matter". End with the blind re-run offer.

No per-user pricing is a real advantage for firms used to seat licences, and a public explanation shortens procurement. The $15,000-50,000 anchor comes from Stilta's own research, so the comparison is already cited.

• Published prices anchor big deals low

  Publish the model and a per-project range only; keep site licences quoted.

• Your websiteHomepage proof and offer: H1 "AI agents for patent analysis", "Trusted by 100+ IP teams", logos including KUKA, Alfa Laval, Mannheimer Swartling, Bergenstråhle, Loza & Loza and Desmarais LLP; testimonials from KUKA, Spotify's Head of IP, Bergenstråhle and Mannheimer Swartling; six practice areas (telecommunications, life science and medtech, chemistry and materials, mechanical and industrial, robotics and automation, electronics and software); pricing only in the FAQ: a "per-project basis, paying only for the cases you bring through" or a flat-fee site licence, and "We do not charge per user"; the FAQ also says "Stilta is ISO 27001 certified, GDPR compliant, and EU AI Act aligned." There is no pricing page (/pricing returns 404) and no newsletter signup.Source
• Your websitePTAB study (February 17, 2026): 40 IPR institution decisions from February 2023 to January 2026; "71% petition recall across 40 cases, with a median search time of 17 minutes per patent", while the header shows "21m" and callouts of "+31pp" vs. search tools and "+49pp" vs. LLMs; the five tools and three LLMs compared are not named; Stilta also surfaced 161 references not cited in the petitions; IPR prior-art search typically costs $15,000–50,000 and takes 4–8 weeks; the authors write "We are publishing the full benchmark dataset - including IPR identifiers and challenged patents", but the page has no download link; CTA "Want to run this benchmark on your own patents?"Source
• Your websiteDeep Analysis page: five workflows on one page (invalidity, infringement, freedom to operate, patentability, standard essentiality) reached by on-page anchors; a table contrasting "Days to weeks per search" with "Results in minutes" and "High cost per search" with "About a third of the cost"; no testimonials; a "Book a demo" button in every section.Source
• Competitor researchIPRally homepage: "The patent engine for humans and agents", with its engine available through MCP and API; 5,000 professionals and 6 patent offices; case studies such as RPX finding 40% more cited references; a linked pricing page; demo replies promised within 2 business days.See the research Source

Pick the accurate status and use one sentence everywhere: for example, "ISO/IEC 27001:2022 certified (certificate in the trust center)", or "aligned to ISO 27001, certification in progress". Do the same for the EU AI Act wording ("aligned" vs. "built to the standard") and GDPR. Add the certificate's scope and issue date to the trust center summary so reviewers don't have to request access just to confirm the status.

Security review is often the slowest step in enterprise legal deals. Consistent wording removes a question from every questionnaire, and Patlytics, Solve and IPRally all state their ISO status plainly on their homepages.

• Downgrading the wording looks like bad news

  Accurate wording now costs much less than a reviewer finding the gap later.

• Your websiteHomepage proof and offer: H1 "AI agents for patent analysis", "Trusted by 100+ IP teams", logos including KUKA, Alfa Laval, Mannheimer Swartling, Bergenstråhle, Loza & Loza and Desmarais LLP; testimonials from KUKA, Spotify's Head of IP, Bergenstråhle and Mannheimer Swartling; six practice areas (telecommunications, life science and medtech, chemistry and materials, mechanical and industrial, robotics and automation, electronics and software); pricing only in the FAQ: a "per-project basis, paying only for the cases you bring through" or a flat-fee site licence, and "We do not charge per user"; the FAQ also says "Stilta is ISO 27001 certified, GDPR compliant, and EU AI Act aligned." There is no pricing page (/pricing returns 404) and no newsletter signup.Source
• Your websiteSecurity page: "Our information security management system is aligned to ISO 27001."; a SOC 2 Type II audit and monthly external penetration tests; Azure hosting with EU customers in Sweden and US hosting for US customers; no training on customer data; SSO and audit log export; reports in the trust center.Source
• Your websiteTrust center (trust.stilta.com, on SafeBase): lists SOC 2 Type 2 and ISO/IEC 27001:2022 under Compliance; documents sit behind "Get access".Source
• Competitor researchPatlytics homepage: "The Premier AI-Powered Patent Platform"; 160M+ patents and 300M+ literature items; a $40M Series B; ROI case studies such as an Am Law 100 firm finishing a 120-patent project in 20 hours instead of 100 for about $38,000 in added margin; SOC 2 Type 2, ISO 27001 and ISO 42001; no published benchmarks on the homepage.See the research Source

Serve the hero and logo images through Next.js image optimisation (AVIF/WebP at display size), lazy-load everything below the fold, split or defer any script the hero needs so the headline text renders first, and fix the two accessibility issues (colour contrast and touch-target size).

A lighter page keeps first impressions fast on mobile and on slower connections as US traffic grows, and the fixes take hours.

• Brand visuals lose sharpness

  Check AVIF quality on retina screens before shipping.

• PageSpeedHomepage, mobile: performance 63/100 (desktop 84); lab largest paint 15.8 s on a slow phone, but real visitors pass Core Web Vitals (LCP 2.1 s, INP 95 ms, CLS 0); 5.3 MB page weight, 4.1 MB of it images, with about 2.7 MB saveable through better image delivery and 260 KB of unused JavaScript (about 1.2 s); two accessibility issues (colour contrast and touch-target size).Source

Add a one-field "Monthly updates" signup to /news, the footer and the end of every post. Send each product update and benchmark as a short email: what shipped, one screenshot, one benchmark, one link back into the app. Segment customers, pilots and prospects so pilots get a "try this on your matter" prompt.

Stilta already writes the content, and email brings pilot users back into the product, which helps pilots convert and keeps partners at firms thinking of Stilta.

• Consent rules for EU contacts

  Only email contacts who opted in or are existing customers, with unsubscribe links.

• Your websiteNews page: 15 posts from January to November 2026, including product updates in June, July, August, September and November, the KUKA and Maersk benchmarks, the LexisNexis and NVIDIA partnerships, the funding round and the October 13 rebrand.Source
• Your websiteHomepage proof and offer: H1 "AI agents for patent analysis", "Trusted by 100+ IP teams", logos including KUKA, Alfa Laval, Mannheimer Swartling, Bergenstråhle, Loza & Loza and Desmarais LLP; testimonials from KUKA, Spotify's Head of IP, Bergenstråhle and Mannheimer Swartling; six practice areas (telecommunications, life science and medtech, chemistry and materials, mechanical and industrial, robotics and automation, electronics and software); pricing only in the FAQ: a "per-project basis, paying only for the cases you bring through" or a flat-fee site licence, and "We do not charge per user"; the FAQ also says "Stilta is ISO 27001 certified, GDPR compliant, and EU AI Act aligned." There is no pricing page (/pricing returns 404) and no newsletter signup.Source
• Your websiteFunding post (May 19, 2026): $10.5M led by Andreessen Horowitz with Y Combinator; since launching in February 2026 Stilta "signed Roche, Alfa Laval, and Maersk" and "Three of the five largest IP firms globally are customers or in active pilots"; the money hires engineers, go-to-market staff and patent experts in Stockholm and New York; the CEO: "When one company starts using AI for patent enforcement, every competitor has to follow."Source

Build /invalidity, /infringement, /freedom-to-operate, /patentability and /standard-essentiality. Each page gets the workflow in four steps, a redacted sample output built from public patents (claim chart, ranked references, documented reasoning), the matching benchmark (PTAB for invalidity, KUKA for infringement), the export formats (PDF and Word, from the September update) and a blind re-run CTA.

Separate pages match how practitioners search and brief each other, and a sample output answers the first question every attorney asks: what does it actually produce?

• Sample outputs read as legal advice

  Label them as illustrative analyses of public patents.

• #1 Put every benchmark on one cited /benchmarks page, and make the homepage numbers match it

• Your websiteDeep Analysis page: five workflows on one page (invalidity, infringement, freedom to operate, patentability, standard essentiality) reached by on-page anchors; a table contrasting "Days to weeks per search" with "Results in minutes" and "High cost per search" with "About a third of the cost"; no testimonials; a "Book a demo" button in every section.Source
• Your websiteSitemap (23 URLs, checked November 9, 2026): no lastmod dates; 2 product pages, 13 news posts and legal pages; missing the three newest posts linked from /news (/news/introducing-polhem, /news/product-update-november-26, /news/product-update-september-26) and /webinar; no pricing, benchmark, workflow, practice-area or comparison pages.Source
• Your websiteSeptember product update (September 30, 2026): a project Overview that runs from assessment to references, evidence and claim chart, with team notes and questions; editable reports exported to PDF or Word.Source

Create one page per practice area. Each page covers the typical questions (e.g. SEP essentiality in telecoms, FTO before launch in medtech), the sources that matter (scientific publications for life science, product brochures and videos for mechanical), the published benchmark or customer quote that fits, and a sample analysis on a public patent in that field.

Heads of IP think in technology areas. A page that shows proof from a peer in their field converts better than a general claim, and gives outbound emails a specific link.

• Customers decline to be named

  Use the anonymised format of the investment-bank story.

• Your websiteHomepage proof and offer: H1 "AI agents for patent analysis", "Trusted by 100+ IP teams", logos including KUKA, Alfa Laval, Mannheimer Swartling, Bergenstråhle, Loza & Loza and Desmarais LLP; testimonials from KUKA, Spotify's Head of IP, Bergenstråhle and Mannheimer Swartling; six practice areas (telecommunications, life science and medtech, chemistry and materials, mechanical and industrial, robotics and automation, electronics and software); pricing only in the FAQ: a "per-project basis, paying only for the cases you bring through" or a flat-fee site licence, and "We do not charge per user"; the FAQ also says "Stilta is ISO 27001 certified, GDPR compliant, and EU AI Act aligned." There is no pricing page (/pricing returns 404) and no newsletter signup.Source
• Your websiteFunding post (May 19, 2026): $10.5M led by Andreessen Horowitz with Y Combinator; since launching in February 2026 Stilta "signed Roche, Alfa Laval, and Maersk" and "Three of the five largest IP firms globally are customers or in active pilots"; the money hires engineers, go-to-market staff and patent experts in Stockholm and New York; the CEO: "When one company starts using AI for patent enforcement, every competitor has to follow."Source
• Your websiteKUKA re-run (September 24, 2026): KUKA's IP team gave Stilta a completed infringement case with only the patent ID and public evidence; one autonomous run of about 20 minutes scanned 500+ references, reached the same no-infringement verdict and found the same decisive gaps, in 85–90% less time than the manual assessment.Source
• Your websiteSitemap (23 URLs, checked November 9, 2026): no lastmod dates; 2 product pages, 13 news posts and legal pages; missing the three newest posts linked from /news (/news/introducing-polhem, /news/product-update-november-26, /news/product-update-september-26) and /webinar; no pricing, benchmark, workflow, practice-area or comparison pages.Source

Publish /evaluate-patent-ai: a vendor-neutral checklist for IP teams choosing a tool (recall on your own closed matters, citation traceability, steerability, data residency, pricing model, exports), with the PTAB method as a do-it-yourself protocol. Then add short, factual "Stilta and X" pages for the rivals buyers raise most (&AI, Patlytics, Solve Intelligence, IPRally), each pointing back to the evaluation protocol and the blind re-run offer.

Buyers comparing tools are close to a decision, and a fair, testable method suits careful lawyers better than a feature grid. It also corrects the "unconfirmed" framing with evidence.

• Comparison pages invite counter-claims

  Stick to rivals' own published facts and Stilta's documented method.

• #1 Put every benchmark on one cited /benchmarks page, and make the homepage numbers match it

• Competitor research&AI's own "&AI vs Stilta" page calls Stilta "a newer entrant (founded 2026)", says its court usage should be confirmed, suggests it may fit US IPR and district court work less well because it is not US-headquartered, and offers a 7-day free trial of &AI.See the research Source
• Your websitePTAB study (February 17, 2026): 40 IPR institution decisions from February 2023 to January 2026; "71% petition recall across 40 cases, with a median search time of 17 minutes per patent", while the header shows "21m" and callouts of "+31pp" vs. search tools and "+49pp" vs. LLMs; the five tools and three LLMs compared are not named; Stilta also surfaced 161 references not cited in the petitions; IPR prior-art search typically costs $15,000–50,000 and takes 4–8 weeks; the authors write "We are publishing the full benchmark dataset - including IPR identifiers and challenged patents", but the page has no download link; CTA "Want to run this benchmark on your own patents?"Source
• Your websiteSitemap (23 URLs, checked November 9, 2026): no lastmod dates; 2 product pages, 13 news posts and legal pages; missing the three newest posts linked from /news (/news/introducing-polhem, /news/product-update-november-26, /news/product-update-september-26) and /webinar; no pricing, benchmark, workflow, practice-area or comparison pages.Source
• Competitor researchSolve Intelligence homepage: "AI for patents" across drafting, prosecution, charts and litigation; "700+" in-house IP teams and law firms; it is also bringing LexisNexis global patent litigation data into its workflows; SOC 2 Type II, ISO 27001 and ISO 42001.See the research Source

Publish a /europe hub. It covers how Stilta supports UPC revocation actions and EPO oppositions, testing evidence against claim constructions from European litigation history, a short quarterly "UPC in numbers" note drawn from the case data (with LexisNexis's permission), and a European benchmark: re-running Stilta on decided EPO oppositions, as was done with PTAB.

Being the European specialist is a position Stilta can win now, with local customers, data residency and its own data, while the US rivals concentrate on US IPR work.

• Licensing limits on the data

  Publish method and Stilta's own benchmark if aggregate figures aren't allowed.

• Your websiteLexisNexis integration (October 21, 2026): patent full text, citations, families and legal status, plus about 93,000 European patent cases from 13 countries with 200+ fields each, including the Unified Patent Court.Source
• Competitor research&AI's own "&AI vs Stilta" page calls Stilta "a newer entrant (founded 2026)", says its court usage should be confirmed, suggests it may fit US IPR and district court work less well because it is not US-headquartered, and offers a 7-day free trial of &AI.See the research Source
• Competitor researchSolve Intelligence homepage: "AI for patents" across drafting, prosecution, charts and litigation; "700+" in-house IP teams and law firms; it is also bringing LexisNexis global patent litigation data into its workflows; SOC 2 Type II, ISO 27001 and ISO 42001.See the research Source
• Your websiteSecurity page: "Our information security management system is aligned to ISO 27001."; a SOC 2 Type II audit and monthly external penetration tests; Azure hosting with EU customers in Sweden and US hosting for US customers; no training on customer data; SSO and audit log export; reports in the trust center.Source

Agree a one-page scorecard with each pilot before it starts: three closed matters to re-run blind, two live matters, the metrics (recall against the firm's own references, hours saved, partner rating of claim-chart quality), a named partner sponsor, and a decision date. Share a mid-pilot readout using the new Home view and editable reports, and end with a written results memo the champion can forward.

Pilots without agreed success criteria drift. A scorecard gives the internal champion the document they need to get a budget signed, in the evidence format their own lawyers trust.

• Firms resist formal criteria

  Present it as a tool for their own partner sign-off.

• #2 Make "bring a closed matter" the demo: a blind re-run of a case the buyer already knows the answer to

• Your websiteFunding post (May 19, 2026): $10.5M led by Andreessen Horowitz with Y Combinator; since launching in February 2026 Stilta "signed Roche, Alfa Laval, and Maersk" and "Three of the five largest IP firms globally are customers or in active pilots"; the money hires engineers, go-to-market staff and patent experts in Stockholm and New York; the CEO: "When one company starts using AI for patent enforcement, every competitor has to follow."Source
• HiringCareers page and open application (posted August 18, 2026): Stilta is "now hiring the founding team"; the open application covers Stockholm and New York City with visa sponsorship; no sales, marketing or patent-expert roles are posted yet.Source
• Your websiteNovember product update (November 4, 2026): a new interface with a Home view of items needing attention; search "up to 10–50x faster"; Analyst (beta) over prosecution and litigation history; Portfolio analysis (beta) with full release later in November; Boards (coming soon) to monitor "patents, assignees, and topics"; a live walkthrough at stilta.com/webinar.Source
• Your websiteSeptember product update (September 30, 2026): a project Overview that runs from assessment to references, evidence and claim chart, with team notes and questions; editable reports exported to PDF or Word.Source

Ship Boards with a default weekly digest: new filings from watched assignees, new products that may read on the customer's claims, and changes in legal status, each with a one-click "analyse this" into Deep Analysis. Seed every new account with a Board of its top three competitors during onboarding, so the first digest arrives in week one.

Recurring alerts give customers a reason to log in between matters, surface new paid projects (each alert can become an analysis) and justify site licences over per-project buying.

• Noisy alerts get ignored

  Rank by claim relevance and cap at five items a week.

• Your websiteNovember product update (November 4, 2026): a new interface with a Home view of items needing attention; search "up to 10–50x faster"; Analyst (beta) over prosecution and litigation history; Portfolio analysis (beta) with full release later in November; Boards (coming soon) to monitor "patents, assignees, and topics"; a live walkthrough at stilta.com/webinar.Source
• Your websiteDeep Analysis page: five workflows on one page (invalidity, infringement, freedom to operate, patentability, standard essentiality) reached by on-page anchors; a table contrasting "Days to weeks per search" with "Results in minutes" and "High cost per search" with "About a third of the cost"; no testimonials; a "Book a demo" button in every section.Source
• HiringPlatform Engineer (Stockholm, on-site, posted September 28, 2026): customers give Stilta "agents reasoning for hours across millions of pages"; the role scales the platform from hundreds to thousands of concurrent agent sessions on Azure, with isolation between customers, compliance and data residency.Source

Package Portfolio Analysis as an annual subscription priced per patent family, with tiers by portfolio size: monthly re-mapping to the customer's and competitors' products, strength ranking, and pruning recommendations, plus a quarterly review with a patent expert. Start with an "entry" portfolio health check (for example, 500 families) that converts into the subscription.

Per-family pricing matches how IP departments budget (renewal fees are paid per family), creates recurring revenue alongside per-project work, and builds on a cost advantage Stilta has measured.

• Underpricing locks in low revenue

  Price on the value of pruning savings and found opportunities, not compute cost.

• #5 Publish a "how pricing works" page next to the $15,000-50,000 search-firm benchmark

• Your websitePortfolio-scale model (October 15, 2026): Stilta post-trained an NVIDIA model on litigation claim charts and EPO examiner decisions; 5,000 claim charts per hour on one GPU, "8–13x faster than open reasoning models"; $0.65 per 1,000 comparisons versus $132 for a leading frontier model, "about 200x more cost-efficient"; a 10,000-family portfolio analysed every month for about $1,600 a year instead of $320,000; early access is limited to a small number of enterprises and firms.Source
• Your websitePortfolio Analysis page: "Your whole portfolio, read at once"; infringement, validity, taxonomy, product–patent mapping and SEP mapping for each patent; "Compare strength" and "Map the landscape" views; no testimonials or prices.Source
• Your websiteMaersk benchmark (September 10, 2026): from patent IDs only, Stilta mapped patents to Maersk products with 88% accuracy (143 of 163) and 97% (30 of 31) where public evidence taught every limitation, across 500+ patents, 150+ invention families and 2,000+ claim limitations; competitor products are the next stage.Source
• Your websiteHomepage proof and offer: H1 "AI agents for patent analysis", "Trusted by 100+ IP teams", logos including KUKA, Alfa Laval, Mannheimer Swartling, Bergenstråhle, Loza & Loza and Desmarais LLP; testimonials from KUKA, Spotify's Head of IP, Bergenstråhle and Mannheimer Swartling; six practice areas (telecommunications, life science and medtech, chemistry and materials, mechanical and industrial, robotics and automation, electronics and software); pricing only in the FAQ: a "per-project basis, paying only for the cases you bring through" or a flat-fee site licence, and "We do not charge per user"; the FAQ also says "Stilta is ISO 27001 certified, GDPR compliant, and EU AI Act aligned." There is no pricing page (/pricing returns 404) and no newsletter signup.Source

Create a co-delivery programme. Partner firms get a site licence, firm-branded report templates and a referral fee or discount when they bring an in-house client onto Stilta. Their clients get a direct seat to view and steer matters with the firm. Publish a /partners page listing the participating firms, and a one-page brief firms can send to clients.

Each partner firm brings several in-house clients, and their trust carries over to Stilta. It also turns the potential conflict (AI replacing outside counsel) into a shared offer.

• Channel conflict with direct in-house sales

  Register referred accounts to the firm for 12 months.

• Your websiteHomepage proof and offer: H1 "AI agents for patent analysis", "Trusted by 100+ IP teams", logos including KUKA, Alfa Laval, Mannheimer Swartling, Bergenstråhle, Loza & Loza and Desmarais LLP; testimonials from KUKA, Spotify's Head of IP, Bergenstråhle and Mannheimer Swartling; six practice areas (telecommunications, life science and medtech, chemistry and materials, mechanical and industrial, robotics and automation, electronics and software); pricing only in the FAQ: a "per-project basis, paying only for the cases you bring through" or a flat-fee site licence, and "We do not charge per user"; the FAQ also says "Stilta is ISO 27001 certified, GDPR compliant, and EU AI Act aligned." There is no pricing page (/pricing returns 404) and no newsletter signup.Source
• Your websiteSeptember product update (September 30, 2026): a project Overview that runs from assessment to references, evidence and claim chart, with team notes and questions; editable reports exported to PDF or Word.Source
• Your websiteContact page: a form (name, company, email, message) for partnerships, press, careers and general questions, with "Looking for a product walkthrough? Book a demo instead."; no office addresses and no stated response time.Source

Create a /us page that puts these together: the PTAB benchmark, US hosting and data residency, US firm customers (with quotes where allowed), US workflows (IPR petitions, district court invalidity contentions, FTO) and a New York contact once the first hire starts. Use it in all US outreach and conference follow-up.

It answers the most likely US objection on one page with evidence Stilta already has, before the first New York salesperson has to handle it on calls.

• A US page with no US team rings hollow

  Lead with US customers and the US-based benchmark, not with an office.

• Your websiteFunding post (May 19, 2026): $10.5M led by Andreessen Horowitz with Y Combinator; since launching in February 2026 Stilta "signed Roche, Alfa Laval, and Maersk" and "Three of the five largest IP firms globally are customers or in active pilots"; the money hires engineers, go-to-market staff and patent experts in Stockholm and New York; the CEO: "When one company starts using AI for patent enforcement, every competitor has to follow."Source
• HiringCareers page and open application (posted August 18, 2026): Stilta is "now hiring the founding team"; the open application covers Stockholm and New York City with visa sponsorship; no sales, marketing or patent-expert roles are posted yet.Source
• Competitor research&AI's own "&AI vs Stilta" page calls Stilta "a newer entrant (founded 2026)", says its court usage should be confirmed, suggests it may fit US IPR and district court work less well because it is not US-headquartered, and offers a 7-day free trial of &AI.See the research Source
• Your websiteSecurity page: "Our information security management system is aligned to ISO 27001."; a SOC 2 Type II audit and monthly external penetration tests; Azure hosting with EU customers in Sweden and US hosting for US customers; no training on customer data; SSO and audit log export; reports in the trust center.Source
• Your websiteContact page: a form (name, company, email, message) for partnerships, press, careers and general questions, with "Looking for a product walkthrough? Book a demo instead."; no office addresses and no stated response time.Source

Launch an annual "IPR recall benchmark". Use a fresh, held-out set of instituted IPR decisions each year, with a larger sample, run or audited by a neutral academic or practitioner group, and invite every vendor to submit runs. Publish recall, time and cost per tool, plus the dataset and method. Stilta enters like everyone else.

Buyers in this market want independent proof, and the vendor that creates the benchmark gets cited every time it is discussed. It also answers the "study by the tool's developers" limitation directly.

• Stilta doesn't win an edition

  The credibility of setting the standard is worth more; publish what the team learned.

• Rivals refuse to enter

  Run public tools anonymously as the original study did.

• #1 Put every benchmark on one cited /benchmarks page, and make the homepage numbers match it

• Your websitePTAB study (February 17, 2026): 40 IPR institution decisions from February 2023 to January 2026; "71% petition recall across 40 cases, with a median search time of 17 minutes per patent", while the header shows "21m" and callouts of "+31pp" vs. search tools and "+49pp" vs. LLMs; the five tools and three LLMs compared are not named; Stilta also surfaced 161 references not cited in the petitions; IPR prior-art search typically costs $15,000–50,000 and takes 4–8 weeks; the authors write "We are publishing the full benchmark dataset - including IPR identifiers and challenged patents", but the page has no download link; CTA "Want to run this benchmark on your own patents?"Source
• Your websitePolhem 1.0 (November 5, 2026): Stilta's own 345M-parameter search model over more than 170 million patent documents, trained on examiner citations; on 3,969 held-out applications, 72.6% of examiner-cited art is in the top 1,000 results, 11.5 points above the strongest open model tested (over 20 times as many parameters); first on the public DAPFAM and perf200 benchmarks.Source
• Competitor researchPatlytics homepage: "The Premier AI-Powered Patent Platform"; 160M+ patents and 300M+ literature items; a $40M Series B; ROI case studies such as an Am Law 100 firm finishing a 120-patent project in 20 hours instead of 100 for about $38,000 in added margin; SOC 2 Type 2, ISO 27001 and ISO 42001; no published benchmarks on the homepage.See the research Source

Expose Polhem retrieval (and later single-claim charting) as a metered API and an MCP server, with the same tenant isolation and EU/US residency as the app. Price per thousand queries and include it in site licences. Target firms' innovation teams and in-house legal-tech teams, and use API usage to spot accounts ready for the full Deep Analysis product.

It reaches buyers who won't adopt another app, turns Stilta's search advantage into revenue that doesn't depend on seats or projects, and builds a habit inside the tools lawyers already use.

• The API cannibalises app revenue

  Expose retrieval only; keep agentic analysis and claim charts in the app.

• Engineering focus is split

  Start only after a paid commitment from a beta firm.

• Your websitePolhem 1.0 (November 5, 2026): Stilta's own 345M-parameter search model over more than 170 million patent documents, trained on examiner citations; on 3,969 held-out applications, 72.6% of examiner-cited art is in the top 1,000 results, 11.5 points above the strongest open model tested (over 20 times as many parameters); first on the public DAPFAM and perf200 benchmarks.Source
• Competitor researchIPRally homepage: "The patent engine for humans and agents", with its engine available through MCP and API; 5,000 professionals and 6 patent offices; case studies such as RPX finding 40% more cited references; a linked pricing page; demo replies promised within 2 business days.See the research Source
• HiringAI Engineer (Stockholm, on-site, posted April 21, 2026): owns the multi-agent architecture for invalidity search, infringement analysis and freedom-to-operate research, retrieval over hundreds of millions of patents and papers, and the evaluation systems that show the output is reliable; owns features from customer conversations through production use.Source
• HiringPlatform Engineer (Stockholm, on-site, posted September 28, 2026): customers give Stilta "agents reasoning for hours across millions of pages"; the role scales the platform from hundreds to thousands of concurrent agent sessions on Azure, with isolation between customers, compliance and data residency.Source

Offer in-house teams a programme: Stilta maps the entire portfolio against competitor products using public evidence, ranks the strongest infringement reads with claim charts, and hands them to the company's chosen counsel. Price it as a modest platform fee plus a success fee on licensing revenue or settlements it leads to, structured with the client's counsel and within professional rules in each jurisdiction.

It turns a cost centre's dormant assets into revenue, aligns Stilta's price with the outcome, and puts the CEO's enforcement thesis into a product customers can buy.

• Success fees raise regulatory or ethics issues

  Take legal advice per jurisdiction and offer a flat-fee alternative.

• Public evidence is thin for some products

  Report evidence strength per read, as in the Maersk full-evidence subset.

• #15 Sell Portfolio Analysis as an annual subscription priced per patent family

• PressTechCrunch (May 19, 2026): the CEO describes the output as "litigation-grade: a report and claim charts with pinpoint citations to every piece of evidence"; the aim is to lower the cost that leaves many companies' patents unenforced, unlicensed and unanalyzed.Source
• Your websiteFunding post (May 19, 2026): $10.5M led by Andreessen Horowitz with Y Combinator; since launching in February 2026 Stilta "signed Roche, Alfa Laval, and Maersk" and "Three of the five largest IP firms globally are customers or in active pilots"; the money hires engineers, go-to-market staff and patent experts in Stockholm and New York; the CEO: "When one company starts using AI for patent enforcement, every competitor has to follow."Source
• Your websiteMaersk benchmark (September 10, 2026): from patent IDs only, Stilta mapped patents to Maersk products with 88% accuracy (143 of 163) and 97% (30 of 31) where public evidence taught every limitation, across 500+ patents, 150+ invention families and 2,000+ claim limitations; competitor products are the next stage.Source
• Your websitePortfolio-scale model (October 15, 2026): Stilta post-trained an NVIDIA model on litigation claim charts and EPO examiner decisions; 5,000 claim charts per hour on one GPU, "8–13x faster than open reasoning models"; $0.65 per 1,000 comparisons versus $132 for a leading frontier model, "about 200x more cost-efficient"; a 10,000-family portfolio analysed every month for about $1,600 a year instead of $320,000; early access is limited to a small number of enterprises and firms.Source

Package a "portfolio diligence" offer: a fixed-price report on a target portfolio (infringement evidence per family, validity risk, product mapping) delivered within days. Sell it to M&A advisers, IP brokers and litigation funders, with an option to re-run it after the deal for the buyer. Write the investment-bank case up as the flagship example.

Deal teams have budgets and deadlines and reward speed, and each report opens a door to the portfolio's new owner as a long-term customer.

• Reports relied on as valuations

  Present them as evidence for counsel, not opinions on value.

• Your websiteCustomer story (March 18, 2026): a Scandinavian investment banking group used Stilta for infringement and validity analysis of a music and AI patent portfolio in the US and Europe ahead of a potential securitization and sale; about 30 minutes per patent versus about 20 hours manually.Source
• Your websitePortfolio Analysis page: "Your whole portfolio, read at once"; infringement, validity, taxonomy, product–patent mapping and SEP mapping for each patent; "Compare strength" and "Map the landscape" views; no testimonials or prices.Source

Pick ten well-known expired SEPs from older cellular or video standards and publish Stilta's claim-to-specification-clause mapping for each as an open sample, with the cited clause text. Invite SEP practitioners to dispute a mapping, and use the corrections to improve the product. Bring printed samples to SEP events.

Expired patents carry no live dispute risk, the outputs show real precision to a specialised audience that rarely sees public examples, and SEP licensing is a high-value niche where a single client can be worth a great deal.

• A public mapping is wrong

  Show the cited clauses and invite corrections openly.

• #9 Give each of the five Deep Analysis workflows its own page with a sample claim chart

• Your websiteDeep Analysis page: five workflows on one page (invalidity, infringement, freedom to operate, patentability, standard essentiality) reached by on-page anchors; a table contrasting "Days to weeks per search" with "Results in minutes" and "High cost per search" with "About a third of the cost"; no testimonials; a "Book a demo" button in every section.Source
• EventsIAM Live: SEP Summit Global 2026 lists Stilta as a sponsor, describing source-cited invalidity, infringement and freedom-to-operate assessments.Source
• Your websitePortfolio Analysis page: "Your whole portfolio, read at once"; infringement, validity, taxonomy, product–patent mapping and SEP mapping for each patent; "Compare strength" and "Map the landscape" views; no testimonials or prices.Source
• Your websiteHomepage proof and offer: H1 "AI agents for patent analysis", "Trusted by 100+ IP teams", logos including KUKA, Alfa Laval, Mannheimer Swartling, Bergenstråhle, Loza & Loza and Desmarais LLP; testimonials from KUKA, Spotify's Head of IP, Bergenstråhle and Mannheimer Swartling; six practice areas (telecommunications, life science and medtech, chemistry and materials, mechanical and industrial, robotics and automation, electronics and software); pricing only in the FAQ: a "per-project basis, paying only for the cases you bring through" or a flat-fee site licence, and "We do not charge per user"; the FAQ also says "Stilta is ISO 27001 certified, GDPR compliant, and EU AI Act aligned." There is no pricing page (/pricing returns 404) and no newsletter signup.Source