Speak Diligence Research Report
Proceed to confirmatory diligence only if data-room materials validate ARR quality, cohort retention, gross margin after model/platform costs, enterprise repeatability, privacy/audio-data compliance, cap-table economics and a durable product moat against AI-native and incumbent competitors.
Speak Diligence Research Report
Speak has credible public evidence of unicorn financing, strong consumer-language app traction, a differentiated speaking-first AI narrative, OpenAI/Accel investor validation and active B2B expansion. The investment case remains highly diligence-dependent because current ARR, margins, retention, customer concentration, cap table, AI supplier economics, privacy posture, IP/data rights and legal/insurance status are not public.
Speak competes with Duolingo and multiple AI conversation/tutor startups with overlapping claims.
Diligence request: Run win/loss calls, pricing comparisons, feature benchmarks and CAC channel analysis.
Public valuation reportedly doubled from $500M to $1B in roughly six months; growth must justify a premium private-market mark.
Diligence request: Test downside cases, runway, burn multiple, forward ARR and financing terms.
OpenAI/Azure access and GPT-4/Whisper usage may be central to product quality and cost structure.
Diligence request: Review supplier contracts, fallback models, cost curves, latency SLOs and data-processing terms.
Speech/audio, user content, contact info, identifiers, minors restrictions and country-specific notices create a multi-jurisdiction compliance burden.
Diligence request: Request DPIAs, DPA templates, subprocessors, retention/deletion controls, age-gating, incident history and legal opinions.
Public sources disclose financing and partial ARR/subscriber signals, but audited financial statements, current ARR, margins, burn, retention and customer concentration are not public.
Diligence request: Open accounting data room; reconcile ARR to billing, cash, contracts and GAAP revenue.
B2B claims cite 200+ customers/brands but public materials do not name customers, ACV, retention or seat usage.
Diligence request: Request named customer list, top contracts, references, pipeline, quota attainment and expansion/churn metrics.
Company operates across Korea, Japan, Taiwan, Europe/LatAm expansion and multiple languages; local privacy, employment, payment and content compliance are material.
Diligence request: Review country P&Ls, local counsel memos, contractor/FTE classification and tax/payment compliance.
Team reportedly ~130 with 40 listed open roles across functions and geographies; org design, attrition and compensation are not public.
Diligence request: Review HRIS, attrition, comp bands, option plan, hiring plan and management succession.
Public evidence strongly supports Speak’s funding milestones and unicorn valuation, but financial statements, current ARR quality, burn, margins and cap structure remain non-public.
not publicly verifiable confidence: low
Public record contains dated ARR/subscriber signals, app-store monetization indicators and funding context, but no audited financial statements or quarterly P&L.
| signal | public value | source read | financial relevance | gap |
|---|---|---|---|---|
| ARR/subscribers | Double-digit million ARR; nearly 100,000 paying subscribers in Series B era | Company financing post | Only direct public revenue signal | Current ARR, NRR, churn and revenue recognition unknown |
| Subscribers | Hundreds of thousands of subscribers in 30+ countries | Careers page | Suggests growth after Series B | Paid definition, active status and platform/country split unknown |
| Usage | >1B sentences in 2024 and 25M personalized lessons | Series C / Accel | Engagement may support retention and upsell | Free versus paid and cohort retention unknown |
| App-store monetization | Google Play in-app purchases and monthly/annual subscriptions | Google Play | Confirms consumer subscription monetization | Net receipts after platform fees and refunds unknown |
| topic | public evidence | why it matters | request |
|---|---|---|---|
| Current ARR and margins | Only dated ARR/subscriber signals are public | Determines valuation support and burn multiple | Monthly ARR/MRR, billings, revenue recognition, gross margin, COGS and churn |
| Cap table and preferences | Series C valuation and funding total are public | Economic ownership and downside value depend on preferences | Cap table, charter, investor rights, side letters, liquidation waterfall |
| Cash/runway and budget | Raised $78M Series C and >$150M total | Expansion plan may consume substantial cash | Cash balance, burn, runway, debt, budget-to-actuals and hiring plan |
| Platform and model costs | App stores and OpenAI/Azure dependencies public | Platform fees and inference costs affect gross margin | Supplier spend, platform fees, refunds, cloud commitments and cost-per-lesson |
Valuation points are company-reported financing valuations, not public market prices.
not publicly verifiable confidence: low
Public sources imply aggressive international and B2B expansion, but no board-approved plan, budget, runway or hiring model is public.
partially verified confidence: medium
Financing milestones are well publicized through the $78M Series C at $1B valuation, but cap table, preferences and reconciliation of funding totals are non-public.
| date | round | amount | valuation | investors or sources | diligence read |
|---|---|---|---|---|---|
| 2022 | Series B | $27M | Not disclosed publicly in source | OpenAI Startup Fund led; Founders Fund participated | Validates early strategic OpenAI relationship and double-digit million ARR signal. |
| 2023 | Series B-2 | $16M | Not disclosed publicly in source | Lachy Groom led; OpenAI Startup Fund participated | Funded U.S./international expansion; total funding reported as $63M. |
| 2024-06-18 | Series B-3 | $20M | $500M | Buckley led; OpenAI Startup Fund and Khosla participated | Sets midpoint for six-month valuation doubling to unicorn round. |
| 2024-12-10 | Series C | $78M | $1B | Accel led; OpenAI Startup Fund, Khosla Ventures, YC named | Core verified unicorn claim; request financing documents. |
| 2026-06-01 | Funding-total database profile | $165.81M total raised | Private company / revenue hidden | CB Insights | Slightly above company $162M to date; reconcile cap table. |
Public amounts may include different treatment of small rounds or secondaries.
not publicly verifiable confidence: low
Tax, banking, debt, payables, off-balance-sheet commitments, refunds and platform fees are not disclosed publicly.
Speak’s public product is a speaking-first AI language tutor sold through consumer app subscriptions and Speak for Business; product efficacy, margins and roadmap require data-room validation.
partially verified confidence: medium
Speak publicly offers a mobile AI language tutor, subscriptions and Speak for Business; product breadth and app-store ratings are visible, but product-level revenue, technical efficacy and roadmap remain private.
| product or sku | target user | public features | monetization | evidence read |
|---|---|---|---|---|
| Speak mobile app | Consumer learners | AI tutor, speaking practice, pronunciation/intonation/fluency feedback, supported language learning | Monthly/annual auto-renewing subscription and in-app purchases | Visible in app stores and homepage |
| Speak for Business | Organizations/employees | AI language-learning access for employee language training | Enterprise/B2B subscription not priced publicly | 200+ brands/customers claim; ACV unknown |
| Localized language/curriculum content | Learners across Korea, Japan, Europe, LatAm, U.S. and other markets | 15+ languages in job descriptions; app metadata shows 16 language codes and language-learning targets | Supports consumer and B2B retention/expansion | Content/localization hiring and app metadata visible |
| dependency or risk | public indicator | impact | diligence test |
|---|---|---|---|
| AI model and speech stack | GPT-4/Whisper and OpenAI/Azure resource access reported | Quality, cost, uptime and data-use obligations | Benchmark against alternatives; review supplier terms and fallback architecture |
| App-store platforms | Google Play and Apple are core distribution surfaces | Platform fees, policy risk, refund exposure and ranking dependency | Review net revenue by platform, refunds and policy incidents |
| International language/content quality | 15+ languages and multiple country launches | Localization cost, content QA and compliance complexity | Review language-level retention, content production costs and local counsel memos |
Customer evidence shows strong app-store and company-reported engagement plus early enterprise claims, but top customers, revenue concentration and supplier economics are not public.
partially verified confidence: medium
Consumer adoption signals are strong, including app-store scale, Korea penetration and company-reported learning activity, but top users/customers are not identifiable from public evidence.
| metric | public value | source | diligence read |
|---|---|---|---|
| Total downloads | 15M+ on homepage; 10M+ Google Play bucket; B2B page says 10M global downloads | Speak/Google Play | Corroborates scale but not active/paid users |
| Ratings/reviews | 4.8 homepage; Google Play 4.7/114K reviews; Apple 4.83394/46,399 ratings | Speak/app stores | Positive quality signal; may vary by market and cohort |
| Learning activity | >1B sentences in 2024; 25M personalized lessons | Series C / Accel | Engagement metric requires active paid/free split |
| Geography | 10M+ learners in 40+ countries; nearly 6% Korean population; hundreds of thousands of subscribers in 30+ countries | Company posts/careers | Korea-led traction with international expansion |
Not a single-unit KPI chart; intended for public-evidence scale context.
partially verified confidence: medium
OpenAI, Accel, Khosla, YC and business customers are strategic to the story, but the legal/economic terms of these relationships are private.
| relationship | role | public evidence | key diligence question |
|---|---|---|---|
| OpenAI Startup Fund / OpenAI / Azure resources | Investor and AI infrastructure/model dependency | Series B led by OpenAI Startup Fund; GPT-4/Whisper and OpenAI/Microsoft Azure resources reported | What are model, data-use, pricing, exclusivity and termination terms? |
| Accel | Series C lead investor | Accel led Series C and published investment rationale | What governance/information/control rights did Accel receive? |
| Apple / Google app stores | Consumer distribution and payments | Live app listings with subscriptions/in-app purchases | What net revenue, refunds and policy incident history exist by platform? |
| B2B customers/brands | Enterprise revenue expansion | 200+ customers/brands and 85% adoption claims | Which accounts are paid, recurring, referenceable and expanding? |
not publicly verifiable confidence: low
No public customer-level revenue data exists; Speak likely mixes consumer subscription revenue with emerging enterprise accounts.
not publicly verifiable confidence: low
No public evidence identified severed customer, investor, supplier or platform relationships; absence of disclosure is not proof of absence.
partially verified confidence: medium
OpenAI/Microsoft Azure are specifically reported as technology resources; broader cloud, app-store, payment and content suppliers are not public.
Speak competes in a crowded language-learning and AI tutor market; public differentiation centers on speaking-first AI and Korea traction, but durable moat and market share are unproven.
partially verified confidence: medium
Speak competes in consumer language learning, AI speaking tutors and enterprise language training against scaled incumbents and AI-native challengers.
| segment | examples | speak position | risk read |
|---|---|---|---|
| Scaled language-learning incumbents | Duolingo and other broad app platforms | Speaking-first AI tutor with Korea traction | Incumbents have massive brand, free user base and language breadth |
| AI speaking tutors | Loora, Praktika.ai, Tutor Lily, Toko and similar AI conversation products | Large funding, OpenAI relationship and app-store traction | Feature differentiation may compress as LLM voice agents improve |
| Local/regional language apps | Yanadoo, ELSA and regional learning brands cited in press | Strong Korea adoption; expanding internationally | Local incumbents may defend language-specific niches and acquisition channels |
| basis | public speak signal | competitor pressure | diligence test |
|---|---|---|---|
| Speaking-first efficacy | AI speech feedback, WER improvements and >1B sentences | AI tutor competitors and incumbents can add speaking features | Independent proficiency/retention study versus alternatives |
| Scale and distribution | 10M+/15M+ downloads and high ratings | Duolingo has far larger public learner scale | CAC, retention and paid conversion by channel/country |
| Enterprise readiness | 200+ B2B customers/brands and B2B hiring | Corporate training vendors and incumbents may bundle offerings | Enterprise references, security/compliance packet and ACV retention |
Scores are analyst qualitative placements from public evidence, not market-share estimates.
GTM appears to be evolving from consumer app-store growth into enterprise and geographic expansion; channel economics and sales productivity are not public.
partially verified confidence: medium
Public GTM appears to combine app-store consumer acquisition, Korea-led brand strength, content/localization and expanding enterprise sales.
| channel | public evidence | likely role | diligence gap |
|---|---|---|---|
| App stores | Google Play and Apple listings with ratings, reviews and subscriptions | Consumer acquisition and payment conversion | ASO, paid/organic split, refunds and platform net revenue |
| Localized market launches | Korea penetration, 30+/40+ country claims and 15+ languages | Country-by-country growth wedge | Country P&L, launch CAC, language-level retention |
| Enterprise outbound/sales | Speak for Business and 11 B2B open roles | Higher-ACV diversification | Pipeline, quota, ACV, payback and enterprise churn |
| Investor/AI brand halo | OpenAI Startup Fund, Accel, Khosla, Founders Fund references | Credibility and recruiting/customer trust | Attribution and durability if model ecosystem changes |
| signal | public value | interpretation | request |
|---|---|---|---|
| B2B roles | 11 listed roles in B2B department | Active investment in enterprise GTM | Sales org plan, quotas, pipeline and attainment |
| Regional sales roles | Korea SDR, Japan AE/SDR, Mexico City/LATAM SDR | International enterprise expansion | Country sales productivity and compliance plan |
| B2B customer claim | 200+ brands/customers; 85% employee adoption in company/investor sources | Demand signal but no named logos or ACV | Customer list, contracts, usage, renewal and references |
Counts mix different units; used to highlight conversion gaps.
partially verified confidence: medium
Speak cites 200+ business customers/brands, but public materials do not name major customers or disclose account economics.
partially verified confidence: medium
New business appears to come through app stores, localized content/language launches, Korea/Japan expansion, enterprise outbound and brand/investor visibility.
not publicly verifiable confidence: low
Public hiring shows an enterprise sales buildout, but quota capacity, pipeline, ACV, payback and sales efficiency are not public.
not publicly verifiable confidence: low
Speak raised substantial capital, but marketing budget sufficiency cannot be assessed without burn, CAC, country plan and runway.
Speak’s R&D story is AI speech/personalization plus content/localization, supported by OpenAI-linked technology signals and engineering hiring; proprietary benchmarks, roadmap and IP need validation.
partially verified confidence: medium
R&D appears centered on AI speech, personalization, mobile engineering and content/curriculum, with public engineering and ML hiring; exact org and spend are private.
| area | public indicator | diligence read |
|---|---|---|
| Speech and AI | GPT-4/Whisper usage; WER reduction and speedup claims | Differentiation depends on reproducible speech benchmarks and model orchestration. |
| Engineering roles | AI Product Engineer, ML Engineer Voice, backend, platform, mobile and full-stack roles listed | Hiring supports in-house technical investment. |
| Learning and curriculum | Learning & Curriculum roles and bilingual actors/script supervision | Content pipeline is a key operational capability and cost center. |
| Vendor architecture | OpenAI/Microsoft Azure resource access reported | Supplier terms, fallback, data privacy and inference economics are key. |
partially verified confidence: medium
Public posts and listings imply ongoing language expansion, speech-recognition improvements, real-time AI tutoring and B2B features, but pipeline timing and economics are private.
| theme | public evidence | validation needed |
|---|---|---|
| Speech recognition quality | >60% WER reduction and 20% speedup claim | Benchmark datasets, baselines, language coverage and statistical confidence |
| Language expansion | App metadata includes 16 language codes; jobs mention 15+ languages and bilingual content roles | Language-level retention, launch cost and content quality metrics |
| B2B productization | Speak for Business and B2B hiring | Admin features, reporting, SSO/security, procurement and ROI proof |
| Real-time AI tutoring | Homepage and app descriptions emphasize real-life conversation and instant speech feedback | Latency, uptime, safety, fallback and model-cost benchmarks |
Founders and team footprint are publicly visible, with about 130 employees and 40 listed roles; management depth, compensation, attrition and equity plans remain diligence gaps.
not publicly verifiable confidence: low
Only high-level team footprint and founder identities are public; no reporting-line org chart was available.
| person or role | public role | evidence | diligence request |
|---|---|---|---|
| Connor Zwick | Co-founder | Listed by YC and Accel; referenced by TechCrunch; careers page cites Flashcards+ exit with co-founder history | Employment agreement, equity, current responsibilities, succession and references |
| Andrew Hsu | Co-founder | Listed by YC and Accel; referenced by TechCrunch; careers page cites Flashcards+ exit with co-founder history | Employment agreement, equity, current responsibilities, succession and references |
| Board/investor representatives | Not fully public | Accel led Series C; OpenAI Startup Fund, Khosla, YC and Founders Fund named across sources | Board minutes, investor rights, observers, consent rights and governance policies |
| metric | public value | diligence interpretation |
|---|---|---|
| Team size | Team of 130 across San Francisco, Seoul, Tokyo, Taipei and Ljubljana | Current headcount anchor; requires HRIS validation |
| Open roles total | 40 listed jobs | Large hiring signal relative to stated team size |
| Largest hiring functions | B2B 11; Engineering 10; Learning & Curriculum 9 | Investment in enterprise, product/AI infrastructure and content supply |
| Largest hiring locations | San Francisco 16; Seoul 9; Los Angeles 8; Tokyo 4; Taipei 1; remote/LATAM 2 | Distributed operational complexity and regional GTM/content expansion |
| External profile | LinkedIn profile: San Francisco software-development company with 25,803 followers | Brand/recruiting visibility signal, not audited headcount |
Open-role counts are access-date snapshots.
partially verified confidence: medium
Careers materials state ~130 team members and Ashby shows 40 open roles; historical/projected headcount by function and location is not public.
partially verified confidence: medium
Founders Connor Zwick and Andrew Hsu are publicly corroborated; broader executive bench, board and functional-lead bios are incomplete publicly.
not publicly verifiable confidence: low
No public compensation arrangements, founder employment agreements, severance terms or bonus plans were available.
not publicly verifiable confidence: low
No option plan or equity-incentive details are public; repeated financings likely require option-pool diligence.
not publicly verifiable confidence: low
No public employee-relations problems were identified in reviewed materials, but no legal/HR docket review was performed.
| area | public status | risk if unresolved | request |
|---|---|---|---|
| Lawsuits against/by company | No material matter verified in reviewed public materials; not exhaustive | Hidden claims, reserves or injunctions | Counsel-led docket search under all aliases and jurisdictions |
| Insurance | No policies or claims history public | Uncovered cyber/privacy/IP/employment claims | Policy schedule, limits, exclusions and claims history |
| Material contracts | Customer/supplier/app-store/model contracts not public | Unfavorable termination, data-use or minimum-commitment terms | Material contract schedule and top agreements |
| Regulatory agency issues | No agency correspondence public | Privacy, consumer auto-renewal, AI or minors enforcement | Regulatory correspondence, complaints and compliance audits |
Likelihood unknown means public evidence was insufficient, not low probability.
not publicly verifiable confidence: low
No turnover metrics are public; open-role volume could represent growth, replacement hiring or both.
Public terms and privacy materials show a multi-jurisdiction consumer/B2B legal surface; litigation, contracts, insurance, IP registrations and regulatory correspondence are not public.
inconclusive confidence: low
No material pending lawsuits against Speak were verified from public materials reviewed; this is not a comprehensive docket conclusion.
| request category | documents | priority reason |
|---|---|---|
| Corporate and financing | Charter, bylaws, board minutes, investor rights, side letters, consents | Validate governance, controls and financing economics |
| IP/data/content | IP assignments, trademark/patent schedules, data licenses, actor/content releases, OSS inventory | Speech/curriculum AI business depends on usable data and content rights |
| Privacy/security | DPIAs, DPAs, subprocessors, pen tests, SOC/ISO evidence, incident logs, deletion/retention policy | App handles user content/speech and international personal data |
| Commercial/supplier | Top customer MSAs, app-store/payment terms, OpenAI/Azure/cloud agreements, SLAs | Controls revenue durability, gross margin and supplier dependency |
inconclusive confidence: low
No public evidence showed Speak-initiated litigation; enforcement, IP or collection matters remain unknown.
inconclusive confidence: low
As a software/mobile AI company, environmental exposure appears limited publicly; employee safety and distributed-office compliance remain non-public.
partially verified confidence: medium
Speak owns/controls public service IP under its terms, but detailed patents, trademarks, model/data licenses, curriculum rights and open-source compliance are not public.
| topic | public evidence | diligence read |
|---|---|---|
| Privacy/user data | Country-specific privacy notices and App Store privacy categories for contact info, identifiers, usage data, user content and diagnostics | Requires DPIA, subprocessors, retention/deletion and data-residency review |
| Minors/consumer terms | Terms restrict minor use absent supervision and include arbitration/class waiver | Consumer subscription and age-gating compliance should be tested |
| IP/user content | Terms define User Content and reserve Speak/licensor rights | Training data, curriculum/media licenses and output rights need review |
| AI/vendor contracts | OpenAI/Azure usage and app-store distribution visible | DPAs, model terms, SLAs and fallback rights are material |
not publicly verifiable confidence: low
Insurance coverage is not public; AI speech, privacy, cyber, E&O, media/IP, employment and international operations create material coverage questions.
not publicly verifiable confidence: low
Public materials imply material contracts with investors, app stores, OpenAI/Azure/cloud, B2B customers, content providers and employees, but contract terms are not public.
inconclusive confidence: low
No regulatory agency problems were verified publicly, but multi-jurisdiction privacy, AI, consumer subscription and minors exposures require formal counsel review.
| ID | Claim | Status | Sources |
|---|---|---|---|
| EC-001 | Speak positions itself as an AI language-learning app focused on getting users speaking and states 15M+ downloads and a 4.8 rating. | partially verified medium | SRC-001 |
| EC-002 | Speak announced a $78M Series C at a $1B valuation led by Accel, with OpenAI Startup Fund, Khosla Ventures and YC among existing investors. | verified high | SRC-002SRC-022 |
| EC-003 | Speak reported $162M raised to date, valuation doubling in six months, >1B sentences spoken in 2024, 25M personalized lessons, and Speak for Business with 200+ customers and 85% employee adoption. | partially verified medium | SRC-002 |
| EC-004 | Speak announced a June 2024 $20M Series B-3 at a $500M valuation, $84M total funding, 10M+ learners in 40+ countries, five years of doubled learners, and speech-recognition improvements. | partially verified medium | SRC-003 |
| EC-005 | Speak announced a $16M Series B-2, $63M total funding, operation in >20 countries, nearly 6% Korean population penetration, and a No. 1 Education App Store rank in South Korea. | partially verified medium | SRC-004 |
| EC-006 | Speak announced a $27M Series B led by the OpenAI Startup Fund, double-digit million ARR, nearly 100,000 paying subscribers, almost 15M lessons started in 2022, and >10M pronunciation feedback items. | partially verified medium | SRC-005 |
| EC-007 | CB Insights lists Speak as founded in 2016, headquartered in San Francisco, alive/private-equity stage, revenue hidden, formerly Speakeasy Labs Korea Co., Ltd., and having raised $165.81M. | partially verified medium | SRC-012 |
| EC-008 | Accel publicly describes Speak as an AI-powered language-learning tutor and corroborates >10M downloads, 25M personalized lessons, 1B sentences in 2024, ~10% Korea trial, 200+ corporations, 85% employee adoption and 2025 expansion priorities. | partially verified medium | SRC-006SRC-007 |
| EC-009 | Speak careers materials state it is the No. 1 English app in South Korea, has hundreds of thousands of subscribers in 30+ countries, raised over $150M, and has a team of 130 across San Francisco, Seoul, Tokyo, Taipei and Ljubljana. | partially verified medium | SRC-008 |
| EC-010 | Speak for Business page states 200+ brands, 4.8 rating based on 100K+ ratings, 10M downloads globally and 100K+ ratings. | partially verified medium | SRC-009 |
| EC-011 | Google Play lists Speak by Speakeasy Labs with in-app purchases, 4.7 rating, 114K reviews, 10M+ downloads, Everyone rating, and monthly/annual subscriptions. | verified high | SRC-010 |
| EC-012 | Apple App Store/iTunes metadata lists Speak: Language Learning by Speakeasy Labs, Inc. with 4.83394 average rating, 46,399 ratings, version 4.50.0, Education category, 16 language codes and public privacy categories. | verified high | SRC-011 |
| EC-013 | TechCrunch corroborated the founders, $16M B-2 round, $63M total funding, 100K+ subscribers, roughly 3M Korean users, GPT-4/Whisper use, Azure/OpenAI resource access and competitor set. | partially verified medium | SRC-013 |
| EC-014 | Y Combinator lists Speak as a Winter 2017 active AI/Education company in San Francisco founded by Andrew Hsu and Connor Zwick. | verified high | SRC-014 |
| EC-015 | LinkedIn lists Speak as a San Francisco software-development company with 25,803 followers and an AI Language Tutor positioning. | partially verified medium | SRC-015 |
| EC-016 | Speak had 40 listed public Ashby roles at access time, led by B2B, Engineering and Learning & Curriculum, with roles across San Francisco, Los Angeles, Seoul, Tokyo, Taipei and remote locations. | verified high | SRC-016 |
| EC-017 | Speak global terms include mandatory individual arbitration, class-action/jury-trial waiver language, eligibility/minor restrictions, organizational-account provisions, user-content terms and subscription-plan references. | verified high | SRC-017SRC-019 |
| EC-018 | Speak maintains global and country-specific privacy notice links, indicating multi-jurisdiction privacy obligations. | partially verified medium | SRC-018SRC-011 |
| EC-019 | CB Insights alternatives list Loora, Toko, Tutor Lily, Praktika.ai and Duolingo; Duolingo publicly markets 40+ languages and very large learner scale. | partially verified medium | SRC-020SRC-021 |
| EC-020 | The public record contains strong fundraising and usage signals but lacks audited financial statements, current ARR, burn, gross margin, retention, customer concentration, cap table and board materials. | not publicly verifiable high | SRC-002SRC-005SRC-012 |
| EC-021 | Public fundraising totals differ slightly: Speak states $162M to date in its Series C post while CB Insights reports $165.81M raised. | inconclusive medium | SRC-002SRC-012 |
| EC-022 | OpenAI is both a named investor/round lead and a technical dependency signal through GPT-4, Whisper and Microsoft Azure/OpenAI resource access. | partially verified medium | SRC-005SRC-013 |
| EC-023 | Speak publicly claims technical speech-recognition improvements including >60% WER reduction and 20% speedup. | partially verified medium | SRC-003 |
| EC-024 | Speak is expanding beyond consumer Korea-led adoption into B2B and new geographies, evidenced by public B2B claims and B2B/GTM hiring. | partially verified medium | SRC-002SRC-009SRC-016 |
| EC-025 | No public source reviewed disclosed audited lawsuits, safety liabilities, insurance coverage, material contracts, patents/trademarks ownership or regulatory-agency disputes; these remain diligence gaps rather than cleared risks. | inconclusive low | SRC-017SRC-018SRC-019 |
This report is a public-evidence diligence snapshot, not investment advice. Important financial, legal, technical, and contractual facts remain non-public and should be verified directly with management and primary documents before any investment decision.