When it comes to IT projects, it doesn’t get much better than this: Procter & Gamble Co. saves US$300 million annually on an investment of less than one per cent of that amount.
Indeed, P&G’s use of agent-based modeling helped it transform its supply chain system so fundamentally that the company no longer even calls it a supply chain. The Cincinnati-based maker of Tide, Crest, Pringles, Pampers, Clairol and 300 other products now calls its connections to five billion consumers in 140 countries a “supply network.”
“Chain connotes something that is sequential, that requires handing off information in sequence,” says Larry Kellam, P&G’s director of supply network innovation. “We believe it has to operate like a network, like an Internet, so everybody has visibility to the information.”
Many of the insights that have enabled P&G to transform a chain into a network come from agent-based computer models it developed with BiosGroup Inc. in Santa Fe, N.M. Their work is a real-world example of what mathematicians call “agent-based modeling of complex, adaptive systems,” a discipline pioneered by BiosGroup and other mostly Santa Fe-area companies, laboratories and think tanks.
The idea is that many systems that are enormously complex overall are in fact made up of semiautonomous local “agents” acting on a few simple rules.
For example, in a supply chain system, a rule in a warehouse might be, “Fill orders on a first-in, first-out basis,” or “Don’t send this truck out on delivery until it is full.” Dozens or hundreds of these local “agents” — truck dispatchers, say — acting autonomously produce complex behaviour by the system as a whole. It’s possible to simulate this complex behaviour by programming software agents with a few rules and letting them interact with one another. By optimizing the agents’ activities at a local level, it’s possible to improve the performance of the system as a whole.
Agent-based modeling, while not yet commonplace, is catching on, especially at companies with large, complex supply or transportation networks. In addition to P&G, the following companies have tried it and cite benefits that include cost savings, reduced inventories and better customer service:
– Southwest Airlines Co. used software agents to optimize cargo routing.
– Air Liquide America LP, a Houston-based producer of liquefied industrial gases, reduced both production and distribution costs with agent-based modeling.
– Merck & Co. used agents to help it find more efficient ways to distribute anti-HIV drugs in Zimbabwe.
– Ford Motor Co. used agents to simulate buyer preferences, suggesting packages of car options that optimized the trade-offs between production costs and customer demands.- Edison Chouest Offshore LLC, an offshore service company in Galliano, La., used agents to optimize its deployment of service and supply vessels in the Gulf of Mexico.
Computer simulations
In P&G’s computer simulations, software agents represent the individual components of the supply system, such as trucks, drivers, stores and so on. The behaviour of each agent is programmed via rules that mimic actual behaviour, such as, “Dispatch this truck only when it is full” or “Make more shampoo when inventory falls to x days’ demand.”
The simulations let P&G perform what-if analyses to test the impact of new logistics rules on three key metrics: inventory levels, transportation costs and in-store stock-outs. The models considered alternate rules on ordering and shipping frequencies, distribution centre product allocation policies, demand forecasting and so on.
“Some of the conclusions were surprising, and some confirmed what we believed but didn’t have the data to support,” Kellam says.
For example, he says, the models showed that it would often be advantageous to send out trucks with less than full loads, something P&G almost never did before. Although transportation costs would be higher as a result, P&G could more than make that up by reducing the frequency of in-store stock-outs, which often result in lost sales.
“Agent-based modeling convinced us of some changes we fundamentally had to make if we were to be flexible and adaptable,” Kellam says.
P&G uses supply chain management software from SAP AG, but it turned to a tiny New Mexico company when its long efforts to decrease inventory levels produced only marginal improvements. “We went to BiosGroup because they think very differently from the way we do,” Kellam explains. “But most of the supply chain experts we went to thought very similarly to the way we do.”
Computer modeling of supply chain operations, like that done by BiosGroup and P&G, today requires a combination of custom software development and consulting. But that could change as a result of a development agreement that P&G fostered between SAP and BiosGroup.
Predicting stock-outs
SAP has already demonstrated a prototype agent capability in its replenishment software. Agents predict the probability of stock-outs – based on current inventory, scheduled receipts and expected demand – and when that probability exceeds a certain threshold, a replenishment signal is triggered, according to Christian Knoll, vice president of global supply chain management at SAP.
SAP may introduce the prototype technology in its products, Knoll says, but for now it’s helping a few key customers try it out on a project basis.
Navi Radju, an analyst at Forrester Research Inc., says the supply network that P&G operates is just the sort of environment that lends itself readily to agent-based modeling.
“It is exposed to a high degree of variability, involves multiple partners and requires a high degree of coordination and collaboration,” he says. “When the whole process is not owned by a company, you need a bottom-up approach to controlling, managing and optimizing the integrated process.”
Radju predicts that such bottom-up, agent-based optimization will increase in popularity – slowly. “P&G is a very forward-looking company, one willing to try new technology and learn from it,” he says. “But the mainstream companies say, ‘Let’s not invest in unproven technologies.'”
But Radju says big software companies – especially SAP and IBM Corp. – over the next two years will roll out agent-based supply chain optimization packages. Then, he says, the technology will shed its image as the intellectual domain of Ph.D. mathematicians.
Meanwhile, P&G says that by 2008, software agents will enable another leap forward in supply network management. While agents have so far been used just for modeling, they will increasingly be deployed in P&G’s operational software, Kellam says.
P&G made three broad changes
Relaxation of rigid rules, often counter-intuitively, in order to improve the overall performance of the supply network. That required some cultural changes, such as convincing freight managers that it’s sometimes OK to let a truck go half full.
More flexibility in manufacturing. As a result of insights gained by the models, P&G is “fundamentally re-tooling” its manufacturing processes so that it no longer produces long runs of a single product but instead is able to produce every product every day. The benefits include fewer stock-outs and happier customers.
More flexibility in distribution. For example, it’s possible to restock a retailer in 24 hours rather than the customary 48 to 72 hours.
