Leeroo growth playbook

Pre-render each blog post to static HTML at build time and serve it from the existing S3 and CloudFront setup, with its own title, meta description, Open Graph image, canonical tag, publish date and author role. Add all four posts to the sitemap with lastmod dates. Make unknown paths return a real 404 page with links to the product, the blog and the audit, instead of the homepage. Then give each post a closing call to action for the free audit, and a short note on how the older posts (Orchestrator, multi-model systems) led to Uniforge, so readers arriving from old links reach today's product.

For an enterprise sale, essays are what champions forward to colleagues. A share that previews as a generic homepage, or a link an assistant can't read, loses that forward. Pre-rendering is a small change on a static site, and it lets every post rank and be quoted on its own topic.

[ ] Plain fetch of each post returns the article title and first paragraph
[ ] Unique <title> and meta description per post
[ ] Open Graph title, description and image per post
[ ] Self-canonical tag
[ ] Publish date and author role visible
[ ] Post URLs in sitemap.xml with lastmod
[ ] Unknown paths return 404, not the homepage
[ ] Audit call to action at the end of each post

• Breaking existing client-side routes

  Pre-render only /blogs/* first and keep the app shell for everything else.

• Old posts describe an earlier product

  Add a dated editor's note at the top rather than deleting them.

• Your websiteBlog: /blogs lists four posts: "Organizational Superintelligence" (2025.11.10), "End2end Multi-Agent finetuning: Open Source Meets SOTA" (2024.12.19), "The First-Principle Multi-Model System" (2024.04.02) and "Leeroo Orchestrator V1: Towards an AI Operating System" (2024.01.24). Fetched twice on November 9, 2026 without running scripts, each post URL returned the homepage title "Leeroo | A Living Model of Your Business, Built by AI" and homepage text instead of the article; made-up paths such as /privacy return the same homepage.Source
• Your websiteSitemap: five URLs (/, /product, /company, /contact-us, /blogs), none with a lastmod date and no blog post URLs.Source
• Hacker News"Organizational Superintelligence" was submitted with the leeroo.com blog post URL and has 1 point.Source

Build a 4-6 page sample audit report from the $1B enterprise engagement, anonymised and approved by the customer: the systems connected, the conflicts and duplicates found by system pair, the top five dollar-quantified findings, and the ranked ML opportunities. Put a preview of it above the form on /contact-us, and offer the full PDF as a download with no email gate.

Enterprise champions have to explain a vendor to their security team and their boss before they ask for anything. A concrete sample makes the free audit a known quantity and sets expectations that the real audit then meets.

• The customer declines

  Build the sample from a synthetic dataset and label it as such.

• Buyers expect the same dollar figures

  Show ranges and explain what drives them.

• Your websiteFree audit page: "See what your data is really telling you."; "FREE AUDIT · 48 HOURS"; connects "2–3 of your systems" with "No setup on your side" and "Runs entirely in your cloud"; finds duplicates, conflicts and missing links plus "ML opportunities in your data, quantified by value"; the form asks for name, work email, company, company size, "Which systems should we look at?" and "Anything else?"; the team will "reply within one business day"; no sample report, testimonial or customer name.Source
• Your websiteHomepage (checked November 9, 2026): H1 "A living model of your business. Built by AI. Run in your cloud."; Uniforge "connects to every system you operate" and ships agents; CTAs "Request a free audit", "Read the specification" and "Watch the demo"; proof from an unnamed "$1B enterprise" with 35+ systems: 9×10⁹ rows learned, a 62% cut in cross-system conflicts and about $20M of trapped value surfaced; runs in the customer's VPC on AWS, Azure or GCP, or fully on-prem and air-gapped; logos (Meta, Microsoft, Amazon, Hugging Face and others) shown as companies whose engineers use the team's open-source work; London, backed by Y Combinator.Source

Publish a one-page "What the audit needs from you" guide, linked from the form. It covers what gets deployed (and where), the read-only roles or service accounts per system for Salesforce, Oracle, ServiceNow, Zendesk, Workday and warehouses, network egress (none), data retention and deletion after 48 hours, who at Leeroo can see what, and a pre-filled answer sheet for common security questionnaire questions. Add a checkbox on the form: "Send this guide to my security team".

Removing the unknowns early shortens the gap between request and access, which is where free enterprise pilots usually stall. It also turns the security team into a reviewer of a known document, not a blocker of an unknown vendor.

• Publishing too much architecture detail

  Describe scopes and data flow, not internal endpoints or keys.

• Your websiteFree audit page: "See what your data is really telling you."; "FREE AUDIT · 48 HOURS"; connects "2–3 of your systems" with "No setup on your side" and "Runs entirely in your cloud"; finds duplicates, conflicts and missing links plus "ML opportunities in your data, quantified by value"; the form asks for name, work email, company, company size, "Which systems should we look at?" and "Anything else?"; the team will "reply within one business day"; no sample report, testimonial or customer name.Source
• Your websiteProduct page: "Learns your business. Acts on it."; six learning agents (Entity Resolution, Lineage Mapping, Process Mining, Event Learning, Skill Learning, LoRA Training) and six action agents (Billing-integrity, Data-quality, Churn sentinel, Cross-sell, Contract-leakage, Fraud sentinel); integration logos Oracle, ServiceNow, SAP, Salesforce, Workday and "G Cloud", while the diagram shows Salesforce, Oracle Billing, ServiceNow, Zendesk, Workday and a warehouse ("6 SYSTEMS · 113 TABLES · READ-ONLY"); worked examples such as "National Riverbend Services" and "EST. $5.0M/YR" not labelled as illustrative; "Days, not quarters"; "Approvals where you want them. Full audit log."; no pricing, named case study or security page.Source

Create /security covering the deployment modes (VPC on AWS, Azure or GCP, or on-prem and air-gapped), the data boundary, read-only connectors, approvals and audit log, model hosting in sovereign mode, sub-processors, incident contact and the current state of any certifications (with honest timelines if they are in progress). Publish a privacy policy and link both pages from the footer and the audit form.

Procurement and security teams check these pages before they reply to a vendor. Publishing them signals maturity beyond a team of three and removes a common reason for silence after an audit request.

• Claims that outrun reality

  Only state controls that exist today; mark the rest as planned.

• Your websiteSite scan (homepage and /contact-us): Google Analytics, PostHog and Google Tag Manager installed; no live chat; no online booking; no pricing page; no privacy or cookie policy linked; no trust center.Source
• Your websiteBlog: /blogs lists four posts: "Organizational Superintelligence" (2025.11.10), "End2end Multi-Agent finetuning: Open Source Meets SOTA" (2024.12.19), "The First-Principle Multi-Model System" (2024.04.02) and "Leeroo Orchestrator V1: Towards an AI Operating System" (2024.01.24). Fetched twice on November 9, 2026 without running scripts, each post URL returned the homepage title "Leeroo | A Living Model of Your Business, Built by AI" and homepage text instead of the article; made-up paths such as /privacy return the same homepage.Source
• Your websiteProduct page: "Learns your business. Acts on it."; six learning agents (Entity Resolution, Lineage Mapping, Process Mining, Event Learning, Skill Learning, LoRA Training) and six action agents (Billing-integrity, Data-quality, Churn sentinel, Cross-sell, Contract-leakage, Fraud sentinel); integration logos Oracle, ServiceNow, SAP, Salesforce, Workday and "G Cloud", while the diagram shows Salesforce, Oracle Billing, ServiceNow, Zendesk, Workday and a warehouse ("6 SYSTEMS · 113 TABLES · READ-ONLY"); worked examples such as "National Riverbend Services" and "EST. $5.0M/YR" not labelled as illustrative; "Days, not quarters"; "Approvals where you want them. Full audit log."; no pricing, named case study or security page.Source

On form submit, show a calendar with 30-minute scoping slots with the founders or the Founding Engineer, pre-filled with the systems the requester named. Send a confirmation with the sample report (idea 2) and the access guide (idea 3). Anyone who doesn't book gets the usual email reply.

Booking at peak intent removes a round of email tag, and the 48-hour promise starts from a fixed date the buyer chose.

• Unqualified bookings fill the calendar

  Show slots only for company sizes in the target range.

• #2 Show buyers what the free 48-hour audit delivers: publish an anonymised sample report from the $1B enterprise
• #3 Explain how "no setup on your side" works when the audit "runs entirely in your cloud": a one-page access guide for the buyer's security team

• Your websiteFree audit page: "See what your data is really telling you."; "FREE AUDIT · 48 HOURS"; connects "2–3 of your systems" with "No setup on your side" and "Runs entirely in your cloud"; finds duplicates, conflicts and missing links plus "ML opportunities in your data, quantified by value"; the form asks for name, work email, company, company size, "Which systems should we look at?" and "Anything else?"; the team will "reply within one business day"; no sample report, testimonial or customer name.Source
• Your websiteSite scan (homepage and /contact-us): Google Analytics, PostHog and Google Tag Manager installed; no live chat; no online booking; no pricing page; no privacy or cookie policy linked; no trust center.Source

Add an "Illustrative example" tag to the demo figures. Give the $1B enterprise a short descriptor (industry, region, years of history, systems involved) and the time it took to reach the 62% and $20M results. Keep the open-source logo wall, but put its accurate caption (engineers at these companies use our open-source work) right next to it.

Clear labels make the real numbers more believable, and an industry descriptor lets buyers in that sector see themselves in the result.

• The customer won't allow an industry descriptor

  Use a broader sector and region.

• Your websiteProduct page: "Learns your business. Acts on it."; six learning agents (Entity Resolution, Lineage Mapping, Process Mining, Event Learning, Skill Learning, LoRA Training) and six action agents (Billing-integrity, Data-quality, Churn sentinel, Cross-sell, Contract-leakage, Fraud sentinel); integration logos Oracle, ServiceNow, SAP, Salesforce, Workday and "G Cloud", while the diagram shows Salesforce, Oracle Billing, ServiceNow, Zendesk, Workday and a warehouse ("6 SYSTEMS · 113 TABLES · READ-ONLY"); worked examples such as "National Riverbend Services" and "EST. $5.0M/YR" not labelled as illustrative; "Days, not quarters"; "Approvals where you want them. Full audit log."; no pricing, named case study or security page.Source
• Your websiteHomepage (checked November 9, 2026): H1 "A living model of your business. Built by AI. Run in your cloud."; Uniforge "connects to every system you operate" and ships agents; CTAs "Request a free audit", "Read the specification" and "Watch the demo"; proof from an unnamed "$1B enterprise" with 35+ systems: 9×10⁹ rows learned, a 62% cut in cross-system conflicts and about $20M of trapped value surfaced; runs in the customer's VPC on AWS, Azure or GCP, or fully on-prem and air-gapped; logos (Meta, Microsoft, Amazon, Hugging Face and others) shown as companies whose engineers use the team's open-source work; London, backed by Y Combinator.Source

Rewrite the YC profile description around Uniforge, the free audit and the $1B enterprise result. Update the Founding Engineer post to name Uniforge, the six source systems, the audit and the action agents, and the intern post to tie the Learning Engine research to Uniforge's learning agents. Add the roles to the careers section on /company, which currently lists none.

YC's directory and job board are where investors, candidates and some buyers first meet a YC company. One consistent story makes every introduction cheaper.

• Older pilots know the earlier pitch

  Explain in the launch post how trainable agents became Uniforge.

• Y CombinatorYC profile: Spring 2025 batch, team size 3, London; describes Leeroo as building trainable AI agents for data and AI teams that run on-prem or in a customer's VPC, with target pilot partners in banking, finance and healthcare.Source
• HiringFounding Engineer (London or remote, $10K–$15K a month, listed on Y Combinator): "Be a customer-first problem solver", working with customers from discovery through delivery and turning requests into scoped plans with clear ROI; joins pre-sales and discovery calls; connects Leeroo's agents to customer systems and feeds reusable patterns back to research and product; nice to have: privacy, security and model governance such as EU AI Act readiness. The post does not mention Uniforge or Kapso.Source
• Your websiteCompany page: Leeroo builds Uniforge, "the platform that learns your business and acts on it"; a careers section, "Help us build AI that builds AI", with no listed roles and applications by email; open-source projects Kapso, SuperML, Leeroopedia and Leeroopedia MCP; "1.5M+ Hugging Face downloads" across the team's open-source models.Source

Change the footer to "Powered by Leeroo, makers of Uniforge" with a link to the audit. On pages about entity resolution, deduplication, data quality, lineage and tabular ML, add a small contextual box: "See how many duplicates and conflicts your own systems hold: free 48-hour audit, in your cloud." Tag the links so audit requests from Leeroopedia can be counted.

Practitioners who read the wiki are the people who feel cross-system data pain every day and can bring Uniforge to their team. A relevant, quiet box converts better than a generic banner.

• The community sees it as an ad

  Keep boxes small, relevant and only on matching pages.

• Open sourceLeeroopedia: 27,667 pages of ML and data engineering knowledge (1,118 workflows, 7,697 principles, 16,832 implementations, 1,245 heuristics, 909 environments); footer "Powered by Leeroo — building Organizational Superintelligence"; links to SuperML, the Leeroopedia MCP and Kapso; no mention of Uniforge or the free audit.Source
• Your websiteHomepage (checked November 9, 2026): H1 "A living model of your business. Built by AI. Run in your cloud."; Uniforge "connects to every system you operate" and ships agents; CTAs "Request a free audit", "Read the specification" and "Watch the demo"; proof from an unnamed "$1B enterprise" with 35+ systems: 9×10⁹ rows learned, a 62% cut in cross-system conflicts and about $20M of trapped value surfaced; runs in the customer's VPC on AWS, Azure or GCP, or fully on-prem and air-gapped; logos (Meta, Microsoft, Amazon, Hugging Face and others) shown as companies whose engineers use the team's open-source work; London, backed by Y Combinator.Source

Create /kapso-enterprise: what the enterprise edition adds (private lesson banks, deployment in the customer's cloud, support), how Kapso powers Uniforge's ML generation agent, the benchmark results, and two paths: "Talk about Kapso at work" and "Run a free audit on your data". Point the README and docs links there.

It turns open-source interest into a named enterprise conversation instead of a dead end, and shows engineers how the tool they like maps to the platform their company would buy.

• Splits focus between two products

  Position Kapso as the engine inside Uniforge, not a separate line.

• #1 Make every blog post its own readable page: today each post URL, including the one shared on Hacker News, shows crawlers and link previews the homepage

• Open sourceKapso on GitHub: MIT licence, 139 stars, 23 forks; the README reports RelBench classification at 81.2 AUROC against 79.3 for KumoRFM-v2 and recommendation at 18.4 MAP against 9.3, plus an IOAI 2026 score of 536.07 and the Grand Master Trophy; it offers "Kapso for Enterprise" through a "talk to us" link to leeroo.com; Uniforge is not mentioned.Source
• Your websiteSitemap: five URLs (/, /product, /company, /contact-us, /blogs), none with a lastmod date and no blog post URLs.Source

Publish a server-rendered /benchmarks/relbench page on leeroo.com: the tasks, the scores against named baselines, the method, and how to reproduce them, with a plain-English section on what relational prediction means for churn, cross-sell and fraud (Uniforge's action agents). End with "Run this on your own tables" leading to the audit.

Benchmark pages attract technical evaluators searching by name, and a reproducible method earns trust. Tying the scores to churn and fraud turns a research result into a business case.

• Comparisons with a named rival get disputed

  Cite the public leaderboard and versions, and date the page.

• #1 Make every blog post its own readable page: today each post URL, including the one shared on Hacker News, shows crawlers and link previews the homepage

• Open sourceKapso on GitHub: MIT licence, 139 stars, 23 forks; the README reports RelBench classification at 81.2 AUROC against 79.3 for KumoRFM-v2 and recommendation at 18.4 MAP against 9.3, plus an IOAI 2026 score of 536.07 and the Grand Master Trophy; it offers "Kapso for Enterprise" through a "talk to us" link to leeroo.com; Uniforge is not mentioned.Source
• Competitor researchkumo.ai now redirects (301) to an NVIDIA documentation page for Kumo Relational, "the relational foundation model for predictive AI on relational data".See the research Source
• Open sourceDocs (docs.leeroo.com): covers Kapso only: installation, first build, harnesses, deploy targets, CLI reference and benchmark pages for IOAI 2026, RelBench, MLE-Bench and ALE-Bench; no Uniforge documentation.Source

Add a Uniforge section to the existing Mintlify docs: architecture overview, deployment Modes A (VPC) and B (sovereign), the connector reference per system with required scopes, the living-model objects (entity graph, lineage graph, event memory), each action agent's inputs, outputs and approval settings, and the audit log format. Link it from "Read the specification".

Architects often decide on a shortlist from documentation before talking to sales. Clear docs also make the Founding Engineer's discovery calls shorter.

• Docs lag the product

  Version pages and update with each release.

• Open sourceDocs (docs.leeroo.com): covers Kapso only: installation, first build, harnesses, deploy targets, CLI reference and benchmark pages for IOAI 2026, RelBench, MLE-Bench and ALE-Bench; no Uniforge documentation.Source
• Your websiteHomepage (checked November 9, 2026): H1 "A living model of your business. Built by AI. Run in your cloud."; Uniforge "connects to every system you operate" and ships agents; CTAs "Request a free audit", "Read the specification" and "Watch the demo"; proof from an unnamed "$1B enterprise" with 35+ systems: 9×10⁹ rows learned, a 62% cut in cross-system conflicts and about $20M of trapped value surfaced; runs in the customer's VPC on AWS, Azure or GCP, or fully on-prem and air-gapped; logos (Meta, Microsoft, Amazon, Hugging Face and others) shown as companies whose engineers use the team's open-source work; London, backed by Y Combinator.Source
• Your websiteProduct page: "Learns your business. Acts on it."; six learning agents (Entity Resolution, Lineage Mapping, Process Mining, Event Learning, Skill Learning, LoRA Training) and six action agents (Billing-integrity, Data-quality, Churn sentinel, Cross-sell, Contract-leakage, Fraud sentinel); integration logos Oracle, ServiceNow, SAP, Salesforce, Workday and "G Cloud", while the diagram shows Salesforce, Oracle Billing, ServiceNow, Zendesk, Workday and a warehouse ("6 SYSTEMS · 113 TABLES · READ-ONLY"); worked examples such as "National Riverbend Services" and "EST. $5.0M/YR" not labelled as illustrative; "Days, not quarters"; "Approvals where you want them. Full audit log."; no pricing, named case study or security page.Source

Publish one canonical /integrations list stating which systems are supported today, read-only, and which are on request. Then write five pages on the most valuable system pairs (Salesforce and Oracle Billing, Salesforce and SAP, ServiceNow and Workday, Zendesk and CRM, warehouse and CRM): the typical conflicts, what the agents find, the dollar impact, and a call to action to audit that exact pair.

Pair-specific pages match how operations and finance teams describe their pain ("our CRM and billing don't agree") and qualify audits by naming the 2-3 systems the form asks for.

• Trademark use of vendor logos

  Follow each vendor's naming and logo rules.

• #1 Make every blog post its own readable page: today each post URL, including the one shared on Hacker News, shows crawlers and link previews the homepage

• Your websiteProduct page: "Learns your business. Acts on it."; six learning agents (Entity Resolution, Lineage Mapping, Process Mining, Event Learning, Skill Learning, LoRA Training) and six action agents (Billing-integrity, Data-quality, Churn sentinel, Cross-sell, Contract-leakage, Fraud sentinel); integration logos Oracle, ServiceNow, SAP, Salesforce, Workday and "G Cloud", while the diagram shows Salesforce, Oracle Billing, ServiceNow, Zendesk, Workday and a warehouse ("6 SYSTEMS · 113 TABLES · READ-ONLY"); worked examples such as "National Riverbend Services" and "EST. $5.0M/YR" not labelled as illustrative; "Days, not quarters"; "Approvals where you want them. Full audit log."; no pricing, named case study or security page.Source
• Your websiteFree audit page: "See what your data is really telling you."; "FREE AUDIT · 48 HOURS"; connects "2–3 of your systems" with "No setup on your side" and "Runs entirely in your cloud"; finds duplicates, conflicts and missing links plus "ML opportunities in your data, quantified by value"; the form asks for name, work email, company, company size, "Which systems should we look at?" and "Anything else?"; the team will "reply within one business day"; no sample report, testimonial or customer name.Source

Package a 30-day paid pilot with a published starting price: one action agent (Billing-integrity or Contract-leakage first, because they recover cash) on the system pair from the audit, a target agreed from the audit's dollar findings, weekly reviews, and a credit towards an annual licence if the target is met. Describe it on a /pilot page and in every audit report.

A small, priced, outcome-tied step fits within budgets a director can approve. Starting with cash-recovery agents makes the pilot pay for itself in the buyer's eyes.

• Price anchors the annual deal too low

  Show the pilot as a fraction of the annual scope.

• #2 Show buyers what the free 48-hour audit delivers: publish an anonymised sample report from the $1B enterprise

• Your websiteSite scan (homepage and /contact-us): Google Analytics, PostHog and Google Tag Manager installed; no live chat; no online booking; no pricing page; no privacy or cookie policy linked; no trust center.Source
• HiringFounding Engineer (London or remote, $10K–$15K a month, listed on Y Combinator): "Be a customer-first problem solver", working with customers from discovery through delivery and turning requests into scoped plans with clear ROI; joins pre-sales and discovery calls; connects Leeroo's agents to customer systems and feeds reusable patterns back to research and product; nice to have: privacy, security and model governance such as EU AI Act readiness. The post does not mention Uniforge or Kapso.Source
• Your websiteProduct page: "Learns your business. Acts on it."; six learning agents (Entity Resolution, Lineage Mapping, Process Mining, Event Learning, Skill Learning, LoRA Training) and six action agents (Billing-integrity, Data-quality, Churn sentinel, Cross-sell, Contract-leakage, Fraud sentinel); integration logos Oracle, ServiceNow, SAP, Salesforce, Workday and "G Cloud", while the diagram shows Salesforce, Oracle Billing, ServiceNow, Zendesk, Workday and a warehouse ("6 SYSTEMS · 113 TABLES · READ-ONLY"); worked examples such as "National Riverbend Services" and "EST. $5.0M/YR" not labelled as illustrative; "Days, not quarters"; "Approvals where you want them. Full audit log."; no pricing, named case study or security page.Source

Write three sector pages. Banking: reconciling customer records across core banking, CRM and billing, plus the fraud sentinel. Financial services: contract leakage and billing integrity. Healthcare organisations: record unification across operational systems. Each page leads with sovereign Mode B and the audit log, names the usual systems in that sector, and links to a sector version of the sample audit. Use them in targeted outreach to data and operations leaders.

Regulated buyers look for proof that a vendor understands their constraints, and sovereignty is Leeroo's sharpest point against cloud-first rivals.

• Regulatory claims without review

  Describe controls, not compliance, unless counsel confirms it.

• #2 Show buyers what the free 48-hour audit delivers: publish an anonymised sample report from the $1B enterprise

• Y CombinatorLaunch post "Leeroo (Trainable AI Agents)": "trainable AI agents that learn much like human colleagues"; asks for pilot partners in banking, finance and healthcare that need data sovereignty, and for introductions to innovation leaders working on data and AI problems.Source
• Your websiteHomepage (checked November 9, 2026): H1 "A living model of your business. Built by AI. Run in your cloud."; Uniforge "connects to every system you operate" and ships agents; CTAs "Request a free audit", "Read the specification" and "Watch the demo"; proof from an unnamed "$1B enterprise" with 35+ systems: 9×10⁹ rows learned, a 62% cut in cross-system conflicts and about $20M of trapped value surfaced; runs in the customer's VPC on AWS, Azure or GCP, or fully on-prem and air-gapped; logos (Meta, Microsoft, Amazon, Hugging Face and others) shown as companies whose engineers use the team's open-source work; London, backed by Y Combinator.Source
• Your websiteSitemap: five URLs (/, /product, /company, /contact-us, /blogs), none with a lastmod date and no blog post URLs.Source

Write an anonymised (or named, if the customer agrees) case study page: the starting situation, the systems connected, the time to a working model, what the learning agents found, which action agents ran, how approvals were set, and how much of the $20M has been recovered so far. Include one quote from an operations or finance leader. Produce a one-page PDF version for outreach.

A detailed story answers the questions a one-line claim raises (how long, how much effort, how much recovered) and gives champions something to forward internally.

• Customer approval takes months

  Start anonymised and upgrade to named later.

• Your websiteHomepage (checked November 9, 2026): H1 "A living model of your business. Built by AI. Run in your cloud."; Uniforge "connects to every system you operate" and ships agents; CTAs "Request a free audit", "Read the specification" and "Watch the demo"; proof from an unnamed "$1B enterprise" with 35+ systems: 9×10⁹ rows learned, a 62% cut in cross-system conflicts and about $20M of trapped value surfaced; runs in the customer's VPC on AWS, Azure or GCP, or fully on-prem and air-gapped; logos (Meta, Microsoft, Amazon, Hugging Face and others) shown as companies whose engineers use the team's open-source work; London, backed by Y Combinator.Source
• Competitor researchTamr: "The Agentic Solution for Trusted AI-Ready Data", an AI-native master data management platform with entity resolution, an enterprise knowledge graph and MCP connectivity; "Request Demo" and customer stories.See the research Source

Add an optional question to Leeroopedia API key signup and the Discord welcome flow: "Using this at work? Tell us your company and we'll send you the enterprise notes." For those who opt in, send a short technical note on how Kapso powers Uniforge, and offer the free audit to their data team. Review opted-in companies monthly against the target sectors.

Bottom-up developer adoption is a cheap way into large accounts when the team is three people. An opt-in keeps trust with the community.

• The community feels mined

  Strictly opt-in, clear purpose, easy to ignore.

• #9 Build the "Kapso for Enterprise" page the GitHub README already sends people to

• Open sourceKapso on GitHub: MIT licence, 139 stars, 23 forks; the README reports RelBench classification at 81.2 AUROC against 79.3 for KumoRFM-v2 and recommendation at 18.4 MAP against 9.3, plus an IOAI 2026 score of 536.07 and the Grand Master Trophy; it offers "Kapso for Enterprise" through a "talk to us" link to leeroo.com; Uniforge is not mentioned.Source
• Open sourceSuperML on GitHub: an ML engineering plugin for coding agents (Apache-2.0, 197 stars, 18 forks) that gives agents access to Leeroopedia; reports a with-and-without comparison on 37 ML tasks.Source
• Open sourceLeeroopedia: 27,667 pages of ML and data engineering knowledge (1,118 workflows, 7,697 principles, 16,832 implementations, 1,245 heuristics, 909 environments); footer "Powered by Leeroo — building Organizational Superintelligence"; links to SuperML, the Leeroopedia MCP and Kapso; no mention of Uniforge or the free audit.Source

Find the heavy assets (likely the demo video or animation), serve a poster image and load the video only on click or when it scrolls into view, with a mobile-sized version. Set long cache lifetimes on hashed static files. Inline the critical CSS and fonts for the hero, and defer non-essential scripts so the headline paints first.

A lighter page loads reliably on mobile networks and uses less of the visitor's data allowance, and the headline (the LCP element) appears sooner, which matters most on first visits from social links.

• The demo gets fewer plays

  Use a clear play button and track plays.

• PageSpeedGoogle PageSpeed (November 2026): mobile performance 78, desktop 95; lab mobile LCP 3.4 s; the homepage downloads 20.3 MB in 29 requests, with about 8 MB savable through longer cache lifetimes, 1.8 s of render-blocking requests and 316 KB of unused JavaScript; Google Tag Manager (192 KB) and a jsDelivr-hosted script (183 KB) are the largest third parties. Real Chrome visitors: LCP 1,987 ms at the 75th percentile and CLS 0, both inside Google's "good" range.Source

Pick PostHog as the single source for the funnel and decide whether Google Analytics is still needed. Define six events: audit page viewed, form submitted, call booked, access granted, report delivered, pilot signed (the last three logged by the team). Tag sources (blog, Leeroopedia, Kapso, sector pages) and review the funnel weekly.

With a few dozen leads a quarter, every drop-off is visible and fixable. One tool avoids conflicting numbers and saves page weight.

• Losing historical GA data

  Export before removing anything.

• #5 Let audit requesters book the scoping call on the spot instead of waiting "one business day"

• Your websiteSite scan (homepage and /contact-us): Google Analytics, PostHog and Google Tag Manager installed; no live chat; no online booking; no pricing page; no privacy or cookie policy linked; no trust center.Source
• PageSpeedGoogle PageSpeed (November 2026): mobile performance 78, desktop 95; lab mobile LCP 3.4 s; the homepage downloads 20.3 MB in 29 requests, with about 8 MB savable through longer cache lifetimes, 1.8 s of render-blocking requests and 316 KB of unused JavaScript; Google Tag Manager (192 KB) and a jsDelivr-hosted script (183 KB) are the largest third parties. Real Chrome visitors: LCP 1,987 ms at the 75th percentile and CLS 0, both inside Google's "good" range.Source

Package a limited Uniforge build (entity resolution and data-quality agents only, two systems, read-only) as a deployable template or cloud marketplace listing. It produces a summary report in the buyer's account: duplicates, conflicts and missing links by system pair, with dollar estimates where billing data is present. The full audit, ML opportunities and action agents stay with the team-led audit and the pilot.

It removes the team's time from the top of the funnel, lets security teams inspect what they deploy, and turns curious engineers into qualified leads who arrive with their own numbers.

• Free tool cannibalises the paid pilot

  Keep action agents and ML opportunities in the paid path.

• Support load

  Private beta first, with docs and a community channel.

• #3 Explain how "no setup on your side" works when the audit "runs entirely in your cloud": a one-page access guide for the buyer's security team
• #11 Add Uniforge documentation to docs.leeroo.com, which today covers only Kapso

• Your websiteFree audit page: "See what your data is really telling you."; "FREE AUDIT · 48 HOURS"; connects "2–3 of your systems" with "No setup on your side" and "Runs entirely in your cloud"; finds duplicates, conflicts and missing links plus "ML opportunities in your data, quantified by value"; the form asks for name, work email, company, company size, "Which systems should we look at?" and "Anything else?"; the team will "reply within one business day"; no sample report, testimonial or customer name.Source
• Your websiteHomepage (checked November 9, 2026): H1 "A living model of your business. Built by AI. Run in your cloud."; Uniforge "connects to every system you operate" and ships agents; CTAs "Request a free audit", "Read the specification" and "Watch the demo"; proof from an unnamed "$1B enterprise" with 35+ systems: 9×10⁹ rows learned, a 62% cut in cross-system conflicts and about $20M of trapped value surfaced; runs in the customer's VPC on AWS, Azure or GCP, or fully on-prem and air-gapped; logos (Meta, Microsoft, Amazon, Hugging Face and others) shown as companies whose engineers use the team's open-source work; London, backed by Y Combinator.Source
• Y CombinatorYC profile: Spring 2025 batch, team size 3, London; describes Leeroo as building trainable AI agents for data and AI teams that run on-prem or in a customer's VPC, with target pilot partners in banking, finance and healthcare.Source

With customer consent built into the audit terms, collect anonymised aggregate statistics from each audit (conflict rates by system pair, duplicate rates, value per million records). Publish an annual report with a methods section, an interactive benchmark ("how does your CRM and billing conflict rate compare?") and a press and LinkedIn launch. Offer each participant their own percentile.

Original benchmark data earns press, analyst and AI-assistant citations, and the percentile comparison gives every reader a reason to run an audit.

• Too few audits for credible statistics

  Publish ranges and sample sizes, or wait a cycle.

• #2 Show buyers what the free 48-hour audit delivers: publish an anonymised sample report from the $1B enterprise

• Your websiteHomepage (checked November 9, 2026): H1 "A living model of your business. Built by AI. Run in your cloud."; Uniforge "connects to every system you operate" and ships agents; CTAs "Request a free audit", "Read the specification" and "Watch the demo"; proof from an unnamed "$1B enterprise" with 35+ systems: 9×10⁹ rows learned, a 62% cut in cross-system conflicts and about $20M of trapped value surfaced; runs in the customer's VPC on AWS, Azure or GCP, or fully on-prem and air-gapped; logos (Meta, Microsoft, Amazon, Hugging Face and others) shown as companies whose engineers use the team's open-source work; London, backed by Y Combinator.Source
• Your websiteFree audit page: "See what your data is really telling you."; "FREE AUDIT · 48 HOURS"; connects "2–3 of your systems" with "No setup on your side" and "Runs entirely in your cloud"; finds duplicates, conflicts and missing links plus "ML opportunities in your data, quantified by value"; the form asks for name, work email, company, company size, "Which systems should we look at?" and "Anything else?"; the team will "reply within one business day"; no sample report, testimonial or customer name.Source
• Competitor researchCelonis: "Industrialize Enterprise AI"; a Context Model of process data pitched as "The missing piece of your Enterprise AI stack"; over 1,400 companies; "Try for free" next to "Join a demo".See the research Source

Recruit two or three mid-sized systems integrators. Offer the 48-hour audit as a pre-migration data check they resell or bundle, a revenue share on Uniforge licences they source, and training so their consultants can run deployments. Start with one integrator and one system pair (CRM and billing), and co-write one joint case study.

Integrators already have the executive relationships and the budget lines (migration, data quality). A pre-migration check gives them a reason to bring Leeroo into accounts it couldn't reach alone.

• Partners see Uniforge as replacing their services

  Position it as making their projects faster and less risky.

• #11 Add Uniforge documentation to docs.leeroo.com, which today covers only Kapso
• #12 One consistent integrations list, with a page per system pair Uniforge reconciles (for example Salesforce and Oracle Billing)

• Your websiteProduct page: "Learns your business. Acts on it."; six learning agents (Entity Resolution, Lineage Mapping, Process Mining, Event Learning, Skill Learning, LoRA Training) and six action agents (Billing-integrity, Data-quality, Churn sentinel, Cross-sell, Contract-leakage, Fraud sentinel); integration logos Oracle, ServiceNow, SAP, Salesforce, Workday and "G Cloud", while the diagram shows Salesforce, Oracle Billing, ServiceNow, Zendesk, Workday and a warehouse ("6 SYSTEMS · 113 TABLES · READ-ONLY"); worked examples such as "National Riverbend Services" and "EST. $5.0M/YR" not labelled as illustrative; "Days, not quarters"; "Approvals where you want them. Full audit log."; no pricing, named case study or security page.Source
• HiringFounding Engineer (London or remote, $10K–$15K a month, listed on Y Combinator): "Be a customer-first problem solver", working with customers from discovery through delivery and turning requests into scoped plans with clear ROI; joins pre-sales and discovery calls; connects Leeroo's agents to customer systems and feeds reusable patterns back to research and product; nice to have: privacy, security and model governance such as EU AI Act readiness. The post does not mention Uniforge or Kapso.Source
• Y CombinatorYC profile: Spring 2025 batch, team size 3, London; describes Leeroo as building trainable AI agents for data and AI teams that run on-prem or in a customer's VPC, with target pilot partners in banking, finance and healthcare.Source

Make the intern's first project a public, synthetic multi-system entity-resolution and conflict benchmark (CRM, billing, ITSM and support tables with realistic noise), released on Hugging Face with a leaderboard and a paper. Report Uniforge's learning agents on it alongside open baselines, and link the dataset card to the audit.

A benchmark defines the problem in Leeroo's terms, attracts researchers and engineers from target enterprises, and gives buyers an independent-looking yardstick, the way RelBench already works for Kapso.

• Synthetic data is seen as unrealistic

  Base noise patterns on anonymised statistics from real audits.

• HiringResearch Scientist Intern, PhD (AI Agentic Systems) (London or remote, 3 to 6 months, $6.5K–$10K a month): "Design and run state-of-the-art machine-learning experiments to extend our agentic Learning Engine"; "Publish results through papers, datasets, and open-source code"; successful prototypes "move quickly into production".Source
• Open sourceHugging Face organization "leeroo": one public model, last updated April 28, 2024; no public datasets; lists the KAPSO paper (arXiv 2601.21526).Source
• Your websiteProduct page: "Learns your business. Acts on it."; six learning agents (Entity Resolution, Lineage Mapping, Process Mining, Event Learning, Skill Learning, LoRA Training) and six action agents (Billing-integrity, Data-quality, Churn sentinel, Cross-sell, Contract-leakage, Fraud sentinel); integration logos Oracle, ServiceNow, SAP, Salesforce, Workday and "G Cloud", while the diagram shows Salesforce, Oracle Billing, ServiceNow, Zendesk, Workday and a warehouse ("6 SYSTEMS · 113 TABLES · READ-ONLY"); worked examples such as "National Riverbend Services" and "EST. $5.0M/YR" not labelled as illustrative; "Days, not quarters"; "Approvals where you want them. Full audit log."; no pricing, named case study or security page.Source

Add a one-page "For your CFO" summary to every audit: the total dollar value found, the three largest items, the confidence level, what a 30-day pilot would recover, and a single next step. Make it easy to forward (PDF and a short email template), and offer the champion a 20-minute joint call with their finance lead.

Enterprise deals move when the budget holder sees money, and the audit already quantifies it. Internal referral from champion to CFO is the cheapest path to a pilot decision.

• Overstated dollar figures damage trust with finance

  Show confidence ranges and the method in one line.

• #2 Show buyers what the free 48-hour audit delivers: publish an anonymised sample report from the $1B enterprise

• Your websiteFree audit page: "See what your data is really telling you."; "FREE AUDIT · 48 HOURS"; connects "2–3 of your systems" with "No setup on your side" and "Runs entirely in your cloud"; finds duplicates, conflicts and missing links plus "ML opportunities in your data, quantified by value"; the form asks for name, work email, company, company size, "Which systems should we look at?" and "Anything else?"; the team will "reply within one business day"; no sample report, testimonial or customer name.Source
• HiringFounding Engineer (London or remote, $10K–$15K a month, listed on Y Combinator): "Be a customer-first problem solver", working with customers from discovery through delivery and turning requests into scoped plans with clear ROI; joins pre-sales and discovery calls; connects Leeroo's agents to customer systems and feeds reusable patterns back to research and product; nice to have: privacy, security and model governance such as EU AI Act readiness. The post does not mention Uniforge or Kapso.Source
• Your websiteHomepage (checked November 9, 2026): H1 "A living model of your business. Built by AI. Run in your cloud."; Uniforge "connects to every system you operate" and ships agents; CTAs "Request a free audit", "Read the specification" and "Watch the demo"; proof from an unnamed "$1B enterprise" with 35+ systems: 9×10⁹ rows learned, a 62% cut in cross-system conflicts and about $20M of trapped value surfaced; runs in the customer's VPC on AWS, Azure or GCP, or fully on-prem and air-gapped; logos (Meta, Microsoft, Amazon, Hugging Face and others) shown as companies whose engineers use the team's open-source work; London, backed by Y Combinator.Source