Stop saying, "That's what I should've done"
Real-time performance information helps, but must be carefully filtered for relevance
By Tony Baer, senior contributing editor -- Manufacturing Business Technology, 3/1/2005 12:00:00 AM
In the late 1990s, DaimlerChrysler's MOPAR Parts Division, Centerline, Mich., updated its systems for demand forecasting, dealer demand, global parts ordering, and distribution resources planning. As the sole supplier of parts to dealerships, the unit processes roughly 200,000 order lines daily. But with all that data, the company needed a way to see the forest through the trees when problems such as inventory imbalances or delivery bottlenecks occur.
"It's not like we've ever lacked reporting systems," recounts Jerry Quell, senior manager of inventory and planning. "But when issues cropped up, our people relied on ad hoc reports."
With its new supply chain systems in place, MOPAR took a chance on the next step: investing in a performance management system from a relatively new vendor, SeeCommerce, to help it isolate the information needed to manage by exception.
Marrying business intelligence (BI) capabilities with event-detection techniques borrowed from the process control world, the SeeCommerce system offers both a data warehouse and the ability to monitor streams of events captured as transactions. Thus it can be used to do traditional look-back analyses of past patterns, and also supports an event-based workflow engine for analyzing disruptions as they occur.
"Using business intelligence to make current operational decisions is a departure from traditional systems that deliver reports from a data warehouse," explains Henry Morris, a VP and analyst with International Data Corp. (IDC), Framingham, Mass.
While transaction systems for ERP, CRM, or supply chain execution operate as systems of record—reporting "what" an enterprise is doing—business intelligence answers questions such as "how" and "why" through reports garnered from production systems or analyses that are run against data warehouses. In most cases, these analyses look at past performance, using historical trends to gain insight that can be applied here and now.
It's better to supplement historical analyses by intercepting exceptional events as they happen.
"I recall touring a Ford engine-lock facility and saw SPC [statistical process control] charts on the wall that must have been up there for 30 years," says Scott Hicar, CIO and VP, worldwide IT, for disk-drive manufacturer Maxtor Corp., Milpitas, Calif. "What's new is to marry SPC techniques with business activity monitoring or workflow to selectively alert an organization to a critical operational issue."
Supply chain performance monitoring lends itself especially well to such an event-based, process control-like approach. "What's unique in the supply chain is that as you implement plans that look out in the future, you can get a view of events that provides the opportunity to avoid issues before you have to bump up against them," says Stephen Brown, Toronto-based supply chain planning lead for New York-based Deloitte & Touche.
A.O. Smith Water Products, Ashland City, Tenn., sells millions of dollars worth of water heaters to homes and commercial buildings, and uses a Logility supply chain performance management system daily to compare demand forecasts with sales, production, and inventory. According to Jim Hutzel, logistics and distribution director, the solution not only nips potential problems in the bud, but also acts as a time-saver. "The nice thing is that the system is exception-based, so you don't have to mine data out of it," he says.
From dashboards to analysis
Today, business activity monitoring (BAM) dashboards display corporate performance parameters that, in most cases, are derived from near-real time information in data warehouses. Recent start-ups such as Celequest Corp. apply streaming technology to intercept live transaction data to populate the dashboards. Nonetheless, while BAM tools show how an organization is performing in the here and now, they don't necessarily answer the "why" questions.
At the other end of the spectrum, most business intelligence and enterprise application vendors support the ability to marry data warehouse queries with drill-downs to transaction systems that add current context to trend reports or analyses. Nonetheless, adding drill-down capabilities to production systems isn't the same as proactive approaches that publish analyses as current events unfold.
In most cases, the hurdle has little to do with whether a system can intercept strictly live production data in real time. The technology for trickle-feeding data warehouses with near-real time data from production systems is well established, allowing some data warehouses to cover recent events up through the previous hour. However, for manufacturers, these capabilities are probably overkill outside the plant floor because, at most points—from upstream product development to inventory management and logistics—events do not require split-second reactions.
"Most of the time, the value proposition there is to be alerted daily, so you don't have to wait until the end to find out your month is toast," says Tom Kozenski, distribution product leader at supply chain execution vendor RedPrairie.
Naturally, there are always exceptions. If an out-of-stock condition exists in the distribution chain for a shaving product during its promotion the week of the Super Bowl, that might merit an alert within hours, especially if there's a possibility of diverting in-transit merchandise to meet immediate needs.
Consequently, the success of an event-based system—such as an analytic or reporting solution—depends on what you define as an "event," and then determining the speed of its impact. For instance, a demand forecast may be set in stone a month in advance. But if it misses the mark, it could have near-real time impact, if poorly planned production or out-of-stocks cause a delivery to Wal-Mart to fall out of its expected six-hour window.
Nonetheless, as any veteran of process control or RFID pilot projects can testify, as you sample data more frequently, data storage volumes, management burdens, and costs snowball. "It would be nice to get BI alerts in hours," says Deloitte's Brown. "But my concern is that you could end up drowning in way too many alerts."
Filtering out the noise
The key to managing event-based business intelligence is to take a cue from the process control world: devise a means for eliminating nuisance alerts, so your screen isn't constantly bombarded with alarm reports. According to A.O. Smith's Hutzel, this is a process that requires serious trial and error.
"You have to look at it item by item, and product family by product family, to find out what works," he says. "In some cases, I might want to know about a 10-percent deviation from forecast, while in others, I don't want to hear about it until it reaches 50 or 60 percent."
The variations also could extend to customers, freight carriers, or suppliers. Are all of them equal, or do some play more important roles or rate higher priority than others? Depending on the complexity of your business, thresholds could vary widely.
And then the question arises, is the variation situational? Take the example of a late delivery that was held up because of an accident that tied up I-95 in the Baltimore-Washington area. Was this event a freak occurrence or part of a larger pattern that until now has been just under the radar? Maybe that highway is so prone to accidents and backups that logistics planners should avoid the route entirely, or maybe the root cause is the marginal performance of the freight carrier, whose deliveries typically border on lateness under normal conditions anyway.
At A.O. Smith, the Logility system picked up such an exceptional event in the first few months after going live in 2004. Detecting a spike in demand for 75-gallon water heaters—which evidently are popular with homeowners installing hot tubs and spas—the result would have otherwise been obscured by overall sales figures indicating only modest growth for the year. A further analysis revealed that the sales spikes were largely concentrated along the east and west coasts, enabling the company to identify the new market opportunity and adjust production and distribution plans to best capitalize on it.
In most cases, customers using BI, scorecard, or other reporting or analytic solutions to help them manage supply chains have not gotten to the point of figuring out all the rules or workflows, or in many cases, where to even start.
"In most cases, customers set key performance indicators [KPIs] and thresholds, and may have a modest amount of analysis performed when an exception occurs. But it's pretty unusual to have an expansive decision tree to look for unexpected events," says Paul Rodwick, a VP with CRM supplier Siebel Systems.
Similarly, says Maxtor's Hicar, don't attempt a build-it-and-they-will-come strategy. If your business intelligence system can isolate current trends, it takes trial and error to determine if they truly add value, or simply add to the clutter. "Expect the process to be iterative. Don't try to build the perfect application from the start."
MOPAR's Quell echoes that sentiment. "We struggled a bit to define the rules for what constitutes an alert. Unless you know your business perfectly, it's a trial-and-error process."
Furthermore, because automated workflows change the way people work, people can easily get overwhelmed with the new powerful tools at their disposal.
"You can't do enough preparation," Quell says. "You have to get people thinking ahead of time what they could do with the information, and you have to do a lot of role playing." Quell adds that over the five years that the system has been in production, usage has grown from an initial core group of six to eight people to roughly 75 to 100 people now.
Admittedly, not everyone is using the system's proactive reporting or analysis functions. "It takes getting used to," maintains Quell. "This isn't ad hoc reporting. It's all about how you run your business and set your goals. This system proactively tells you when you are not meeting your goals."
In MOPAR's case, the system certainly has helped out. Since first going live, order fill rates have improved by 2 percent. With a volume of more than 55 million order lines per year, that has translated to millions of dollars in annual savings.
