P&G let suppliers bid bundles and discounts; optimization saved $295m on $3bn

By the early 2000s Procter & Gamble, a $57bn-revenue company, bought trucking, packaging, raw materials and services through electronic reverse auctions that had become procurement's standard: suppliers bid a single price per line item, lowest price won. That mechanic finds price concessions but is blind to synergies. A carrier with a natural backhaul can serve two lanes cheaply together — but only if it wins both, and the auction never lets it say so. Suppliers hated the format because every bid exposed them: quote a low price and hope volumes arrive.

P&G flipped the format with CombineNet, a Pittsburgh startup built on Tuomas Sandholm's combinatorial-optimization work. The trial came in 2001: P&G handed over a trucking-services instance a competing optimization product solved in 30 minutes; CombineNet's ASAP engine solved it optimally in 9 seconds. P&G then moved sourcing events to expressive bidding. Suppliers could submit package bids across lanes, conditional discounts ('15% off if you award at least 20 loads'), discount schedules, even off-specification offers.

P&G's own constraints and preferences were modeled as business rules in a Scenario Builder; the optimizer then assembled the cheapest award that satisfied them. In the Spring 2005 sourcing of all North American truckload — $885m, 7,161 lanes, 137 carriers — bids reached 160,797, including 720 package bids and 33 conditional discounts plus six discount schedules. P&G and its two largest customers, Walmart and Target, let carriers package lanes across all three shippers to build backhauls. The team walked through more than 550 scenarios before the final award, which carried 146,294 constraints.

The numbers held up where it matters. By March 2005, two and a half years in, P&G had sourced over $3bn through expressive commerce and banked $294.8m of implementable savings — 9.6%, down from 16.9% ($517m) unconstrained once P&G's business rules were applied. By category: services 29.8%, packaging 23.3%, indirect materials 23.9%, raw materials 10.9%, transportation 3.6%. The truckload event itself cleared at just 0.7% ($6m) — remarkable only because truckload prices that year rose 6–9%, so P&G beat the market by moving first on efficiency rather than price.

Suppliers bid bundles; optimization assembles the award

Expressive bids expose synergies a line-by-line auction cannot see: a carrier only reveals its cheap backhaul if it can win both lanes together.

Savings come from better matches rather than squeezed margins, so suppliers benefit too — which keeps them bidding instead of gaming the event.

The buyer's real constraints sit inside the optimization, so the reported savings were implementable, not theoretical.

It only worked because the clearing engine was fast enough: the 2001 trial solved optimally in 9 seconds where a competing product took 30 minutes.

Savings can come from matching, not squeezing: give bidders a language to reveal their efficiencies, then let optimization assemble the pieces under your real constraints.

Expressive competition scaled well past the pilot. Between December 2001 and December 2006 CombineNet hosted 447 highly combinatorial sourcing events worth $35bn for more than 60 Global-1000 buyers, with over 12,000 supplier companies bidding, and delivered $4.4bn of savings. The work earned AAAI's IAAI-06 Deployed Application Award, and the joint P&G–Walmart–Target truckload event became the canonical example of buyers pooling demand so carriers can harvest backhauls. Package bids, conditional discounts and optimization-driven awards are now standard features of strategic-sourcing suites.