System records capture the price; the real procurement battle lies outside the system
Behind the procurement price numbers are dozens of meetings, hundreds of emails, and thousands of suppliers—the advantage of AI-native companies is taking over this entire invisible work, not just recording the result.
The video won't play here. Listen to the audio instead:
The argument · tap a timestamp to hear it
The system records only prices; the actual work happens outside the system
Seema's argument: legacy software can only handle what's recorded in the system. When a customer complains about overcharges, the answer requires account statements, chat logs, and contracts—no single system holds all three. Procurement is even more extreme: "aluminum $8K" in an ERP is just the result. Behind it are 30 meetings, 500 emails, 20 Excel spreadsheets, and negotiations where suppliers originally demanded $10K. The opportunity for AI-native companies: take over this entire end-to-end work outside the system.
— Seema AmbleInvoices are 20 percent of the work; exceptions account for 80 percent
Leo started with invoice processing: extract information, cross-check across documents, push into SAP or Oracle—a clean workflow. This is the entire invoice software market today. But Vlad pointed out this is only 20% of the work; the other 80% is exceptions: "what if the invoice is forged?" or "what if the numbers don't reconcile?" Three years ago, Leo amazed customers with basic extraction. Leo quickly moved to exception handling, repeating the playbook of always being one step ahead of customer expectations.
— Vlad KyleHuman review layers scale trust from five figures to six figures
No enterprise adopts fully automated negotiation agents on day one—not a technology problem, but a trust problem. Leo's approach: have humans review every step, feed feedback back to the agent, train it on how a specific Fortune 10 customer actually operates. Trust builds incrementally: start with five-figure deals, move to six figures. Negotiations involving 3D models and technical drawings—worth millions—deliberately keep experts in the loop. The agent runs for hours, then stops to ask the cost engineer for input, then continues. Not one-shot full automation.
— Vlad KyleOne missed email in the inbox can cost hundreds of millions
Aircraft manufacturing and data-center construction require coordinating with thousands of suppliers. One component arriving two weeks late cascades into project delays costing hundreds of millions. But the system shows only a date: "not next Wednesday, the Wednesday after." The real signal is buried in a procurement manager's Outlook among 500 daily emails. Miss one, lose hundreds of millions. Leo's agents don't just surface the email; they judge whether it matters, how much, and how to fix it.
— Vlad KyleOne supplier contract alone can represent a billion dollars in value
Direct procurement isn't notebooks and pencils (50K suppliers worldwide) but 100 to 2,000 strategic suppliers, where one contract can be worth ten billion. These negotiations don't run on automation: three months, a 10-person team. Cost engineers dismantle drawings, track aluminum and oil prices, to negotiate ten billion down to nine billion. Vlad imagined a real-time version: during talks, the agent alerts—oil prices up 10%, but this part is only 30% oil, so the price rise should be 4%, not the 10% the supplier is asking.
— Vlad KyleMoats don't come from executing a predetermined six-step roadmap in order
Seema argues that moats can't be designed into the product roadmap upfront—"execute steps 1-6, moat appears at step 7" doesn't work. In the best companies, what you see early is simply winning customer trust and selling more. The real signal: dependence. Old CRM is just a ledger of completed sales. New sales AI agents do the actual work—prospect research, outbound contact. Customers depend not on "recording transactions" but on "getting work done." That dependence is the moat.
— Seema AmbleBuilding in-house to 70 percent doesn't mean 70 percent automation
Three years ago, Leo's first product—importing quote data into SAP—is now a hiring challenge; candidates solve it in 8 hours. Proof this technology is easy to replicate. So why doesn't every enterprise build it in two months? Because 70% feature completeness isn't 70% automation: someone still has to verify all 100% of the results, often harder than before. Seema's Fortune 500 example: they spent three months internally on a cash-recovery tool; recordings and screenshots piled up but context quality was poor; they acquired another company and inherited more ERP mappings to maintain; they abandoned the project.
When both sides deploy agents, negotiations actually close faster
You'd expect suppliers to adopt agents first, but the opposite is true: suppliers lead in recording tools but lag in agent deployment. Large industrial buyers already dictate supplier systems and RFQ response formats. Leo gets to deploy agents to both sides simultaneously, owning both ends of the transaction. Price looks like a zero-sum game, but Vlad says it's the outcome of 5,000 other tasks—both sides want faster delivery, less friction. Deploy agents to both sides, and negotiations close faster.
— Vlad KyleIn their own words · checked verbatim
Someone sends a confirmation of like, hey, sorry, like this part is going to arrive two weeks later. And if they miss this email, hundreds of millions of them.
Vlad Kyle0:00
No company and no enterprise starts with fully autonomous negotiation agents from day one. Why? Because they don't trust us and they don't trust the technology from day one.
Vlad Kyle0:00
the opportunity for the AI native startup is to say, we're going to own that entire end-to-end arc.
Seema Amble4:16
if there's like one specific part which arrives two weeks later, this can have like a damage of like hundreds of millions of dollars and postpone and postpone the project.
Vlad Kyle22:41
it's really, really hard to forecast your moat going forward. If you look back at all the best businesses at the, at the early stages, they were, they were just thinking about, okay, I'm winning customer trust.
Seema Amble38:03
you will only reach 70 percent, let's say, like the of the performance. And the problem is 70 percent of performance or accuracy or however you measured it, it depends really on the task, doesn't mean 70 percent automation.
Vlad Kyle41:17
So to get like 1% margin increase, um, you need to make 10% more revenue, 10% more sales.
Vlad Kyle57:43
Figures
| Coordination behind a single procurement negotiation | 30 stakeholder meetings, 500 emails, 20 spreadsheets | 5:17 |
| Trust-building negotiation scale progression | From 10,000-yuan-level to 100,000-yuan-level negotiations | 16:24 |
| Potential project loss from a two-week component delay | Hundreds of millions of dollars | 22:41 |
| Emails in a procurement manager's inbox daily | Approximately 500 | 22:41 |
| Number of suppliers in direct procurement | 100 to 2,000 | 32:01 |
| Maximum procurement spend from a single supplier (example) | Approximately 10 billion dollars | 32:01 |
| Results and investment of a major negotiation | Negotiated from 10 billion to 9 billion dollars; invested 3 months and 10 full-time staff | 33:01 |
| Performance ceiling of in-house-built systems | Approximately 70 percent (not equivalent to 70 percent automation) | 41:17 |
| Percentage of engineers at Leo | 85 percent | 50:34 |
Glossary
- system of record
- The authoritative central system storing business data for one function, such as an ERP or CRM
- RFQ
- Request for Quote: a formal request from a buyer to a supplier asking for pricing on specified goods or services
- should-cost modeling
- Estimating the fair cost of a component from its drawings and manufacturing process, used in price negotiations
- direct procurement
- Procurement of components directly used in manufacturing; fewer suppliers but individually high-value contracts
- principal agent
- The top tier in a four-level agent framework with authority to make strategic business decisions like approving additional spending
How to listen
Founders concerned with how AI-native companies compete against software giants, enterprise software product managers, and investors watching supply-chain and procurement digitalization.
The segment from 14:22-15:23 consists of garbled transcription with repeated filler words and can be skipped.