Clawback

Company: Clawback

Self-reported purpose: A mobile-first financial housekeeping tool that uses AI to parse trial emails and track credit card perks, aiming to prevent unexpected charges by surfacing next actions before money slips away.

What changed: The project is a self-contained, deployed product built over a hackathon period (as per author's account), incorporating AI-powered email parsing via GPT-5.6, structured output validation, and user-controlled task management. It is not evidenced to have traction, revenue or customers.

Single most important open question: Does the AI extraction pipeline reliably surface actionable financial tasks from real-world emails, or does it fail in edge cases that would undermine user trust?

The description states that Clawback is a mobile-first application built with Expo and React Native, with web support. It allows users to paste financial emails (e.g., trial or credit card benefit notifications) into the app, where GPT-5.6 extracts structured data such as provider, task type, deadline, amount, and action URL.

Key technical components include:

• Supabase for backend authentication and storage
• PostgreSQL for data persistence
• Edge Functions for AI processing
• NativeWind and React Native Reanimated for UI
• Expo Router for navigation

The app does not access inboxes directly; it only parses pasted text. It presents tasks in a dashboard with three metrics: Available, At Risk, and Clawed Back.

Users can manually create tasks or review AI-extracted ones before saving. Tasks are ranked by deadline and financial importance. Completion is handled via swipe-to-strike or explicit button, with an 8-second undo window.

Not evidenced: No revenue model, customer base, usage metrics, or actual user behavior data.

The author states that Clawback aims to be more useful than a generic reminder app. It positions itself as a focused financial housekeeping tool that understands what is at stake and keeps users in control.

It claims to:

• Turn messy financial communications into clear, prioritized tasks
• Surface next actions before money slips away
• Deliver the satisfaction of striking out subscriptions just before they charge

The product is described as not autonomous — it does not cancel subscriptions or redeem benefits. Instead, it organizes work and leaves every financial action under user control.

Inference: The positioning reflects a niche focus on personal finance task management, possibly targeting users who are financially conscious but overwhelmed by scattered deadlines.

The description implies an individual user base — someone who:

• Manages multiple subscriptions
• Receives frequent trial emails or credit card perks
• Values control over financial actions
• Is tech-savvy enough to paste email text into a mobile app

There is no evidence of segmentation beyond this general persona.

Not evidenced: No defined buyer personas, customer interviews, or user research data.

The description does not state any business model or pricing strategy. It also does not mention monetization plans, partnerships, or revenue streams.

Not evidenced: No information on how the product will generate income.

The app is built using:

• Mobile-first architecture (Expo + React Native)
• Supabase for backend services
• GPT-5.6 for AI processing with strict schema output
• Edge Functions for server-side logic
• TypeScript and NativeWind for UI
• Jest for testing (42 test suites, 280 passing tests)

Key engineering decisions include:

• Mandatory human review of AI extractions
• Pessimistic writes to prevent UI inconsistency
• Fail-closed URL handling
• No automatic financial actions
• Local demo mode for testing without backend setup

The system uses deterministic normalization and validation steps to ensure accuracy.

Inference: The technical stack suggests a lean, modern SaaS-like approach with strong emphasis on security and user control.

There is no evidence of:

• Revenue
• Customers
• User engagement metrics
• Product adoption
• Market traction

The project is described as a hackathon submission and a deployed demo. It includes a local demo mode, but there are no signs of production-scale usage or growth.

Not evidenced: No data on user retention, active users, or product performance in real-world conditions.

The description does not mention competitors or market positioning relative to existing tools.

Not evidenced: No competitive analysis, benchmarking, or awareness of similar products.

• AI reliability: The system depends heavily on GPT-5.6 for extraction. If the AI fails to correctly parse emails in edge cases, it could erode trust.
• User control vs. convenience trade-off: While user control is a strength, it may also slow down task completion or reduce adoption if users find it too cumbersome.
• Limited scope: The app only parses pasted emails and does not integrate with inboxes or other financial tools — this limits its utility for many users.
• No monetization strategy: No indication of how the product will be monetized, which raises questions about long-term viability.

1. How often does the AI extraction pipeline fail to correctly identify deadlines or values in real-world emails?
2. What is the expected user journey for someone who wants to use Clawback regularly?
3. Are there plans to integrate with email providers or financial institutions, and what are the technical challenges involved?
4. How do you plan to scale beyond a single developer (Selene Yang)?
5. What are your thoughts on privacy implications of storing pasted email content, even if not persisted?