“See what you can do about scheduling.” That was about all John Sayer said when Morgan Walker joined Sayer’s small study group in DuPont’s Engineering Department near Christmas 1956. The department designed and built the company’s plants and oversaw more than $500 million a year of investment in them. Several of Walker’s friends told him nothing could be done about scheduling and that he had committed political suicide by taking the job.
DuPont asked Remington Rand, maker of the UNIVAC computer, for help, and it sent James Kelley. Their first decision, as they described it in 1959, was to pull apart two things planners usually did at once. Planning meant listing the jobs and the order they had to happen in. Scheduling, putting dates and costs on them, came afterward. Each job became an arrow on a diagram, and a UNIVAC computer worked out the longest chain of jobs through it.
Not everyone welcomed it. Walker recalled that the design division feared construction would use the method “as a club against them.” The chief engineer settled it by limiting the first live test to construction.
That test was a new plant at DuPont’s Repauno Works in New Jersey. Six engineers spent a good two months turning it into a network of 846 arrows, spread across their office wall. The regular schedulers had drawn up their own list of critical deliveries. The team fed in all 156 deliveries instead, and the computer found only seven that could hold up the project; three weren’t on the schedulers’ list. Only about one job in ten was critical. According to the computer, the plant could be finished two months sooner at no extra cost, and two months sooner again for about 1 percent more in variable direct costs. “The intuitive tendency,” Kelley and Walker wrote, “is to dismiss these results as ridiculous.”
A later test went to DuPont’s Louisville Works, which made a material for neoprene that could detonate on its own. Almost no maintenance was possible while it ran, so all of it waited for shutdowns, and every hour down was lost production. The common view was that a shutdown could hardly be planned, since nobody knew what a reactor needed until it was opened. The team found that most jobs came up every time. They built a library of typical shutdown schedules, usually fewer than a dozen, and absorbed surprises in the slack around the few critical jobs.
By March 1959, Kelley and Walker reported, the average shutdown had fallen from 125 hours to 93. Writing thirty years later, Walker said the extra production in the first year more than paid for developing the method.
Walker also recalled what he had been taught to do when a project slipped: beat it “to death” with manpower, across the board. “This across-the-board expenditure was commonplace and stupid,” he wrote. “Until Jim’s parametric model, we could never prove it.”