Work

A few things I've led professionally.

Oatfi · First PM, embedded lending

Automated underwriting. I took what existed, the underwriting endpoints, and made them executable, and made sure underwriting was systematically talking to our ledger. The harder part: business underwriting has no reliable bureau the way consumer credit does, so the real signal lives in financial and bank statements. I designed a system that was friendly to all those sources and standardized the data into a shape our underwriting model could actually use, so we made the most informed decisions. Cut underwriting cost 27% and lifted approvals 10%.

Capital markets. I prototyped and ran the capital-markets operationalization: taking a lender contract, running our loan tape through it to generate a borrowing base, and keeping that loan tape accurate against our ledger. I also made sure the verification pipeline tied to borrowing-base generation fired autonomously, so the team could draw capital from our lenders without doing it by hand.

Banking integration. I led a 3-way integration with a partner's issuing bank, end to end: the bank issued the charge card while we held the underwriting, the capital, and the repayments. It opened up a new white-label charge-card product for partners.

Amptalk · PM, AI

Rebuilding the AI system. I identified that our prompt architecture needed to change and led the charge to fix it: moving from one big monolithic prompt to a dynamic, semantics-driven architecture, and working with engineering to figure out the evals we needed so changes could be measured instead of guessed at. I own model selection, the model P&L, and the workflows we want the agent to handle, and I keep the roadmap and vision aligned with customers by working closely with sales and engineering.

Repositioning around our data. Amptalk's flagship is amptalk analysis, a meeting-transcription tool that connects to Zoom and phone calls so sales teams can revisit transcripts and AI-generated minutes. I pushed the view that this conversation data is our real differentiator, and that the agent should learn from it and surface it inside its workflows rather than sit beside it. So I repositioned it from a narrow "email agent" into the broader amptalk agent, made the bet to have it live natively in the tools reps already use (email first, Slack next), and rebuilt the roadmap and messaging around that.

Getting traction without friction. Rather than pitch the agent as yet another separate tool, I led the push to get existing amptalk analysis customers onto it through free trials, so they could see it in action as a natural extension of what they were already paying for. That built traction without asking anyone to adopt something new from scratch.

Capital One · PM, Enterprise Data

Scaling access. The platform let consumers connect external accounts to Capital One, which meant wrestling with consent capture and data aggregation (continuous vs. one-time). I reworked the consent flow so it could plug into other Capital One systems like Shopping and accounts, which made it easy for any internal team to adopt. That took it from a few thousand to 100K+ new users a month.

Finding the real error rate. We were losing users at multiple steps with no visibility into why. I led an investigation into what was actually happening, then separated genuine platform errors from user error (like MFA) and third-party failures we needed to route to vendors. That brought the true platform error rate down to 0.7%.

Making it a true platform. I led the charge to turn the product into a true platform, in vision, capabilities, and the tech underneath. That meant rebuilding to our internal enterprise standards so it could scale cleanly to more enterprise teams, and drawing a hard line on what actually belonged in the platform (judging each ask by whether all customers would use it, and the revenue at risk if we passed) so we didn't overengineer. Lifted NPV $3M a year.