In the 1990s, silicon chip makers could pretty much dictate terms to their customers. Need a few million microprocessors for your latest line of PCs, video games, or cell phones? Intel, the world's biggest supplier, would tell you if, when and how much you'd get. At a time of massive demand for new high-tech products, it was a seller's market by any definition of the term.
If there's an indisputable lesson to be learned from economic studies, it's that dominant players don't stay that way forever. By the turn of the new century, Intel was beginning to pay a price for the way its operations were set up, in the form of growing customer dissatisfaction. The company's inherent inefficiencies could no longer be ignored.
Intel had designed its supply chain to service a handful of large original equipment manufacturers (OEMs), who required a limited number of SKUs. Order fulfillment was predicated on several guidelines: that constrained supply be disbursed in a "fair and equitable" manner, that commitments for product be backed up by physical supply, and that quantities be fixed as far in advance as possible, to ensure predictability in the chain.
Not a bad plan from a supplier's standpoint. Intel's customers, however, weren't so content. They were required to place orders at least three months ahead of delivery, and their change requests had to pass through several levels of review before being allowed. (Assuming, of course, that they were.) That disadvantage began to chafe as a fragmenting customer base led to more volatile demand, and increased pricing pressures. Order changes stepped up in frequency, and Intel simply couldn't keep up with the flow.
A study of 10 semiconductor companies by IBM Global Services found that most could respond to a supply request on the same day it was made. In 2004-05, Intel needed seven to nine days, a dismal record which made it "worst in class," according to supply-chain strategy program manager Nikhil Chhabra. Or, to put it more bluntly: "Some of our customers hated us."
The year 2005 saw the launch of an initiative that Intel dubbed "Just Say Yes." The name signaled a determination by the company to forge a new approach to customer relations. More than a marketing campaign with a catchy slogan, the plan involved a number of ambitious goals touching upon multiple aspects of the supply chain. First up was the ability to respond quickly to change-order requests. That would be followed by an improvement in committed dock-date performance, a push to reduce errors in demand forecasting, and an overall reduction in inventory levels. (That last achievement was especially important, given Intel's insistence on reaching all the other objectives without boosting safety stock.)
"Just Say Yes" ended up involving hundreds of separate projects. Given the size and scope of Intel - 87,000 employees worldwide, nearly $38bn in revenues and net income topping $5bn - that's hardly surprising. In 2006, the company processed more than 1.5m change orders.
The effort was structured around four "pillars," targeting improvements in responsiveness, inventory balance, delivery performance and forecast accuracy. At bottom, says Chhabra, was the common goal of figuring out "how many times we can say yes in a [business] day."
How It Was
Under the old way of doing business, change-order requests had to pass through four separate departments: three separate reviews for "business validity," followed by an attempt to match demand with actual supply. Customers were understandably frustrated. According to Chhabra, a "no" from Intel often took longer to deliver than a "yes."
The company set out to create a common set of business rules, known as guard-bands, to be utilized by multiple departments. If a change order fell within the parameters of those rules, the customer would be granted access to supply without need for further review. At the same time, customer-service teams were given the power to mix and match cancellations and reschedules with supply requests. A series of performance dashboards highlighted any bottlenecks and ensured that the process stayed on track.
To enhance product availability, Intel scrapped its reliance on spreadsheets and embraced technology that created a real-time, available-to-promise (ATP) environment. For the first time, says Chhabra, supply became visible "on a minute-by-minute basis." In one of those actions that must seem self-evident after the fact, customer-service teams were given direct access to the ATP system. Not only could they say "yes" more quickly, they could offer customers alternatives in the event that orders couldn't be fully accommodated.
"People seen as order takers were essentially negotiating with customers about what they wanted," says Chhabra. "That was a huge cultural change."
Addressing the issue of long order lead times, Intel shrank the ATP horizon to less than that required for orders or backlogs. Big orders that were likely to be changed or canceled no longer took up available supply. Previously, many customers had treated their initial orders as "placeholders," to be adjusted when they figured out what they really needed. A series of new processes and metrics hunted down and exposed unused pockets of supply - not an uncommon problem for a company with sales regions around the globe.
To make up for what it saw as a lack of off-the-shelf technology, Intel built its own system which consolidated all of the data and supported multiple transactions - guard-band testing, ATP mapping and escalation processes, to name a few - by way of a single interface. It eliminated most manual input while allowing third parties to complete order transactions upon confirmation of supply. In essence, the company outsourced the process of order execution, creating "a virtual 24-hour operation."
Results were quick to emerge. Prior to introduction of the new system and processes, Intel was confirming just 21 percent of change orders within one business day. The number rose to 52 percent by the end of 2007, and topped 70 percent by the end of 2008.
Order approval was just the first step; now Intel had to deliver the product on time. It was seriously lagging in committed dock date (CDD), meeting that metric less than 25 percent of the time. The solution lay in creation of a process known as the Metric Wheel. Cross-functional teams followed the six-step drill of (1) making it important, (2) measuring the baseline, (3) proposing targets, (4) identifying improvement projects, (5) funding projects and (6) executing the project road map. In the event, they were able to uncover the root causes of most CDD failures.
According to Chhabra, Intel quickly whipped its CDD performance into shape, achieving best-in-class status by 2007. The company's perfect-order record stood at 96 percent in 2008, and has been "improving sharply" this year. Deviations in performance among sales regions, which had varied by up to 30 percent, were reduced "to acceptable levels."
Focus on Inventory
The first impulse of a business looking to boost product availability might be to increase stock on hand. Not so with Intel. It had already seen a sharp rise in pipeline inventory during 2005 and into 2006, triggered by lengthy production cycles, customer issues and the complexities of launching new products. Now it was determined to take those inventory levels down.
Opportunities existed throughout the chain. In the factory, Intel applied Lean principles to eliminate process waste and cut wafer-fabrication time by 10 to 15 percent. The effort led to a corresponding reduction in work in process (WIP). The practice of building product ahead of demand was curtailed, aided by process improvements and better risk analysis. Greater use was made of postponement practices, whereby product is customized for a specific use or market at the latest possible stage. Lean techniques were also instrumental in reducing planning cycles.
As a result, Intel saw a 62-percent drop in manufacturing cycle times between 2006 and 2009, according to Chhabra. Raw materials, WIP and finished goods fell by 33 percent. And the "health" of finished-goods stock - its relationship to true demand - improved by 40 percent in 2008.
Also on the rise in the early 00's was demand forecast error. Again, Intel's track record was much worse than the industry average. Meanwhile, it was seeing an increase in product variety, emerging markets and competition. The need for accurately predicting the future was greater than ever before.
Intel must have viewed itself as customer-centric in the 1990s. Didn't it strive to allocate product fairly during supply shortages? Didn't it solicit input from customers before making those decisions? The answer in both cases was yes, but that wasn't enough. Most customers couldn't accurately forecast their needs within the three- to six-month horizon that was necessitated by Intel's long manufacturing cycles. in addition to shrinking time to market, the company needed to supplement customer input with a broader view of demand.
Market variability, shorter product lifecycles and a lengthy ramp-up procedure for new products were the culprits behind most forecast errors. So Intel installed a new methodology called the Product Transition Index (PTI), which maps the evolution of products from one generation to the next, and encourages the sharing of knowledge in such areas as market positioning, strategy and uncertainty. Rigorous statistical methods, such as diffusion modeling, were also blended into the PTI.
Forecast metrics changed, too. Intel went from measuring for average error percentage on total volume to Weighted Mean Absolute Percent Error, or WMAPE. And it began judging the rate of success in forecasting demand among product families.
Relying on a tool known as Prediction Modeling, the company sought insights from employees across all functional groups. By aggregating information from a broader range of internal sources, it could lesson the forecaster "bias and gaming" that had previously skewed the results.
Finally, Intel removed the crutch of buffer stock as a means of coping with demand uncertainty. In its place, the company relied on formulas for safety-stock modeling.
As of mid-2008, Intel had managed to match industry median performance in forecasting in six of the previous 11 quarters. In 10 of those quarters, absolute error fell within the acceptable range of deviation. WMAPE levels improved by better than 20 percent, hovering at just above the industry median of 28 percent. All of this took place at time of growing complexity in product mix, Chhabra points out.
Phase II Begins
So ended phase one of the Intel transformation. Just Say Yes Part II was launched in 2008 and continues today. Its goal, says Chhabra is an even more intensive focus on customer responsiveness, calling for a variety of new business processes and technology.
Intel's old method of judging service level was to assess its speed - however slow that might have been - in responding positively to change orders. Now, the company has embraced a series of standard metrics prescribed by the Supply-Chain Operations Reference (SCOR) model. The tool allows the company to benchmark its performance across supply chains. New metrics include Order Fulfillment Lead Time (OFLT) and the ever-popular Perfect Order.
Vendor-managed inventory (VMI) has long been a popular means of delaying the transfer of goods or components from supplier to manufacturer until the moment they are needed on the assembly line, or in the buyer's distribution pipeline. Products that are characterized by a high level of innovation typically don't lend themselves well to a VMI arrangement, which often involves some use of consignment. Intel has nonetheless been able to adopt a flavor of VMI for its larger OEM customers. It is asking suppliers to hold buffer inventory at their regional hubs, close to the customer's manufacturing sites. The presence of those hubs eliminates the need for customers to place orders well ahead of time. In certain cases, change orders aren't required it all, moving Intel in the direction of a true "pull-oriented" supply chain.
On the technology side, new software modules have made Intel's order-management processes more efficient. Supporting business processes were altered accordingly, according to Chhabra.
He says Intel is only about halfway to meeting its goal of applying so-called "Dynamic VMI" principles to all sitting inventory. Already, though, the company has seen some big improvements in the area of order fulfillment. Average OFLT is in the one-day range, versus the several weeks that the company was taking only a year ago. Average Perfect-Order performance for product flowing from the supplier hubs is consistent with industry best-in-class.
Performance in other segments of the distribution chain remains well below industry average, however. Intel plans to attack that area in 2009 and 2010, possibly through the forward-staging of inventory and the providing of better information to customers about the short-term availability of product. Again, the idea is to remove the necessity for buyers to place orders long before they need the product - or even understand precisely what it is that they need. A pilot conducted with two European distributors in 2007 and 2008 yielded some early positive results, says Chhabra. During that time, the order-placement horizon shrank from three months to about week, and change orders were virtually eliminated.
Intel was struck by the discovery that nearly three-quarters of all change-order requests occur in the last four weeks before shipment. In fact, the company says, most customers make the adjustment less than two weeks prior to delivery. Revisions within the two- to four-week range tend to be either requests for more product or outright cancellations, while those in the final weeks are mostly in the nature of date changes. So asking customers to book orders several months prior to actual need "has little or no value to either party," says Chhabra. With this realization in mind, Intel has sought to shorten the order horizon for a handful of customers, resulting in a 40-percent reduction in change-order requests.
All of these efforts have been accompanied by a drastic simplification of Intel's planning process, resulting in 15 out of 21 major process steps either being eliminated or redesigned, in line with industry standard practices. The need for spreadsheets has been eliminated by 95 percent. As Chhabra puts it, the whole effort "is dragging Intel into the supply-chain leadership role."
Timely, incisive articles delivered directly to your inbox.