Accelerating IT innovation.
They are the Oscars of information technology: every year the Smithsonian Institution and Computerworld magazine award prizes to a handful of computer programs that, in the eyes of the judges, exemplify the most innovative applications of technology. In 1994, the prizes went to software that enables more natural use of artificial limbs, a computer simulation of the world's oceans, the programs that created realistic dinosaurs for the movie Jurassic Park, NASDAQ's networked financial markets, and the inventory control system at Mervyn's, the well-known department store. Why Mervyn's? Why inventory control?
Mervyn's executives asked themselves the same questions as they mounted the stage to accept their award. The answer is not simply that their Planned Store Inventory system (PSI) is an interesting and useful piece of software, that adds tens of millions of dollars a year to the company's profits. The real achievement is that Mervyn's has worked out a new way to build systems that, by making the inevitable risks of innovation more manageable, can speed IT-led innovation throughout a business - any business.
The accomplishment of PSI
From its headquarters in Hayward, California, Mervyn's manages a chain of nearly 300 department stores stretched across America, selling over $4 billion-worth a year of mostly middle-market clothing. Its PSI system addresses two problems at the heart of its business. First, how can Mervyn's get the right goods to each of its stores, even as sales fluctuate through local promotions and changing fashions? And second, how can it get its employees to take full advantage of the opportunities created by advances in information technology?
Neither of these problems is unique to Mervyn's, or to retail. Be it inventory management or product development, companies everywhere are struggling to use IT to help them react more quickly to change - and to tailor their services more precisely to local markets. That means they must learn how to change business practices rapidly and enthusiastically to take up the new opportunities that technology brings. On both counts, there is much to learn from Mervyn's experiences.
One of the first things to appreciate about the PSI system is how successful it was in reaching its goals. Instead of distributing inventory on the basis of average sales in a group of stores, as Mervyn's had previously done, PSI distributes to each store a mix of products, sizes, and colors matched precisely to local sales. As a result, Mervyn's now sells more products with less inventory. Overall, the inventory handled by PSI (over half of Mervyn's total) has been reduced by over $100 million-worth of goods. At the same time, PSI halved the frequency with which goods are out of stock when a customer asks for them. The net result has been, even in a very difficult market environment, to add tens of millions of dollars each year to the bottom line.
What makes PSI really merit a prize, however, is the speed with which it was created, as well as the innovative ways in which developers and eventual users worked with each other - and with the technology. From conception to the end of roll-out took less than a year, including three months to test and improve the prototype as it was used to manage the inventories of a few selected items over the busy Christmas selling season. More broadly, Joe Vesce, who was chairman of Mervyn's when PSI was built, credits this rapid systems development process with creating a new company-wide culture of innovation, a new ease in working with technology, and a new excitement about business change.
In the past, Vesce argues, Mervyn's had been held back by its highly complex information systems. Since every part of the business depended on IT, every change initiative had to compete for me scarce resources available for improving computer systems. In effect, the annual meeting to set IT priorities became a sort of central planning session. As a result, nearly every scheme for significant IT-based change became a five-year plan, because that was how long it took to make large system changes. Step by time-consuming step, the process of defining needs in detail, then designing a full system on paper, then coding that system, and finally training users was much too slow to allow managers to respond to rapidly changing business needs.
PSI changed all that. Now systems are built much more quickly, changes in business processes move hand in hand with changes in technology, and IT managers are able to respond to rapid shifts in the business environment. This article describes how Mervyn's transformed both its inventory management system and its approach to technology-led change. To understand the lessons this experience offers, let us start by understanding why Mervyn's management felt they had a problem with inventory in the first place.
Systems follow strategy
The idea that inventory could most usefully be planned for each store individually, rather than for groups of stores, did not originate with the PSI project. Even before the project was conceived, Mervyn's had already committed itself to build systems with such capabilities. But enhancing inventory systems was only part of a $37 million, four-year project to improve all the company's information systems. In practice, PSI was born of the impatience of Mervyn's top executives - notably Joe Vesce, then chairman; Denny Chantland, chief administrative officer; Vivian Stephenson, chief information officer; and Dennis Green, VP of merchandise planning and logistics - to implement new sales programs.
In the summer of 1992, Mervyn's was in a slump. Its chief market, California, was in recession, and the company faced mounting pressure from new competitors. On all sides, the performance challenge mounted. An essential part of what Mervyn's had to do to meet this challenge was to make major changes in the way it managed inventory in order to reassert its competitiveness. What was needed was to reduce the absolute level of inventory and, at the same time, to cut down on the sales lost owing to goods being out of stock.
By managing inventory according to averages, Mervyn's had traditionally made it hard for any store to break out of the rut of past performance. To determine the mix of sizes and colors needed for any particular product, stores were lumped together into 8 to 12 categories according to sales volume. But because individual stores vary greatly in the sizes and colors they sell, these average allocations left some continually short of "smalls" while others could not get enough "larges." Similarly, some stores could sell lots of black jeans while the customers of others preferred traditional blue. Average allocations likewise failed to cope with surges in demand for advertised products - sometimes leaving stores short of the very items they were spending good money to promote.
Not only was the existing system wasteful; it also stymied any attempt to create more flexible and/or more customized services at each store. In other words, it precluded virtually any interesting retail strategy that Mervyn's might come up with. So, rather than wait four years to develop enhanced inventory management capabilities, Mervyn's top managers decided in July 1992 to preempt the usual kind of systems redesign project. What they needed was a computer system that could manage, store by store and item by item, the ordering, allocation, and distribution of all the different products stocked in Mervyn's stores. And they needed it fast.
If a working prototype of the system could be up and running by October, it could demonstrate its worth in the busy Christmas season. That would save money, and also help rally the troops to jump-start the change effort. So the top management group set an ambitious deadline: deliver a working prototype within three months - by October 1.
To drive the process, they assembled a core team composed of both developers and users of the new system: five were inventory planners, whose job it was to manage inventory from day to day - Kathy Ng, Grace Hardy, Connie Wolfe, Lisa Jenkins, and Terry Hall; and one, Matt Beall, was a systems developer from Mervyn's MIS department. The team also included a small group of consultants who were experts in retail strategy and logistics and in the business applications of IT. As a group, they possessed a deep understanding of the business coupled with a conviction of the need to try a new approach, despite its uncertainty. Together, the team was charged with responsibility for creating the prototype, connecting it to existing systems, and preparing for roll-out. But they did not have to seek approval from committees. If they had, they would never have hit their deadline. Things were hard enough as it was.
Building the prototype
The team met for hours with Mervyn's MIS department to discuss how to make the necessary changes to the existing mainframe-based inventory system. In the end, they determined that it simply could not be done. Reprogramming the mainframe would take an extremely long time; most in-house programmers had already been committed to other projects; and the MIS department was worried that any attempt to modify the current system might wreck it, bringing Mervyn's operations to a screeching halt.
Since they could not modify the existing system, the team tried a different tack. They decided to sidestep the mainframe. They would take raw sales and inventory data from the mainframe, calculate the new inventory requirements on personal computers, and then feed the answers back into the mainframe, where they would be acted upon just as though they had originated in the mainframe itself. In retrospect, this decision proved crucial to the success of the project: it allowed the team to modify only what absolutely had to be changed in the old system and, at the same time, to create the new one from a clean slate and with fresh technology.
Having decided to supplement what was already on the mainframe - or to "wraparound" the old system, as the team would later describe their idea - they faced two further problems. They had to figure out how to get data to and from the mainframe, and they had to build a system that would run on PCs but still do the work that the mainframe used to do. Accordingly, they sought the help of the MIS department to jury-rig network links able to move data from mainframe to PC and back again. To address the MIS department's concern that PCs would have difficulty coping with the sheer quantity of data that they were about to receive, the team bought three of the biggest and fastest PCs available. Even so, the demands they were making of the hardware meant that the team had to make frequent modifications to computers, networks, and the like.
In parallel with attending to technical infrastructure, the team also concentrated on getting the prototype to work. In doing so, they broke three cardinal rules in the systems development canon - not because they meant to, but because they did not have the time to do things any other way.
Breaking the rules
First, they did not worry much about what the system would look like to its initial users. Mervyn's MIS department, like most other groups of IT professionals, was accustomed to basing its systems design on the sequence of screens with which users would interact. The PSI team, by contrast, worried only about producing a system that could do the necessary calculations - no matter what the screens looked like. "We worked from the inside out rather than the outside in," noted one member. "We focused initially on the logic of the system: creating it, testing it against real data, and modifying it when we found ways to make it work better. Only later did we add a decent user interface."
Nor did the PSI team concern themselves much with bulletproofing the system against all possible errors, or with ensuring the robustness of the engineering design. Here again, time constraints forced them to concentrate solely on getting the system to work. To do that, they used two software tools with which they were already familiar: Microsoft Excel and Paradox. Although they knew that these tools were woefully inadequate as the backbone of a proper system for managing inventory, they were the only tools that the team knew how to make work in the allotted time.
Finally, the team deferred most requests from would-be users for extra features until the functionality and benefits of the core logic had been proven. Kathy Ng and Grace Hardy played a crucial role in determining which requests had to be dealt with urgently and which could be postponed for a few weeks (or, at most, months) until the crush of building the prototype had abated.
With hindsight, the team recognized that all of this rule-breaking had helped them enormously in building a system that would promote change. As one of them put it, "If we had the time to listen and respond to all of the requests for extra features, it would have been very tempting to build a system that exactly recreated the old way the job was done." Instead, they built a system that could evolve in response to the demands of the new ways of working that the system itself enabled. The team got away with all the rule-breaking for one simple but very important reason: they, the developers, were also the users. They knew what they did and did not need to make the system work. And they knew how to run it - with little or no interface or bulletproofing - because they had built it themselves. Collapsing the old distinction between users and developers proved extremely powerful.
Checks and improvements
In the first week of October, the team switched on the prototype; by mid-October, they were using it to distribute the inventories for 15 categories of goods (more than 1,500 SKUs). Even with the fastest PCs available at the time, the system took over five hours to chug through the necessary calculations for each category, and its user interface was so primitive that the slightest typing error could ruin several hours' worth of work. To guard against catastrophe, Kathy Ng and Grace Hardy carefully checked each distribution list produced by the new system - typically comprising 30,000 different shipment quantities - to make sure that its calculations looked reasonable, before allowing real goods to be shipped against its recommendations. They quickly realized, however, that the new system, clunky though it was, actually worked. It really could help Mervyn's sell more goods with less inventory.
After the basic concept of their system had proved itself, the PSI team spent November and December of 1992 improving it along two dimensions. First, they hired a contract programmer to rewrite the core of the system in C++, a far more suitable programming language for production systems. Not only did C++ give them a more robust system, it also cut processing time from several hours to about 30 seconds. And second, they worked to improve the prototype, adding new features and capabilities suggested by Hardy and Ng or by early experience with the new way of distributing inventory.
Everything was informal and collaborative. As Hardy recalls, "Somebody would have an idea and call together whoever else was around to ask 'What do you think?' We would change the logic and evaluate the results. Then we'd quickly decide 'Yes' or 'No' and go on from there."
Even as these new capabilities were being added, the prototype was being used to ship real inventories to real stores. The hard and risky work of testing the prototype during the busiest season of the year paid off: by the time the system was ready for deployment, both Hardy and Ng were thoroughly convinced - by their own experience - of its abilities and merits. And so were many of Mervyn's buyers, as well as the growing number of executives throughout the company who had worked with the prototype. This conviction would prove crucial during roll-out.
Building the system
By the end of November 1992, the prototype was ready to begin its metamorphosis into a fully-fledged production system. The new method of distributing inventory had fulfilled its efficiency-boosting promise, and the wraparound approach to adding PC-based capabilities to the mainframe had demonstrated its reliability, at least on a small scale. But two big problems remained: building a computer platform capable of managing thousands of categories of goods, instead of the 15 administered by the prototype, and getting Mervyn's analysts to adjust to the new ways of working that the system would require.
At the end of November 1992, while the prototype was being refined, Mervyn's MIS department created a team that brought together the people and skills needed to roll out the system. This led them into the realm of client-server technology, where the company had far less experience than it did with mainframes. Although many in the MIS department welcomed the challenge, others were skeptical.
Old concerns, new value
Not only did the prototype use different technologies - PCs and client-server technology instead of mainframes - it also approached systems development in an unconventional "build it, use it, fix it" fashion. As mentioned earlier, many of the familiar landmarks of orthodox systems development had been sidestepped in building PSI's prototype. More worrying still for MIS managers charged with keeping crucial corporate systems up and running was the concern that the technology of PCs could not provide such rock-solid data security and reliability as a mainframe could.
The vital task of responding to these concerns fell to Vivian Stephenson, Mervyn's CIO. She was in a difficult position. The things her colleagues were worried about were legitimate and important: integrity and security of corporate data and predictability in planning. Yet those concerns could not be allowed to slow the momentum of change. As she commented, "Joe Vesce, Denny Chantland, and I worked hard to make sure that everybody saw that we understood their concerns. But we worked even harder to show them how the prototype created new business value. That was how we demonstrated the benefits of this new way of doing things."
By January of 1993, the future of the PSI system was no longer in doubt. The roll-out team had selected all of the major hardware and software components that would make up the new system. The server had already been installed. Basic network communications were up and running, and the first training classes in the programming languages needed to develop the system took place during the first week of the new year. Most important, however, the documented performance of the pilot over the Christmas season had won the project the strong backing of key senior executives beyond its original sponsors. The questions facing Chris Ehemann and Walt Bizzare, who had been charged by Vivian Stephenson with deploying the PSI system, no longer concerned "if" or "what" - just "how."
Ehemann and Bizzare were chosen for this task because they had had the foresight to take courses in client-server technology. (Mervyn's had long encouraged its MIS managers to train themselves in new technologies that they believed to be important.) Even so, they realized that their level of expertise would limit the speed and efficacy with which they could deploy PSI. To meet a deadline every bit as demanding as that imposed on the prototype team, they needed to supplement what their own courses had taught them and what Mervyn's relatively limited client-server expertise could provide.
So Ehemann added expertise to his hardware and software shopping list. Some he found in technology consultants, whom he hired to help sort out the networking for the new system. In fact, although Mervyn's executives kept responsibility for leading and managing the project, consultants were involved in almost everything else - writing code, suggesting technical architecture, and configuring, installing, and debugging software products. In addition, much expertise was provided by the vendors of the powerful computers and software used to implement the system. Indeed, in Ehemann's eyes, a willingness to educate the company on how to use the new technology was almost as important in choosing a vendor as the capabilities of the technology itself. As Ehemann later noted, "The strategy enabled many of us to learn by doing, knowing that we could get someone to bail us out if needed."
But he wanted more than new technology. Because he recognized the success of the new way of working that created the prototype, he also wanted the process of deploying the system to be more cooperative and innovative. Instead of detailed planning and rigid, lock-step discipline, he hoped to encourage the same kind of continuous, iterative improvement that had characterized the building of the prototype.
As Walt Bizzare points out, this kind of cooperative development can be a delicate balancing act. In the case of PSI, the functionality required to achieve business benefits, which would please senior management, was delivered before robust ease of use, which would please the actual users. Along the way, Bizzare and the team often found themselves asking their colleagues to work with a system that they knew could be better - and would be better eventually, but not just yet.
By early February, the teams that built the prototype and managed the roll-out were merged, and, with the assistance of several MIS support groups such as database administration, technical support, and desktop technologies, a production version of the system had been implemented. Attention then shifted to getting managers to adopt the system and the new working methods that it embodied. Despite backing from senior executives and initial users, the system was not exactly welcomed with open arms by the managers who were to use it. It was not immediately obvious to them that it would make their lives easier. It required analysts to do more planning at the beginning of each selling season - about two to three hours for each category of goods they managed, instead of one hour with the old system. Although PSI promised to repay that initial effort by reducing the time needed to patch up crises later on, convincing busy, skeptical analysts of this payoff was no easy feat.
Analysts' worries were compounded by the disruption that the new system caused in the normal flow of orders to suppliers. For each category of goods switched over to the new system, PSI initially generated a huge order as it filled in gaps in each store's selection of goods in hand. Then it paused, waiting for the excess inventory generated by the old system to be sold off before creating new orders. Not surprisingly, the disruption raised eyebrows - and sometimes blood pressure - both in Mervyn's own ordering department and at some of its suppliers. With hindsight, the prototype team reckoned that they should perhaps have taken steps to ease the transition for suppliers. But at the time they could do nothing but press ahead.
The prototype proved crucial in overcoming doubts about the system. At one early meeting where the system was being introduced to prospective users, it met a barrage of criticism from managers. Some complained about the technology itself; others about the new ways of working that it involved. They were silenced by Grace Hardy, who assured the grumblers that she had already been using the prototype for several months and that it was well worth the effort. Without such experience, the arguments would have dragged on. As it turned out, the demonstrated benefits of the prototype enabled Chris Ehemann, Walt Bizzare, and Kathy Ng to turn their colleagues' energy from fighting the system to improving it.
Kathy Ng probably made the most converts. For several reasons - to get the earliest possible benefit from PSI, to make the most of scarce technical resources, and to encourage a new style of cooperative development - the system was deployed step by step, one category of goods at a time. Ng herself showed many of her colleagues how to operate and tap the benefits of the new system. As time went on, she encouraged others to do the same. The experienced taught the newcomers. Instead of requiring a lock-step, do-as-I-say training program, the PSI system became a catalyst for the sharing of experience and expertise throughout the company. Moreover, these discussions often generated additional suggestions for improvements to the system, many of which were incorporated in new releases.
If anything, according to Ng, the steps Mervyn's took to encourage cooperation worked too well. One of her main concerns during roll-out was trying to decide when to bring more discipline to the process of change management. It was one thing for a small group with similar experience and backgrounds to make changes quickly, because they all knew about the changes and could understand the reasons behind them. As the community of PSI users grew larger and more disparate, however, frequent changes threatened to become annoying instead of liberating. As soon as newcomers mastered some feature of the system, it might be changed, for reasons that their lack of experience prevented them from understanding.
But for Joe Vesce, Vivian Stephenson, Dennis Green, Denny Chantland, and other senior executives, such worries paled beside the impressive speed with which the PSI system was deployed, the benefits it produced, and the enthusiasm that it generated along the way. PSI was completely up and running by fall 1993, just a year after its conception, and Mervyn's mainframe-based systems development program was quietly scrapped to make way for a collection of projects more like PSI. Mervyn's had transformed both the way it approached change and the way it used technology. This, in the end, was the reason Mervyn's was selected for the Smithsonian award.
Looking back, former CEO Vesce says that he was surprised by how much PSI had managed to change Mervyn's approach to new projects. Although he had hoped for morale-boosting success and an effective new way to manage inventory, he had not expected the change to be so dramatic. Nonetheless, he and Vivian Stephenson quickly acted to make the best of a good thing and incorporate the lessons of PSI into Mervyn's normal way of doing business. These lessons included:
Business systems evolve; they do not spring full grown from managers' heads.
* Systems are most effectively built by improving upon a quick and simple prototype: "Build it, use it, fix it." Too much planning creates bureaucracy and often misses the point. It merely recreates the current poorly functioning system behind handsome new screens and interfaces. People cannot usually envisage all the things they will want the new system to do until they actually use it.
* Partnership between IT specialists and line managers is crucial to the development of prototypes, which can capture - and iteratively improve upon - not just computer systems, but the complete business processes that they enable.
Risks are worth taking, even in information technology.
* The risks of introducing an IT prototype can be mitigated by an approach that allows work to be handed back to the old system if the worse comes to the worst.
* The risk of imposing a short, "stretch" deadline is counterbalanced by its ability to focus a team's efforts on what is absolutely essential.
* Client-server technology enables specific functions to be removed from the mainframe and installed on something more up to date - without managers having to completely replace the whole of the mainframe system. This wraparound approach can dramatically speed systems development and deliver real business benefits quickly, while allowing old systems to be phased out step by step. It also encourages continuous improvement.(*)
An emphasis on results - rather than routines - promotes change.
* A proven prototype is the best possible argument for the adoption of a new system and a new way of working. It also provides a crucial focus for discussions aimed at continuous improvement of the system.
* Rapid change entails giving real decision-making responsibility to the team leading the change. If the team that built Mervyn's prototype had had to consult with committees of users and IT experts, they would probably still be in meetings.
* Success should be celebrated. In an environment of continuous change, where things could always be better, it is important to remember that reward and praise encourage renewed effort. Constantly pointing out how much farther there is to go can quickly become discouraging.
To break down the barriers between her department and the rest of the business even further, CIO Stephenson has changed her hiring policies to give more weight to candidates who understand business processes, not just technology. She has also led a migration to client-server technology and is organizing systems development in small fast-moving teams that include both IT professionals and line managers. Meanwhile, her colleagues are more than happy to meet her half-way. As Dennis Green, head of logistics, points out, the idea that IT can be a force for change has come as a welcome surprise to Mervyn's line managers. Instead of struggling to get a new idea past committee after committee, they can simply take it and try it out. Entrepreneurialism may yet replace bureaucracy, after all.
As for PSI itself, the system that sparked off all this change continues as it began: constantly improving. At the end of 1994, it was being enhanced to handle "seasonal basics" - items that are not kept in stock year round, but vary from season to season. A development process similar to that of PSI has been used to create another new system to handle the remainder of Mervyn's inventories, so-called "fashion goods," which are rarely restocked after being introduced. What comes next will be up to the employees who use the new systems. But whatever it is, it will happen fast.
* See Derek L. Dean, Robert E. Dvorak, and Endre Holen, "Breaking through the barriers to new systems development," The McKinsey Quarterly, 1994 Number 3, pp. 3-13, and Richard Heygate, "Avoiding the mainframe trap in redesign," The McKinsey Quarterly, 1993 Number 3, pp. 79-86.
RELATED ARTICLE: A CHRONOLOGY OF THE PSI SYSTEM
Mervyn's commits $37 million to rewrite key business computer systems, including inventory management. The project is scheduled for completion in 1996.
Implementation of Mervyn's new strategy is thwarted by limitations in inventory systems.
Mervyn's undertakes to develop a working prototype of a new inventory management system by October 1, so that it can be tested against Christmas sales.
Prototype team forms: 5 in-house "inventory planners," 1 in-house MIS systems developer, and a small group of outside experts.
After brainstorming with Mervyn's MIS department, team chooses "wraparound" strategy for integrating new inventory control system with existing systems.
Using Microsoft Excel, team develops new logic for system.
Team builds interface to mainframe and develops techniques to clean up and use data on personal computers.
New distribution rule written to allow mainframe to use PC output for real inventory distribution.
Prototype goes live on October 1.
Prototype manages 15 categories of goods (more than 1,500 SKUs) for Mervyn's Christmas season.
November 1992-January 1993
Constant iterative improvement of prototype.
Sales of PSI-managed products increase even as inventory falls.
Roll-out team formed in late November.
Selection of hardware and software for production system and design of client-server architecture.
Server and basic networking capabilities put into place.
Prototype team and roll-out teams merged in early February.
Merged team leads build-up of production system and roll-out of new business process.
System roll-out and inventory planner training completed.
System improvements identified and incorporated into new releases.
Bob Dvorak is a principal, and Derek Dean and Marc Singer are consultants in McKinsey's San Francisco office. Copyright [C] 1994 McKinsey & Company. All rights reserved.
|Printer friendly Cite/link Email Feedback|
|Title Annotation:||Delivering the Value From IT; information technology|
|Author:||Dvorak, Robert; Dean, Derek; Singer, Marc|
|Publication:||The McKinsey Quarterly|
|Date:||Sep 22, 1994|
|Previous Article:||The business promise of 'hub-and-spoke' systems.|
|Next Article:||Being intelligent about 'intelligent' technology.|