Turning GPT-4 Into Product Engine
Geoff Charles, VP of Product at Ramp, on Ramp's AI flywheel
The real edge is not access to GPT-4, it is turning a generic model into a faster product engine and a leaner operating system. Ramp used LLMs as a general parser for messy finance data, receipts, invoices, contracts, merchant names, and support conversations, which let it ship features that previously needed custom models, OCR vendors, or manual review. That reduced product build time, cut low level operations work, and helped Ramp stay much smaller than Brex and Rippling while expanding across more finance workflows.
-
Inside the product, GPT-4 replaced narrow point solutions for turning documents into usable fields. Ramp uses AI to pull renewal dates from contracts, detect minibar charges hidden inside hotel receipts, unify vendor names across card and AP data, and answer support questions. That is generic model capability wrapped in Ramp specific workflows and proprietary spend data.
-
Inside the company, AI let Ramp keep headcount low by automating support deflection, coding assistance, and customer research. Geoff Charles described bots that summarize sales calls and surface why prospects churn or choose competitors. The goal was to keep coordination overhead low and give each employee broader scope, not to match rivals function for function with a larger team.
-
That operating model matters competitively because Ramp was scaling revenue with far fewer people. Ramp reached an estimated $1B annualized revenue by August 2025 with about 500 employees cited in the interview era, versus Brex at $700M annualized revenue with roughly 1,000 employees and Rippling at $570M annualized revenue with roughly 2,000 employees. AI helped Ramp push further into bill pay, procurement, and vendor management without hiring proportionally.
The next step is that finance software stops being a place where people type in data and becomes a system that prepares decisions, flags exceptions, and executes routine work. If Ramp keeps combining generic models with its own transaction, contract, and vendor data, it can keep widening the gap between a lean product led finance platform and competitors that still need more people to deliver the same operational output.