LLMs Power Ramp's Vendor Graph
Geoff Charles, VP of Product at Ramp, on Ramp's AI flywheel
Switching from a custom model to an LLM shows that Ramp’s edge is no longer building narrow classifiers, it is owning the finance data and workflow where generic models can be applied again and again. Once Ramp can reliably tell that a messy Square charge is really Blue Bottle, or that an invoice vendor matches a card merchant, it can turn raw transactions into a unified vendor graph that powers expense controls, bill pay, contract analysis, and negotiation in one product surface.
-
The hard part is not just reading receipts, it is linking the same vendor across cards, invoices, contracts, and email. Ramp describes vendor unification as connecting AP payments and card spend so finance teams can see that the same supplier is being paid through multiple channels, then push more of that spend onto Ramp-managed workflows.
-
This is why LLMs beat the old in-house approach. Classical models worked best when the answer set was already structured, but finance data is full of merchant aliases, line items, renewal clauses, and mismatched documents. Ramp found generic LLMs could reach near vendor specific OCR accuracy at far lower cost, which made more use cases economically viable.
-
The competitive implication is that AI compresses model-building advantage and shifts value to distribution and proprietary spend data. Ramp’s broader goal is to use cards as the entry point, then use unified transaction and contract data to move into bill pay, vendor management, procurement, and enterprise expense software, alongside rivals like Brex and incumbents like SAP Concur.
From here, the winning finance platform is likely the one that turns fragmented payment records into an action layer. As Ramp scales past cards, better vendor resolution should let it automatically flag duplicate software spend, surface renewal deadlines, and draft negotiation emails, making each added workflow strengthen the next one instead of living as a separate product.