Common questions about IrisGo's business model, competition, and Series A fundraising
Our go-to-market is B2B: OEMs pre-install IrisGo on every AI PC they ship. Acer is signed, with additional OEMs in our pipeline. This gives us near-zero customer acquisition cost.
Our revenue is B2C (premium subscriptions starting from $14.9/mo) plus B2B (data annotation services for AI companies).
Think of the Spotify-carrier model: carriers distribute the app on every phone; Spotify converts users to paid plans. OEMs distribute for us. Users pay us.
CAC
≈ $0
OEM pre-installation
ARPU
from $14.9/mo
Premium subscription
Gross Margin
70-80%
Platform business
LTV:CAC
Exceptional
Denominator ≈ zero
Additional revenue streams from Skills Marketplace (30% platform take rate) and Data Annotation Network (B2B) further increase blended ARPU. We use subscription-only for conservative estimates.
Microsoft Copilot: One-way revenue (users pay $30/mo, Microsoft takes all)
IrisGo: Three-way profit platform (users earn, developers earn, we take commission)
Copilot: Windows built-in (Microsoft controls)
IrisGo: OEM pre-install partnership (we keep IrisGo brand, like Intel Inside)
OEMs get differentiation + Skills revenue share
Copilot: Closed system (developers can't participate)
IrisGo: Open platform (Skills Marketplace: 30% platform / 70% to developer)
Platform economics always beat closed products. Android beat Nokia for this exact reason.
IrisGo provides earning platform (users do annotation, earn money)
Copilot is just productivity tool — can't be a unique PC selling point
OEMs get revenue share from Skills sales
We split our 30% commission with OEMs
Transform from "selling hardware" to "hardware + recurring revenue"
"Powered by IrisGo" (like Intel Inside)
Not white-label — OEMs can showcase partnership with innovative AI platform
100M+ OEM pre-install = instant distribution
GPT Store needs organic growth
PC user LTV > mobile app user
Stable subscription model (not just one-time payment)
We're the first AI PC Skills Marketplace
Developers can capture the AI PC category early
We will never do white-label. This is our core principle.
Value accumulates in IrisGo, not individual OEMs
User loyalty: Switch PCs but still use IrisGo (data doesn't get lost)
Developers create IrisGo Skills (cross-OEM)
Not Acer Skills / HP Skills (fragmented)
Intel Inside: Kept Intel brand
Android by Google: Kept Android brand
These prove brand partnership model works
AI-in-cloud is already mature (ChatGPT, Claude)
AI-on-device is the next wave (100M+ AI PCs)
Not just AI assistant (product layer)
It's AI PC's operating layer (platform layer)
Three-engine flywheel creates network effects
Lman (COO): 25+ years cross-domain hardware-software integration, OEM relationship builder
Jeffrey (CEO): ex-Apple Siri Engineering Manager, CMU ECE
Signed Acer within 3 months (execution speed)
100M+ pre-installed users = solve cold start problem
Competitors need $50/user CAC to acquire customers
Three-engine flywheel: Users ↔ Developers ↔ OEMs
More users → More developers → More Skills → Platform value ↑
Annotation data accumulation
User behavior data (context understanding)
Developer best practices (which Skills work best)
Developer ecosystem (investment cost)
User data migration cost
IrisGo = de facto standard for AI PC Skills
10 Years Later:
Just like iOS/Android defined smartphone ecosystems, IrisGo will define the AI PC ecosystem. First-mover advantage + ecosystem = hard to replace.
12 months execution + 6 months buffer
Target: Series B ready ($50M+ ARR)
Dell, HP, Lenovo, Acer, ASUS — full AI PC lineup
Not niche products, mainstream upgrade
Risk: If no more OEMs sign after Acer
Mitigation: HP meeting scheduled, ASUS MOU in review
Risk: If Microsoft restricts third-party AI platforms on Windows
Mitigation: We're complement (supplementary), not replacement
Risk: If developers don't want to create IrisGo Skills
Mitigation: OpenClaw already has 10K+ users (developers love us)
But Opportunity >> Risk:
$454M SAM, we only need 10% to reach $45M ARR.
IrisGo collects the basics needed to run an account: email, phone number, and authentication credentials. That lets you save your history, personalize your experience, and reach support. We also collect limited usage data such as feature interactions and crash logs, plus basic technical details like OS version and hardware, to keep the product stable and compatible.
The part that matters for sensitive work sits apart from all of that. The workflow memory Watch & Learn builds is stored and runs on your device. It does not sync to a shared cloud, so sensitive workflow data never travels across external infrastructure.
Personal account data is deleted when you cancel or terminate your account.
The user does. IrisGo only learns a workflow when you deliberately start Watch & Learn for a specific task. It does not run in ambient monitoring mode and does not observe your activity in the background.
Every automation traces back to a decision you made. There is no silent capture running behind the scenes, which is the opposite of always-on screen recording.
IrisGo runs only what it was taught, in the sequence it was taught. It does not invent new actions on its own or improvise when it meets an unexpected screen state.
Because of that, it cannot be prompted or manipulated into acting outside its learned scope. Compared with generative AI agents that hold broad, open-ended action authority, this bounded-execution model is a real difference in safety and predictability, and it matters more, not less, in regulated settings.
IrisGo runs a hybrid architecture. Core learning and workflow execution happen on-device. The AI reasoning layer is the component that engages a cloud model.
That reasoning layer is model-swappable by design. The on-device execution model stays the same no matter which model sits behind it, so the AI layer can adapt to a customer's requirements, including deployments where the model must be self-hosted or served through an approved provider. Swapping the model does not require re-architecting how IrisGo works.
The workflow models Watch & Learn builds live on your device and can be deleted by you at any time.
For everything else, IrisGo maintains a formal Data Retention Policy: personal data is kept only as long as it is needed to run the service, and is deleted when you cancel or terminate your account. Data is encrypted in transit and at rest, employee access to user information is limited, and MFA is available to protect accounts.
No. The two run on separate tracks, and personal workflow data never crosses between them.
Your Watch & Learn memory stays on your device and is never sold. It is not an input to the annotation network, and it never becomes training data for anyone.
The Data Annotation Network is opt-in and separate by design. A user who chooses to take part completes discrete labeling tasks on data provided for that task, processed on-device, and earns GoT credits for the work. Taking part is an explicit choice, never a background harvest.
That separation is the entire point: your own context stays private and local, while any contribution to model training is deliberate, compensated, and anonymized by design.
We don't compete — they're "developer tools", we're "platform layer".
2028-2030 timeline
$50M+ ARR, 100%+ growth rate
Benchmark: Unity (IPO $13B), Figma (pre-acquisition $10B valuation)
Potential buyers: Google, Meta, Microsoft (ironically)
Acquisition rationale: AI PC platform layer + developer ecosystem
Like Notion, Figma (before acquisition)
Platform companies can operate independently long-term
Our goal is IPO, but we won't rule out strategic offers (if valuation > $5B).
Ready to discuss IrisGo's Series A opportunity?
Let's build the Android for AI PCs together.