The record of a purchase is not the same as the work required to make it happen. In a discussion about AI in procurement, Lio co-founder and CEO Vladimir Keil describes an ERP line item as the visible end of a process that also involves meetings, emails, spreadsheets and negotiation.
Keil uses an illustration involving an aluminum purchase to show the gap. The recorded item may conceal stakeholder discussions, a supplier seeking a higher price and a cost engineer working through technical details. These are examples he uses to explain the workflow, not independently measured statistics about all procurement teams.
That gap is the opening a16z partner Sema Amble identifies for AI-native startups. In her account, incumbent software companies tend to add AI around the system of record they already own, while a new company can attempt to handle the job from beginning to end. The competitive claim is about the scope of work a product takes on, not merely whether it has an AI feature.
Keil describes Lio’s multi-agent approach as a response to the same problem: negotiation cannot be handled in isolation from contracts and other relevant information. Different agents need to share information and work in sequence to complete a purchase. This is his account of the company’s approach, rather than independent evidence that it outperforms competing systems.
The implementation burden remains important. Keil warns that getting an AI workflow close to working is different from making it production-ready, with integrations, memory and vertical data required for the final stretch. His argument therefore points to a larger opportunity than an AI layer on an ERP, but also a harder job than building a convincing demo.