Hewlett-Packard's Imaging and Printing business knew that it had to overhaul its supply chain. The unit constantly struggled with too much of the wrong products and not enough of the right ones, and resellers' complaints were soaring. The answer came in the form of a holistic value chain—an approach that synchronizes the supply chain by working backward from the customer, truly understanding demand, and focusing on business processes instead of technical solutions.

The frustration in the room was palpable. It was May of 1999, and the management team of Hewlett-Packard's (HP) North America Consumer Imaging and Printing operations was holding an offsite meeting near our home office in Vancouver, Wash. Our goal was to agree on a multiyear plan to roll out advanced planning software to improve operational performance.

The division had just gone through a terrible holiday selling season, regularly building too much of the wrong products and not enough of the right ones despite heroic efforts all across the division. In fact, availability and delivery performance had deteriorated, and firefighting had become the order of the day. HP saw excess inventory left homeless by retailers; the retailers saw shelf stock-outs. Increasingly, end customers were buying from competitors when HP products weren't on retailers' shelves.

Internally, morale in the division was low. There was defensiveness and plenty of finger-pointing over lack of cross-functional coordination, forecast accuracy, and supply mishaps. Recent attempts at implementing collaborative planning, forecasting, and replenishment (CPFR) processes and other improvement initiatives had not provided expected benefits.

Trish Stevens-Gemulla was the first to speak up at the meeting. "We can't go through another holiday selling season like this one," she said. "We need solutions delivered this year!" As the leader of the division's North American account operations teams and channel planning function, Stevens-Gemulla was on the front lines, constantly trying to match supply with demand and juggle constraints as far back as the factory. She was also in the line of fire when the angry phone calls came in from retailers' top executives.

Stevens-Gemulla's argument electrified the meeting. Consensus quickly emerged that new planning software was not the solution; what was needed was a complete rethinking of the whole value chain process—beginning with hard commitments to account forecasts. We tabled the plans for a large-scale software system rollout. Instead, we immediately began to put all our resources and energy behind high-impact process improvements, launching a "discovery team" to quickly map out and tear down our current demand and supply management processes.

This group of HP managers would help identify and dissect the North American supply chain problem and then map out a program to restore competitive strength and confidence. A key part of the program was establishing the values and characteristics needed to move from collaborative planning to what HP calls "holistic value chain planning." This closed-loop process begins with capturing channel demand in a collaborative forecast. It then flows through to supply planning operations and circles back with high-confidence commitments to the retail channel.

Five years after rethinking our demand and supply management processes, HP has maintained and built upon its strength in consumer printing products. And, even more importantly, we have regained the confidence of our retail channels, our suppliers, and our employees. We believe that the lessons we've learned can apply to any company looking to review its supply chain planning processes.

The May 1999 offsite meeting brought a few hard truths home to our management team. Our North American resellers had begun to run out of confidence and patience. Many found they could not reliably support promotions with HP product—especially when they faced advertising leadtimes of eight weeks and longer. "If HP could not deliver with high confidence, we were not going to place all our market-share eggs in one ad basket," recalls a Best Buy account representative.

It was quite a slap in the face for HP. After all, we had essentially created the inkjet printer category when we'd launched the Thinkjet in 1984, and we'd always dominated that market. Now we were seeing stiff competition from companies such as Canon, Epson, and Lexmark. At the same time, we were failing to support our resellers with the product they wanted. Our long inbound leadtimes, highly variable demand, and high forecast error were especially problematic when we introduced new products; we often had to discount the older products more steeply than we liked. To make matters worse, the discounts reset expectations for the price of the new product.

Our divisional leadership was fully aware of what was going on. They heard about it directly in the exasperated phone calls from retailers' executives. By early 1999, they had tasked managers such as Trish Stevens-Gemulla and co-authors Steve Radosevich and Ike Harris with fixing the problem. The May 1999 meeting was one of the earliest steps.

We knew we had to hit the problem hard—and fast. The next four weeks were spent assessing all of our demand and supply processes, internally and in cooperation with key channel partners. Several key findings emerged:

• Our CPFR processes didn't connect to upstream supply planning processes. While we had been doing joint planning with major retailers for several years, these initiatives had developed as individual efforts. There was virtually no connection between the detailed planning and forecasting that was happening with CPFR and the production plans against which manufacturing worked.
• Supply planning was a guessing game. The cause: poor-quality inventory data throughout the supply chain. Records for on-hand inventory were riddled with data issues and badly needed to be cleaned up. In-transit inventory and delivery dates were even worse. Distribution centers couldn't count on a delivery being made within 14 days of an expected receipt date.
• Demand changes were slow to be incorporated into production plans. Because the factories in Asia operated on a monthly plan, it took a long time to get demand changes reflected in production plans. As a result, nobody knew how many units were being produced in any given week, and HP missed deliveries or held excessive inventory buffers.
• Forecasting at the SKU level was really difficult. Despite our best efforts, creating accurate weekly forecasts down to the SKU level—down to any granular level, in fact—proved to be a huge challenge. As a result, the supply chain had built in expensive mechanisms to cope with such uncertainty in demand, such as deliberately overbuilding this or that product or sending expedited air shipments.
• Reliable supply and delivery were like gold. We had learned a lot about retail demand patterns—for example, how much they were driven by product introductions, ads and promotions, and pricing actions. But our upstream supply planning processes were not designed to support those needs. Most retailers, we discovered, were committing resources to ad plans and store promotions as much as eight weeks before execution. They made it very clear how critical it was to have the supply at the right time to support these investments.

We also took a hard look at past efforts to improve operational performance. We began to see some systemic flaws. Executive sponsorship had often been poorly defined; business sponsors were rarely actively involved in development and weren't accountable for the results of the efforts. It became clear, too, that those involved in change initiatives often lacked solid project management skills and methods. And many past efforts had focused on glamorous solutions, neglecting the basic building blocks. "Building swimming pools in the attic" was the catch-phrase.

The CPFR program that once had been trumpeted as the breakthrough we needed to handle hard-to-predict demand was not the answer by itself.¹ In fact, all it really did was add a demand signal without any process to integrate the information. Resellers were less than delighted. Having invested in CPFR processes and systems so they could do business more effectively with us, they had naturally expected benefits in the form of increased assurance of supply. Eventually, the whole premise of CPFR was subverted as the retailers began to escalate their supply problems to our top management.

By the middle of the summer of 1999, we had coalesced into a formal team to rewrite our supply chain rule book. Planning architecture chief Steve Radosevich headed up the holistic value chain planning program which we dubbed the "Demand-Supply Management Program" (DSMP). We had a clear idea of our overarching objective, which was to ensure that demand signals flowed back through the supply chain and were met with firm commitments. (See Exhibit 1.) With some of the biggest issues identified, we formed three teams around the major business processes. One focused on demand management (all aspects of predicting demand and product allocation); it was sponsored by Trish Stevens-Gemulla. The other two teams focused on supply planning at all nodes in the supply chain and on data quality management; both were sponsored by Ike Harris. Altogether, we had more than 150 people involved in the program, with about 30 at its core; it was one of the most significant change initiatives our division had ever launched. There can't have been anyone in the division who didn't know what DSMP was trying to do.

We made some good moves at the outset. Of course, it was vital to assemble a cross-functional team—representatives from sales, planning, marketing, and manufacturing—to take a holistic view of the value chain. It was also crucial to make the team members accountable for the outcomes. We deliberately avoid focusing on problems or metrics specific to any one group or transaction. But the DSMP's success really hinged on four other mechanisms:

• We established a high-level vision early on. Our rapid and thorough dissection of the problem allowed us to scope out a detailed concept of how things should work. The vision of demand pulled from the channel partners and of an end-to-end supply chain synchronized to those demand signals became the foundation for all communications with executive sponsors, supply-side managers, channel partners, and suppliers. We emphasized not just collaboration but synchronized planning on weekly cycles with firm commitments to our partners. (Exhibit 2 depicts the weekly planning process.) This vision remains our foundational operational strategy today.
• We locked in executive sponsorship. Learning from earlier project-sponsorship failures, we set up biweekly sponsor reviews with three executives: one from the supply side, one from the demand side, and one from planning. The regular review sessions reaffirmed alignment, led to firm decision making, kept things on schedule, and helped communicate progress within the division and to resellers and suppliers.
• We built analytical probability models to show how supply chains respond to demand signals over time. It is always hard to determine a "return on investment" for planning. Our sophisticated model mapped the probability distribution for expected forecast error. These graphs helped to clearly communicate to managers the DSMP's main value and goals.
• We brought in experts on program management and process development. Our division had a woeful record of implementing anything of success; in general, our change programs had suffered from poor project management. So we recruited top-notch program managers and process development engineers from across the division and from other parts of HP. Then we spent a lot of time training them in program management fundamentals. We didn't choose to bring in a big-ticket consulting firm to solve our problems. We knew we had the talent in-house.

There was some pushback to the DSMP, of course; few people love upheaval. It wasn't easy to get agreement that near-term forecasting would be "owned" by the channel operations team not by marketing—a very unconventional but crucial move. There was also grumbling and some dissension from the more traditional production managers who had long held to monthly measurements. "This is just going to add variability" was a typical complaint. For the most part, though, we had the tacit backing of most people within the division. Everyone knew the problem had to be fixed.

Revamping CPFR

By early 2000, we had revamped our CPFR processes. We acknowledged that our initial efforts had failed to bring real improvements to value chain performance; we now saw that the resulting shared forecast didn't connect to what the factory did. Even so, we still believed that CPFR was a powerful process that could generate significant value—if we utilized it correctly. We knew we still needed a top-down forecast for long-range planning, but we also knew that a bottom-up CPFR process could yield the detailed insights about retailers' plans, promotions, and operational policies that we needed.

So we restructured the CPFR system to enable the account teams to create a detailed weekly workflow. Now they submit a 12-week order forecasts each week. To manage all the incoming data, we set up a database with a simple spreadsheet as the front-end interface. (Remember: We wanted quick implementations.)

Now that we had solid forecasts and promotion plans from our sales force and/or major channel partners, we developed a new position: a channel analyst to aggregate and analyze the demand signals. The role entails short-term demand planning that rolls up the account-level forecasts and then adjusts (up or down) an aggregated signal based on further analysis of sales and channel inventory levels. One of the primary outputs of the channel analyst is a short-term SKU-level order forecast. The new job has become essential to our operations.

As the new processes began to operate fully in 2000, the DSMP team realized that our earlier CPFR deployment had generated conversations only about demand, not supply. That left a lot of room for missed expectations. We were not "closing the loop" to make commitments between the collaborating partners.

Committing to a Commit Plan

Our goal was to have the supply side produce reliable responses to the retailers' demand forecasts—in effect, a commit plan. We wanted to provide a commitment back to resellers in the same week that they had sent demand signals so that all parties would know what could and could not be supported. That way, we would have the basis of a solid "handshake" between demand and supply.

In looking at traditional supply and demand matching tools, we found a glaring hole. The tools never took into account the results from the previous plan. We didn't want to completely reshuffle the deck each week. Rather, we wanted the logic to reward good forecasting and provide stability in supply commitments (a commit that could be counted on!). In 2001, we began to construct a proprietary supply/demand match tool—an allocation engine—with the desired business rules and algorithms, using minimal resources and on a short timeline.

The philosophy behind the matching tool was to honor previous commitments first. Each week, as new forecasts were submitted, the tool first took into account the previous week's commit plan. Once all previous commits had been met, new demand was allocated. When all demand was met or when supply was tapped out across the planning horizon, the commit plan was completed and published to each sales team and retailer at the end of the week. We were not limited to quantities sitting in on-hand inventory; we could commit to units not yet produced in Asia. Our partners now knew exactly what they could expect to receive in the next 12 weeks in response to their weekly forecasts.

Our sales force and resellers caught on fast. They had a clear incentive to provide accurate weekly forecasts—and to move HP product rather than competitors' offerings. Having collaborated on business plans and forecasts, resellers had growing confidence in HP's execution, and they could focus on their own activities. Several now had more of an incentive to make HP displays more visible in store locations, and they were more willing to consider spending additional advertising dollars to support our products. The information provided in the collaboration process no longer fell into a black hole; it was used to determine allocation quantities.

Before long, we began to notice that "issue escalation" by the resellers—those executive-to-executive phone calls—began to dry up. We had started to remove the need for them, and we'd built internal mechanisms to shut them down. With solid processes and business rules in place, the starting point for these conversations would be fact-based—that is, built on data. Our divisional leaders, in the event of such an escalation, would start with two questions: What was our collaborative forecast? And what was our commitment quantity? These simple queries quickly diffused the escalation scenario, refocusing energy on timely communication and problem solving.

Changing Metrics to Get Predictable Supply

We realized that it would be a disaster if the commit plan could not be counted on week in and week out. We figured it had to be executable at least 95 percent of the time. That meant having very high confidence in supply execution or, if necessary, in holding inventory. The goal we established was to create a highly predictable plan of supply that aligned across all nodes in the supply chain—using less inventory than we were holding at the time.

Creating predictable supply started with a data quality management initiative. We cleaned up data system linkages that were poorly maintained and established accountability for on-hand inventory levels and in-transit monitoring. This generated immediate benefits because it led to planner confidence in knowing exact on-hand or in-transit levels at all times. We also improved processes for managing expected delivery dates. We moved from weeks of accuracy to within a day or two of predictability.

It proved more difficult to change the factory's operating policies and metrics. Three fundamental changes were implemented. First, the factory moved to weekly production measures and weekly execution of the delivery plan instead of monthly. By changing to a weekly plan and metrics, we could know with certainty when units would arrive and that data could then be synchronized with the demand plans from the CPFR process.

The second area of factory focus was to trim the "frozen" build horizon—that period of time in which production plans are firmly set—from five weeks to two. We also went to a biweekly cycle for deciding product mix and volumes within specified guidelines. This major change let us take advantage of production upside—or the extra manufacturing capacity available to us—as demand changed in the account forecasts. Moreover, it allowed us to slow down production or switch to another region to minimize excess. The third change was to increase "factory upside responsiveness" guidelines. The more quickly our contract manufacturers in Asia can expand or shrink the capacity available to us, the lower the levels of inventory we need to support demand. We have guideline agreements with our contract manufacturers that provide parameters for increasing and decreasing output over specific time horizons. A rigorous modeling exercise recommended that a 20-percent factory-responsiveness level—or upside capacity—would permit lower inventory targets and take advantage of the weekly forecast. Having some ability to change production more often would give HP increased sales, fewer stockouts, and lower inventory.

Looking back on it, the factory team's changes were significant. Their primary metric had always been product cost. The changes required them to be more flexible and to take on more production volatility in the short term. Such changes could only be accomplished if all parties were working toward system-wide value chain goals.

The DSMP paid off as we successfully navigated several new product introductions and the holiday selling seasons. It didn't make us immune to allocation issues, of course; ensuring 12-week visibility and commit plans didn't mean that we could simply spin the factories faster to keep up with market demand. But it did mean we could operate consistently and predictably with minimal inventory levels. When demand soared in 2000, the new connected processes proved invaluable at keeping product moving quickly through the supply chain—and keeping product on resellers' shelves.

The big opportunity to upgrade the program came in 2001 and into 2002 with what was billed as HP's "Big Bang"—an unprecedented effort to build a range of high-quality imaging and printing products that would sell for very low prices.² Retailers liked what they saw but needed assurance that HP could deliver the volumes required given recent memories of product-allocation problems. So we tweaked the program to extend commitments to six months, incorporating long-term capital-planning considerations. The Big Bang was a huge success: By midsummer, we had placed more than a million new printer models on shelves.

By 2002, the weekly end-to-end process had proven itself in aligning demand and supply from the channel partners back to the factories. The numbers were striking: Compared to our "baseline" numbers from 1999, inventory investment across the value chain had decreased by 20 percent on average. Channel partners held less inventory because of their confidence in our supply commit plans. In addition, our improved SKU-level forecasting, data quality, and factory responsiveness allowed us to drop inventory targets by 50 percent. Delivery performance improved by 80 percent from 1999 to 2002, and our ship-to-plan numbers went up from the 40-percent range to more than 90 percent in the same period. At the same time, the predictability of our 12-week supply plan was consistently above 95 percent. We had achieved this by firmly focusing on business processes, data, and relatively simple technical tools. At the same time, we boosted near-term forecast accuracy by 30 percent.

Since then, we've followed up with a range of refinement initiatives, including a more robust planning database and a reporting platform that provides standardized views for decision making, measuring and reporting performance, and identifying improvement opportunities. We also launched a team to create an "available to sell" capability. The idea was to go beyond the traditional system in which only on-hand uncommitted inventory was made visible. We wanted full visibility of uncommitted supply potential across the entire 12-week horizon so we could feed that information to the sales force. Based on that effort, we began publishing a SKU-level availability report indicating upside levels over time. This gave account teams the information they needed to steer demand and promotions into SKUs and/or time horizons with available supply and away from times when a SKU was in tight supply.

The final piece to the puzzle came when we began to monetize the bottoms-up forecasts and use the result as the basis of business management. Our controller challenged us to achieve end-of-quarter financial predictability within 5 percent of projection. With solid plans aligned between sales, marketing, and operations, this challenge was not as unrealistic as we had thought. Working the processes, we have been able to consistently manage the end of a quarter within this goal. As important: By tying the unit plans to the financial plans, we have created an additional level of rigor and accountability that makes the process that much more robust.

We've stopped dreading product rollovers and the holiday selling period. Now that responsiveness has become a proven capability, we can manage change as it happens. With plans at the SKU level by retailer, we've also seen a dramatic improvement in forecast mix accuracy. New product launches are smoother. Collaborative end-to-end planning lets us communicate properly to keep retail shelves well stocked throughout the transition from old products to new.

Crucially, we've been able to restore and reinforce confidence across the value chain. The firefighting, finger-pointing, and defensiveness have been replaced with teamwork and joint problem solving against shared business objectives. We have far less second-guessing and buffering against uncertainty. Indeed, the focus on predictability has become an organizational core value. People take great pride in consistently delivering on commitments. Our mantra has become "plan-commit-deliver."

Some numbers tell the story. At the outset of the DSMP, we could only meet a four-week channel forecast 70 percent of the time. But by last year, that figure had climbed to 97 percent—and has stayed there. Over the same period, we have halved the number of times when product in the channel dropped below the level needed to support on-shelf availability.

The new process has also enabled us to slash 10 weeks off our total planning cycle time (See Exhibit 3.) The top bar depicts the total time it previously took HP to respond to a change in mean demand from the point at which we noticed the trend, working through a monthly planning cycle, bringing units by ocean carrier from Asia, to postponement (final product configuration), to getting them to end customers. The total planning time was more than half a product life cycle!

Because our resellers now trust that HP product will be shipped to their shelves when they want it, they have responded with stronger order commitments. From a low of 61 percent of committed quantities purchased in 2000, the number neared 90 percent in 2004, giving us confidence that our planned production will be sold in sync with our financial goals. The results have been apparent in our cost structure: Because we can now plan confidently over longer horizons, we've been able to reduce annual product rollover costs (end-of-life discounts) by 60 percent.

Resellers have seen significant gains too. "A longer-term and more detailed view of HP product availability...has allowed us to expand just-in-time inventory management, better manage life cycles, and maintain an average of 96-percent product in-stock. We have adapted the HP approach as a model for collaborating with other vendors," says Kim Miller, Office Depot's technology inventory manager.

If there is one lesson we've learned from our holistic value chain planning overhaul, it's the need to focus first on customers and business processes. It sounds clichéd, but we now know you can only synchronize the supply chain by working backward from the customer, understanding demand and all its fluctuations in detail, and focusing on business processes instead of technical solutions. We now see that a demand-driven organization starts with having the sales force as a connected partner in a formal demand and supply management system. That way, you obtain the greatest insights into demand and establish a clear connection between the accountability of forecasting with the consequence of inventory (excess and shortage).

We also realize that management systems that once worked won't necessarily continue to work as the basis of competition changes. Implementing our CPFR program in the 1990s was a move that was only half-right. Yes, it allowed us to gather plenty of valuable data, but it lacked the end-to-end visibility vital to a flexible supply chain model in the volatile consumer electronics industry.

Lastly, it has been vital to establish and maintain solid project-management skills and to ensure constant support from senior management. We brought in and developed strong project managers to run our holistic value chain planning effort. We instituted regular review meetings with management. Every aspect of scope, schedule, and resources was taken seriously. Rapid issue identification and resolution techniques were deployed to make sure that work didn't slow down and decisions were made quickly.

Having a holistic perspective of the end-to-end supply chain is important. To accomplish this, however, simple but meaningful questions must be answered—questions about how the new demand data will be incorporated into the demand and supply planning processes, about the timing of the end-to-end process, and about the rules for the allocation of product, for example.

Implementing a holistic value chain planning program is no easy task. Nor is it a one-size-fits-all remedy. At HP, it comprised an exhaustive program that called for continuous commitment by dozens of managers across many functions and facilities. It constantly tested our communication skills, and it always forced us back from the temptation of the short-term. It's been absolutely worth it.