by Richard Pastore

Con-Way Cruising With Expert System

Feb 01, 200315 mins
Supply Chain Management Software

Ann Arbor, Mich., 5:09 p.m., a typical Tuesday afternoon in late 1996.

Marty Robinson has been scrambling since his shift started 99 minutes ago. As Con-Way Central Express’s line-haul dispatcher, Robinson has to gather and sort through a foot-and-a-half stack of mainframe printouts showing what freight needs to move between 200 locations across 25 states and Canada during the next 12 hours. Robinson reviews tonnage, available drivers and trucks, and trailer capacity at each of the 40 locations assigned to him. Like the other seven dispatchers sitting in a brightly lit tray of cubicles, he must plot the nightly routes that will transport 50,000 shipments of heavy freight on 2,100 trucks from their origin points to midway freight-assembly centers. (A midnight shift will route the freight from the assembly centers to its final destination.)

Con-Way’s model for line-haul?the movement of freight over long distances from service center to service center for redistribution?forces Robinson into a perpetual tug-of-war between the promise of on-time overnight delivery (a nonnegotiable customer expectation) and efficiency. For efficiency’s sake, he must build routes with the fewest miles, maximize trailer loads and make sure drivers, who are nonunion, can get home when their daily shift ends. The clock ticks insistently on the whitewashed wall that’s covered by maps of the Midwest. Robinson has 51 minutes left to finish the route plan and start typing his instructions to individual freight depots and their waiting drivers. But there’s a snag. He’s got two more freight units to ship from Indianapolis to Louisville, Ky., and he’s out of trailer space.

Robinson can reroute a truck from Terre Haute, Ind., to Indianapolis to grab the two units, but the truck would have to return empty to Terre Haute to meet its incoming load. That’s 130 nonrevenue miles.

He shouts to his neighbor John, visible over the low cubicle wall. “Have you got anyone in your area who can divert to Indianapolis for two units?”

No, John’s already short of drivers for Cincinnati. Robinson could keep checking, but he can’t afford the time. He eats the empty miles and shifts his attention North, to Chicago.

Actually, Con-Way’s line-haul dispatch job is even more complicated than the image drawn so far. For example, Robinson’s working with only 85 percent of the day’s freight orders. The rest come seeping in while he’s wrangling his route plan. And an ice storm he hasn’t yet heard about may shut down one of his cities, throwing another monkey wrench into his plans.

This complexity is the primary reason Enterprise Value Award winner Con-Way Transportation Services, which operates Con-Way Central, Western, Southern and Canada Express, decided back in 1996 that it needed to automate Robinson’s job. The result of this decision, coming after two years of painful development and another two years of deliberately slow deployment, is the line-haul automation system. Each afternoon, this expert system dutifully digests on average 80,000 customer pickup orders and change requests. Its optimization model plots delivery across all the service centers and determines which shipments should be loaded on which truck?all this in seven minutes flat. The incremental efficiencies generated have repaid the $3 million investment within the first two years.

Even without that kind of hard-dollar return, Con-Way would have funded the project simply because it could not sustain the status quo. The stressful dispatch position was difficult to fill and had a learning curve as long as 18 months. The procedures and business rules the dispatchers followed were undocumented, making the company uncomfortably dependent on the knowledge in the dispatchers’ heads. A line-haul automation system with an optimization model at its heart would get the job done faster and more consistently, letting dispatchers use their time to troubleshoot problems that could jeopardize on-time delivery. And Con-Way would be able to extend its cutoff time for overnight shipment requests, letting customers submit orders, when necessary, right up until the end of the business day?a competitive advantage, since businesses would rather not have to arrange their activities around shippers’ schedules.

Adapting a System, Not a Business Model

In 1996, Con-Way Central’s then Line-haul Director John Labrie contacted a University of Michigan academic consulting group to explore the possibilities of an expert system. While FedEx and other companies have had route optimization models for overnight flights, the kind of technology that applied to less-than-full-truckload line-haul was still largely the stuff of university labs. Con-Way quickly discovered that the university group and other outside developers would require the company to modify its fairly unusual business model to suit the system. Con-Way plans all delivery routes from the origin site of the shipment, rather than using a hub-and-spoke model, and all its drivers go home at the end of their daily shift. “The alternatives all required us to change, and we knew our model worked very well and was critical to our competitiveness,” says Douglas W. Stotlar, Con-Way Transportation’s executive vice president and COO.

The practices have helped make Con-Way the perennial earnings leader in its niche of less-than-full-truckload, time- and day-definite freight delivery. So the company made the risky decision to dedicate on-staff industrial engineer (and former Chinese Army rocket scientist) YaFeng Du to develop Con-Way’s own expert system, starting with its biggest and most complex group, Central Express.

“I figured it would take me six months, but I told them it would be a year in order to make more room,” Du says with a sly grin. It ultimately took five years and $3 million to develop, test and deploy at Con-Way Central in late 2000. The delay came from the excruciating challenge of extracting the business rules and logic from the minds of the dispatchers, coding them into the expert system, and getting it to work. Tests produced routes that were far from optimized?a truck would be sent out of its way, or a driver would be stranded too far from home to get back during his shift. The interrelationships of all the variables were so complex that a fix to one flaw would cause two new problems. Because faulty system output would often take as long as a month to surface during testing, each change necessitated a stressful, fingers-crossed waiting period.

User Struggles

It’s been two years since Con-Way deployed its system at Central Express. But Du, a fidgety man with a ready smile, still seems anxious to get back to his computer to work on changes for further implementations. Du says his biggest development mistake was that after talking to just one dispatcher, he rushed a rudimentary system into place after three months as a proof of concept. “I should have spent more time with the dispatchers and should have done some dispatching myself,” says Du, now director of decision technology. After the letdown of the demo, Du struggled to accommodate the various styles and problem-solving preferences of all the dispatchers. They peppered him to add so many rules that the system and its developer were overwhelmed. “We had thousands of rules after a while,” says Du. “It was very frustrating, and I came close to giving up, thinking we’d never get it to work.” Lines of C++ code mushroomed to 250,000 (100,000 lines are now active). In the end, Du’s boss, Stotlar, who looks too young at 41 to be COO of a $2 billion company, told the dispatchers enough is enough. “I had to lean on them a bit; they were throwing everything in but the kitchen sink,” Stotlar recalls.

But Du’s time spent with the dispatchers was not wasted. This group could easily have felt threatened by the project and sabotaged it. Instead, their sweat spent to help build the new system bought equity in it. “This had to be acceptable to the guys here. All you’ll get is alienation if you stuff something down someone’s throat,” says Michael Kucinski, who took over as director of line-haul and freight flow at Con-Way Central when Labrie was promoted to run Con-Way Western. Kucinski acted as a rudder “to keep attitudes straight” and remind his dispatchers that “there was a light at the end of the tunnel.” And with time, Kucinski says, they really learned how much they appreciated line-haul automation after computer crashes forced them to go back to manual routing on three occasions.

“It saves us a lot of time each night, which we use to make sure we’ve got accurate order information, look into problems and handle changes,” says Robinson, who helped test the system and is now line-haul supervisor at Con-Way Central. The system captures customers’ shipment pickup requests throughout the day, including orders coming as late as 5:15 p.m. It flags suspect data for the dispatchers to verify or correct, such as freight dimensions and weight specs that don’t seem to correlate. It then works its optimization magic, taking just seven minutes to generate a routing plan for 95 percent of the day’s overnight shipments. Dispatchers, now trained as analysts on a managerial track, massage the plan via a Web interface, solving any remaining driver shortages and tweaking routes for exceptional circumstances.

The Payoff

The line-haul automation system routinely manages to extract efficiency improvements of 1 percent to 3 percent over results achieved with manual route planning. For example, comparing two single nights moving similar tonnage, Con-Way Central used 111 fewer trucks and 68 fewer drivers, drove 26,530 fewer miles, and boosted the average truck’s load by 370 pounds. Stotlar says he initially guessed the differential would have been higher, but Kucinski and Robinson, not surprisingly, expected only this modest improvement. But all would agree, given Con-Way’s volumes, that the incremental gains would tip any weigh station scale. They add up to between $4 million and $5 million in savings annually from paying fewer drivers, moving trucks fewer miles, packing more freight per trailer and reducing damage from rehandling freight. Con-Way Central has been able to reduce its dispatch personnel by three people (through attrition) and can keep the group small as it adds business.

Customers, in the meantime, benefit in a couple of discreet ways. One is improvement in on-time delivery. While Con-Way is reluctant to attribute its 99 percent on-time performance directly to the line-haul automation system, there’s no doubt freeing dispatchers from the tedium of routing has given them more time to prevent delays.

Another customer benefit is the later cutoff time for submitting orders. “It’s a huge advantage over the competition, and it’s also a great psychological advantage because it’s the end of the business day,” says Kathleen F. Curley, research professor in Information Systems at Boston University’s School of Management and a member of the Enterprise Value Awards application review board. Con-Way doesn’t market the extended time to its customers; it wants to avoid being overwhelmed by last minute calls. But when an urgent situation comes up for a customer, “you have until the end of the day to call Con-Way and say, I really need this shipment to be picked up,” Curley says.

According to Enterprise Value Awards judge John Glaser, in businesses like Con-Way’s, there are only a small number of variables to maximize for strategic value. “Truckload fill rate is the name of the game for them, and they improved it,” says Glaser, vice president and CIO of Partners HealthCare System in Boston.

When it came to bringing the new line-haul system to other regional operating groups, Con-Way also took its time. The system, undergoing constant tweaks and upgrades, ran in parallel to the manual way of doing things for a year at Con-Way Central. Then for more than a year, the live system was monitored to see how consistently it handled Central’s workload. The deliberate pace and choice to implement first at Con-Way’s biggest operation bulletproofed and validated the system. “Nothing is more critical to this company than line-haul; no system failure could bring us to our knees more quickly than line-haul,” Stotlar says.

Even perception of failure was a threat to the field staff’s fragile faith in the project. “We’d laid the groundwork here and in the field that this was the wave of the future,” Kucinski says. “If the system experienced any prolonged failure at any point, that would have been shot. The perception would have been that we just can’t automate this.”

Mending an IT Disconnect

Despite the cautious pace and sagacious involvement of the users, Con-Way Transportation committed a classic business-IT alignment faux pas. Du did the technical design and coding of the optimization model on his own, then, in his words, “I gave it to the IT department and said, ’This is it. You have to figure out everything else.’”

“Everything else” meant integrating Du’s optimization model with the rest of the line-haul automation components, including an existing load-plan application and Oracle database, plus developing new interfaces, providing maintenance and leading subsequent rollouts. The handoff to the 150- person, Portland, Ore.-based IT group was far from the ideal for Jacquelyn Barretta, Con-Way Transportation’s vice president of information services. “We had a lot of concern around that. YaFeng is very talented, and not many people have the same abilities he does. So we would have preferred to start working on this earlier to make sure we understood the model,” Barretta says. She notes that there’s still a lack of documentation, and the model didn’t adhere to IT’s coding standards, making it tougher to support. There’s also the risk that a change to the applications that capture and transmit shipment orders, among others, could have an unanticipated impact on Du’s model. All the more reason to gain a full understanding of its nature and code.

Toward this end, Barretta recently dedicated a staff member to become an expert in the optimization model, and sent the person to classes to gain a better understanding of the disciplines behind it and what makes it tick. In the meantime, IT staffers still see the optimization model as a hands-off “black box” and call on Du to diagnose and fix problems via remote access to their servers.

In spite of those concerns, the four IT staff members dedicated to deploying and maintaining the line-haul automation system have forged ahead. In March 2002, Line-haul Automation Project Coordinator Scott Van Winkle’s team deployed a Web interface that for the first time lets dispatchers view the line-haul plan onscreen rather than in stacks of printouts. And they’ve been spearheading the line-haul automation’s rollout in Con-Way Western Express, due for completion in December 2002, using a dispatcher from Central as an ambassador to smooth user acceptance. And this time there is no grueling extraction of business rules. The Western dispatchers were given parameters and a business logic table, and they were told to fill in the blanks.

Still, because the Western region’s destinations are more spread out than Central’s, the program needed some tweaks. Again, there was pressure from dispatchers to make changes to suit their individual preferences. Even Central was still trying to slip in more modifications. Barretta quickly learned to subject change requests to careful, structured review, requiring documentation of the request and review, and prioritization of the proposed change by line-haul managers in all three regions. Overall, ongoing maintenance on the system should average $100,000 per year.

Another IT responsibility is making sure the mission-critical application never fails. MQSeries performance monitors scan for certain threshold numbers that could indicate trouble in the order pipeline. When a threshold is reached, IT staff pagers start clamoring. A crash in the IBM WebSphere server that powers the system will cause an automatic failover to the server used for development. About 105,000 shipment orders move through the system each day, with a spike in the late afternoon. The architecture, based on J2EE, is built to accommodate volume spikes and overall business growth.

Plans call for building optimization bridges across the company. Once line-haul automation is running in all four Con-Way operating regions, each will have visibility into the others, making it possible to, say, take an available truck and driver from Southern and press them into service for Central. The coding is already complete for this capability and other enhancements, such as adding weather conditions into the routing plan, and simulating what happens if Con-Way adds new terminals, for example.

Faith in Long-Term Strategy

Looking back over seven years to the inception of the line-haul project, Du sees the trust that Stotlar and Con-Way CEO Gerald Detter had in the project as its most critical success factor. “At certain times, things didn’t look that rosy, but they never lost confidence,” Du says. “They told me it didn’t matter how slow it was, just keep working on it. So many people put their trust in me; that was my motive not to fail.”

It helped that enough incremental progress was being made to indicate a probability of success, adds Stotlar. But another reason for Con-Way’s faith was that it wasn’t letting short-term ROI rule the day. In fact, Stotlar doesn’t hesitate to answer “absolutely” when asked if Con-Way would green-light such a protracted initiative in today’s lean times, when many companies won’t invest in anything that doesn’t promise a six-month payoff. “We understood the strategic value of this, we were comfortable with that, and we didn’t get blinded by ROI,” he says. “If you manage purely from an ROI perspective, you might have given up on this one early on. And that would have been a mistake.”