Week 9 of Price Theory III covers VCG. Here’s an interesting application of VCG:
How mechanism design theory helps optimize Amazon-vendor collaboration
Dirk Bergemann, May 5, 2026
Amazon buys products from vendors (like P&G). Today, Amazon dictates the order: “Send 5,000 cases of Tide to Phoenix by Tuesday.” The vendor can only choose how much to ship — not where or when.
This is inefficient because each side has private info the other can’t see:
Amazon knows: customer demand, outbound shipping costs
Vendor knows: their factory locations, trucking routes, production costs
For example, P&G already has a truck running Cincinnati → Dallas every Tuesday, half-empty. Adding 5,000 cases to Dallas costs them almost nothing. Shipping to Phoenix means a dedicated truck across the desert — expensive.
But Amazon asked for Phoenix. P&G’s options today:
- Eat the cost and ship to Phoenix
- Short the order
- Refuse
None find the actually-cheapest answer: ship to Dallas, Amazon trucks it to Phoenix customers.
The fix is Flo Pro a back-and-forth system where:
(1) Both sides’ software agents negotiate without revealing actual costs
(2)They land on the plan that’s cheapest overall
(3) Whoever benefits pays the other a fair compensation for any inconvenience
VCG guarantees truthfulness. ADMM (alternating-direction method of multipliers) guarantees the progress only requires query on cost instead of vendor reporting their “cost function” — so privacy and applicable:
“CPP is a distributed optimization protocol based on the alternating-direction method of multipliers (ADMM). Rather than asking agents to reveal everything at once, CPP works iteratively. A central coordinator proposes a consensus plan and a set of prices. Each agent responds with its preferred plan given those prices — a ‘best response’ that requires solving only the agent’s own local optimization problem. The coordinator then updates the proposal, and the process repeats until convergence.”
I think this is very impressive. It feels like a computational tatonnement. The prices that ADMM iterates on are the Lagrange multipliers (dual variables) on the constraint “Amazon’s plan must equal vendor’s plan.” They function exactly like Walras’s market-clearing prices: they’re the signals that push two independently-optimizing agents toward agreement on a price that is incentive compatible.
The result is: tested for 9 weeks with a real manufacturer. Real cost savings. Could extend to FBA sellers, logistics partners, anywhere two parties need to coordinate but keep secrets.