Rob Belcher · Senior / Principal Product Manager · Austin, TX
I handed them twenty pages on how their system should work. They said yes.
I was a punk kid in the PeopleScout call center, stuck in a DOS app I thought could work so much better. So I drew up how it should work instead, walked up, and handed it over. Over 20 years later I’m doing the same thing — see the dumb system, see the better one, go build it. Today that’s a real-time AI agent layer on live PayPal global support calls, and several AI apps I built and run myself.
What I’m building now
At PayPal I’m spearheading the real-time AI agent layer on live contacts for a 9,000-person contact center serving 400M+ customers. The summarization agent is live and cuts about 20 seconds off every transferred contact; coaching and risk-intervention agents are in pilot. I write the prompts, define each agent’s KPIs, and run the evals against contact history.
At home I build and run my own AI apps in daily use — for meals, documents, mail and jobs, with an iPhone front end over them — on shared sign-in, email and CI services I built underneath. I’m a product manager first; the engineering is the skill I’m building, with Claude as the muscle.
What have I built with teams?
FinTech, Health Care, and Staffing/Recruiting. I’ve built some pretty cool tools for other people too. Same thesis: what’s the best use of their time, and how can systems take over the rest?
I built PeopleScout’s applicant tracking system, Windows first and then the web; and at its sister company, Staff Management, an HRIS platform that brought standalone field offices together into a true connected network.
I build the shared layer, so the special cases stop needing new code. At Staff Management I turned an old DOS system into a networked platform where workforce operations became verbs — actions you could run against a list of employees assembled any number of ways — so a new operational need stopped meaning new code. At StudentScout I built the BI and data-warehouse platform the whole business read its numbers through; deals closed on that alone. At athenahealth I moved a use-case-specific codebase toward modular APIs and reusable widgets, so a new practice assembled from parts instead of custom work. At PayPal I contributed the workflow primitives that let new service journeys snap together from steps that already exist. Same instinct that put shared sign-in and email under my own apps.
When we know what to change, I make “live” the next step, not next quarter. At Vision Web I compressed maintenance cycles from eight weeks to one and automated a deploy process that had been eating a full person’s job. At PayPal I took a routing change from a sprint-and-a-pull-request down to an edit in a dashboard, and modeled campaigns as configuration instead of code — so a new campaign type ships with no engineering at all.
I put the memory in the system, not in someone’s head. This is the whole site’s thesis, and it’s been my job for years. The earliest proof: at Staff Management we packaged the routine back-office work and shipped it to a central National Crew Desk, so local offices could be on the floor with customers instead of chained to a back-room desktop. At PayPal I replaced an alerting system that fired once at login and left teammates to personally remember open issues and live fraud vectors for weeks — the new one surfaces what matters from upstream intent and the workflow they’re actually in — and I’m bringing live call transcription and analysis into the orchestrator so nobody has to hold the whole call in their head. At athenahealth I collapsed a practice’s hundred-plus appointment types down to twelve, so booking stopped needing an expert who remembered how. My apps are the same move, pointed at my own life.
At StudentScout I found that speed to first attempt didn’t just speed up contacts, it permanently lifted the whole contact-rate curve. So I built the call-scheduling engine around that — wrote production dialer list management code myself — and took time-to-first-attempt from over five minutes to twenty-eight seconds, contact rates from around 60% to over 80%, no extra staff.
The range, briefly: nurse staffing from a 60-nurse pilot to 10,000 across seven states during the crisis; payments service ops at 9,000 teammates and 400M+ accounts; healthcare EMR; and contact-center telephony.