Remember that famous scene from 2001: A Space Odyssey when the supercomputer HAL seizes control of the spacecraft, systematically murdering crew members and engaging in a malicious game of cat and mouse with the sole survivor? That same theme's been explored more recently in the Matrix movies, where "thinking" machines running "intelligent" software wield power over what's left of the world with bone-chilling results. Memorable as those images may be, they're hardly an accurate depiction of the state of intelligent software. In the warehouse environment, at least, the machines are still under the control of their human overseers, and visions of a fully automated, hyper-networked supply chain remain just that—a vision.
That's not to say software developers haven't made significant strides toward creating supply chain software that mimics human intelligence. Systems already exist that monitor conditions within a distribution facility or transportation network and report on any abnormalities, or "exceptions," encountered. Someday, they may be able to provide a list of recommendations for humans to act on ... or even take corrective actions on their own.
"It's a brave new world as far as technology is concerned," says Alison Smith, senior research analyst for AMR Research. "[M]ore and more intelligence is being put into devices. We are seeing more intelligent software being embedded into sensors and controls."
Right now, however, the day when thinking machines will be able to make supply chain decisions and reduce the human workload remains far off. At this point, "intelligence" is still largely limited to sensors and controls that monitor and report two key types of information: an item's location and its status. The advantages are obvious: With access to information on an item's location within a DC (and eventually anywhere in the supply chain), a manager has a good idea of whether the product can be expected to ship on time or will be delayed. Some companies are also using transportation management systems (TMS) that can issue status alerts to a computer, pager or cell phone when an order does not make the truck. Information on an order's status provides similar advantages. If a manager is alerted that some of the components in a shipment have failed to come together at a pack station or that there's not enough inventory in a pick face to complete the next wave of orders, he or she can take steps to solve a minor problem before it escalates into a full-blown and costly crisis.
"Intelligence will help us reduce those things in the supply chain that now have more expensive fixes," says Larry Lapide, research director at the Massachusetts Institute of Technology's Center for Transportation and Logistics. Most supply chain managers currently don't have enough information to act quickly, he explains. As a form of insurance, they build up buffer inventories. And when faced with delays, they have little choice but to throw money at the problem, scheduling employees to work overtime or air freighting a shipment at considerable added expense. With good intelligence, problems can be detected earlier, and cheaper fixes made.
This type of monitoring capability has already paid off for a lucky few. Procter & Gamble, for example, recently watched its on-time performance climb after installing a TMS from LeanLogistics that's now being rolled out across its enterprise. LeanLogistics says that before the pilot, P&G, which was looking to bolster its 94-percent on-time delivery rate, chose six "events" within its delivery process to monitor for possible corrective action: Did the carrier accept the assignment? Was the trailer available on time? Did loading begin on time? Did loading complete on time? Did the trailer leave the gate on time? Did the carrier report any delays en route?
In the end, Procter & Gamble discovered that about half the delays could be traced to internal problems and the other half to its carriers, and it used what it learned to fix the problems. In short order, the company, which had gone into the pilot hoping to increase its on-time performance by 1 percentage point, actually upped performance by 3 percentage points—to 97 percent.
Is data fact? But before software developers can get to the next level— that is, creating software that goes beyond simple monitoring—they face an enormous hurdle: gathering, sifting, correlating and analyzing mountains of data that eventually must be distributed to decision makers. As daunting as that task may sound, some experts believe programmers will receive a giant leg up from recent advances in visibility software and radio-frequency identification (RFID) technology.
RFID tags, in fact, have the potential to automate the entire data-gathering process. Even the simplest tags, the read-only models, can report on the status of products as they make their way through the supply chain—announcing to anyone with a reader when and where the item was manufactured, for example. The more sophisticated tags, those with read/write capabilities, allow users to update their information as they move through the chain, providing such valuable tracking data as where each item has been, who touched it, what value-added services have been performed and when each step in the process occurred.
Initially, the tags' information will be used inside the DC, processed through intelligent modules within warehouse and transportation management software suites. With those data, managers will be able to confirm at a glance that, say, replenishment tasks have been completed, orders picked properly, labor deployed where needed and orders shipped on time. Eventually, data from other parts of the supply chain can also be written to the tags, and then reported back to these software systems. This information will allow managers to determine the exact whereabouts of items in transit and even share the data with trading partners.
But that brings us to the next problem, what do you do with the flood of data that RFID can potentially provide? Work on that question is already under way. "Researchers are now studying ways to employ RFID," says Richard Pibernik, professor of supply chain management at the Massachusetts Institute of Technology-Zaragoza International Logistics Program in Spain. For example, Pibernik and his colleagues are looking at ways in which new technologies can provide real-time visibility into order fulfillment. This will give managers, suppliers and customers continuous access to status information throughout the order cycle. A customer who orders a plasma TV, for example, would automatically be advised at the time he places the order whether the item is in stock and if so, when he can expect it on his doorstep.
Still, even if RFID someday goes mainstream, there's no guarantee that the age of the thinking machine will follow close on its heels. The real problem has never been data gathering—Pibernik notes that the basic infrastructure for gathering location and status data already exists with bar codes. The true challenge is the analysis. "[W]e don't have the technology to process the data and filter the important information to make decisions," he says. "We lack the intelligent modules needed to extract and evaluate the data. Most companies are not ready to spend time and resources on it yet."
AMR's Smith adds that a logical next step is an integration of information gathered from sensors and controls into warehousing management and enterprise resource planning systems. But it won't happen tomorrow. "We are looking to 2008 before we see much integration with those systems," she says. "It's a very new market."
First of all, machines simply still have a lot to "learn." "You need a full history to ëpopulate' the learning. Not enough companies have this history yet," says MIT's Lapide.
But even when they've learned all they need to, the machines still must be programmed to respond in a certain way whenever they encounter a situation that can be tied to their history—much the way a so-called self-regulating thermostat is programmed to signal the furnace to kick in once it detects a drop in temperature. That very simple example of a self-regulating response, however, is a far cry from actual machine "thinking," which would require millions of bits of data to be analyzed and compared to its history before determining a precise resolution.
"Once self regulation is proved to work, then we can create adapting systems with learning capabilities, but that's a long way off," says Zaragoza's Pibernik. He says it would mean developing programs that would cover every conceivable situation that could arise in the supply chain.
And it's not at all clear that such an effort would pay off. "You would not get enough value out of the system to replace human intelligence," Pibernik says. There are other obstacles as well, he adds, citing a lack of industry standards, a dearth of corporate resources, and the absence of a clear picture as to what results logisticians want to achieve through intelligence.
For those reasons, most researchers expect breakthroughs in intelligent software to be limited to specific areas and functions. "We will have supply chains that are more automated," says MIT's Lapide. "Computers will [make] some of the routine decisions, but humans will still be handling the exceptions. The software can't know everything. It can support, but not replace."
"With enough time and money, all things are possible," adds AMR's Smith. "But I don't think there will be a financial incentive to have that much automation within the next 10 years."