Clawback

Inspiration

I built Clawback because people often have money sitting in their inbox that they never end up claiming. A price drops after a purchase, a delivery arrives late, or a product is still under warranty, but following up is usually too much effort.

I wanted to build an AI agent that could do that work for you. Clawback reads retail emails, connects related information across different emails, checks whether something is actually claimable, calculates the deadline, and drafts the claim for you to review and send.

Building it taught me that this is more than an email extraction problem. The useful part is connecting information over time. For example, a receipt from last week can become relevant when a sale email arrives today. I also learned how important confidence, evidence, and validation are when an AI system is dealing with people's money.

The biggest challenges were linking emails to the correct purchases, handling changing retailer policies, and making sure the AI never invents amounts or claims. I addressed this with structured extraction, policy checks, confidence scores, evidence tracking, and a deterministic fallback when no AI model is available.

The result is Clawback: an AI agent that finds money you may be entitled to get back and turns it into an actionable claim.