"We don't ask what we can take out of the budget," says Glenn Hodge, the strategic initiative manager for Corporate Services (CS). "Rather, we look at what is necessary to satisfy the customer, and then ask how to do that more efficiently and more cost effectively."

This approach can be applied just as well to an office building or a research facility as to an ultra-high-tech chip factory, affirms Hodge. However, it does entail delving into intricate detail about what drives spending in the particular facility type.

For Intel, the effort is worth it. Proof that the competitive services model is a wining formula is evident in the $30 million year-over-year savings realized as a result of applying the new paradigm, all while maintaining current service levels.

A Universal Metric

To set up a true bottom-up, zero-based budgeting process, Hodge and colleague Kris Goranson, CS senior strategic finance analyst, assembled a subteam of content experts from within Corporate Services' Cost Ops Labor and Productivity Team. The subteam drew on extensive corporate cost data compiled by CS and the manufacturing organization to create a model that could be applied universally to worldwide manufacturing facilities.

The cornerstone of the model is a factory of a predetermined size and output, along with appropriate support buildings. The key component is a hypothetical fab that encompasses 150,000 sf of cleanroom and X-number of wafer starts per week.

"Then we asked: What are the systems and infrastructure driving that business—in other words, where are our headcount resources consumed? What kind of organization and dollars are necessary to support that?" explains Goranson. "The consistent services that we provide for those facilities worldwide constitute our baseline model."

This step departs from the typical top-down tactic of allowing existing headcount to dictate how much money is necessary to run the operation. Instead, Intel determines optimal performance and then figures out the staffing requirements to achieve it.

"The tenet of the model is to focus on the bottom-up approach, accounting for the space and systems that we manage and the services we provide, versus 'this is our headcount and therefore this is how much money we need,'" Hodge emphasizes.

Benchmarking Buckets

The process to determine spending requirements starts by dividing costs into major benchmarking buckets, which Goranson describes as "the functionals important to the facility."

Among the benchmarking buckets that Intel uses are categories like mechanical support, electrical support, ultra-pure water, and waste systems. These are further subdivided into what Intel calls core businesses. For example, under the "mechanical" heading are air handling, chilled water, compressed air, and exhaust systems, etc.

Absent from this model are some typical facility functions like janitorial and landscaping services.

"Custodial services account for a much larger proportion of resources in an office facility, but in a fab they are only a small part of what goes on, nowhere near one of the biggest drivers," says Hodge. "We do have a bucket for custodial services, but we made a conscious decision not to include that here."

Staffing for the benchmarked categories is broken down into three types: administrative, engineering, and technician.

With the services and staffing positions catalogued, Intel could begin to calculate the headcount necessary to support each subsystem in the baseline 150,000-sf model plant.

An Even Playing Field

Further steps are necessary to tally up costs before the baseline spending model can be applied to real-world facilities.

Some factories, especially the new chip fabs, have additional requirements driven by cutting-edge technologies or processes, ranging from infrared (IR) tool scan to special waste treatment systems to total chemical management. Any facility-based spending formula must take these extra services or systems into account.

The subteam's response was to identify two tiers of support to reflect such distinctions. Level Two encompasses the baseline services common to all plants, while Level Three includes the gamut of extra customer- or process-driven needs CS must provide at the more complex sites. The two-tiered structure allows the different support requirements to be expressed quantitatively, essentially evening out the playing field so dissimilar plants are not subjected to the same measuring criteria.

With this full scope of headcount drivers, the team was prepared to assign the appropriate headcount numbers to each benchmarking core business. [See Figure 3: Competitive Services Model.] Once these calculations were completed, the conclusions were submitted to functional content experts who were not part of the team to confirm their validity. This "sanity check" also helped in obtaining internal partner buy-in.

"We asked the experts whether they could operate a facility of this size using that kind of headcount," says Hodge. "We got a lot of input and they confirmed that it could be done."

Burden Rates and Scaling

With more than 10 facilities worldwide targeted in the effort, the Cost Ops subteam needed to factor geographic cost differences into the competitive services model. Using burden rates provided by each site, Intel formulated a payroll algorithm allowing for these differences to produce an expected dollar cost for the minimum headcount service level.

Other expenses—chemicals and gasses, parts and spares, travel and training, and consulting costs—were similarly burdened to forecast an "expected minimum facilities sustaining cost" for the theoretical 150,000-sf cleanroom facility.

To further refine the accuracy of projected spending, the subteam adopted data from the manufacturing organization that stood out as best-in-class in the benchmarking buckets.

Those figures were populated into a model [a sample of which is shown in Figure 4: Expected Costs] that was then scaled to the size of each individual factory. Scaling factors are pegged to either cleanroom gsf (CGSF) or wafer starts per week (WSPW), whichever has a stronger correlation. For example, mechanical, electrical, and ultra-pure water systems are scaled by CGSF, while sub-fab services are scaled by WSPW.

"The more wafers you run, the more tools you have in the fab, and the more tools you have in the fab, the more systems you have directly below in the sub-fab," observes Hodge.

CS found one further tweak to the model was necessary to address multiplant locations, which might benefit from economies of scale not available to freestanding factories. Probing confirmed that headcount and other expenses should increase at less than a linear rate at these "behemoth" sites, so the model was adjusted to reflect this reality.

Closing the Gap

Once all the variables were incorporated, the team was able to use the model to calculate anticipated costs for each site. The budget request from each manufacturing plant was then overlaid on the final model [see Figure 5: Apparent Cost Savings]. Opportunities for cost savings became apparent from the differences between the model spending and the requested spending.

Despite the detail of the model, individual sites are not tied to the designated headcounts specified for their size and production parameters. Rather, they enjoy considerable autonomy in determining how to close the gap between the spending targets and their own requests.

This flexibility has led to a number of operational changes that reduce spending without altering service levels. (In fact, the competitive services model was originally called the "minimum services model" but was renamed to signify that it didn't entail a diminution in services.) For example, Hodge cites one plant that moved from outsourced to in-house facilities technician support, which has proved to be more cost-competitive.

"This is something that from the strategic side we have been aware of for a few years," Hodge comments. "We're excited that the model has led people to reexamine their current practices and make changes."

Another innovation is the maximization of PEMS (for Predictive Excursion Management System), a statistical process control for facilities systems that monitors critical parameters for indications that maintenance needs to be done. Instead of following a rigid preventive maintenance timetable for a piece of equipment, a pump, for instance, technicians would measure vibration and wait until signs of degrading performance are detected before scheduling the work.

Other economies have been produced by eliminating non-conforming shift structures, broadening employees' skill sets, and deploying PDAs.

In addition, the initiative has created a climate that emphasizes constant improvement as sites continue to come up with new ways to trim expenses.

"The model says, 'here's where your spending is and here is where you should be,'" says Goranson. "That makes people ask what's going on at other sites and inspires sharing across the globe."

Cross-Industry Applications

While the competitive services model targets Intel's manufacturing plants (actually, just the fab portion of manufacturing, although assembly is next on the agenda), Hodge is convinced that it can be ported to any kind of facility.

"The basic principles apply to R&D, office, or any other type of building," he points out. "What's special to the model is that we picked the parameters important to the facility involved."

Still, facilities professionals in other disciplines may find it quite a leap to adopt the same yardsticks of headcount and best-in-class materials spending.

"They might not have the tools to get data easily to do something like this," he says, explaining that several years ago Intel realized the need to invest in building a data-gathering network to remain competitive.

Hodge does concede that it would be easier to apply a similar process to office space only because services in that environment are much less complex. However, the requirement for accurate, comparable data across the organization is absolute. A company with decentralized operations where every site does things differently—the classic apples and oranges syndrome—would not be able to benefit from the model.

"The scope of service A at site one might be different at site two," he explains. "You must have the resources and infrastructure in place to get cross-company and external information to do these comparisons."

By Nicole Zaro Stahl