Before the technology bubble burst, lots of people were saying 3Com Corp. CEO and president Bruce Claflin was playing it too safe. Analysts thought he was sitting on too much cash. An executive in the broadband modem group told him forecasted demand was five times too low. Even Claflin's wife wanted to know why he was so worried when smart guys like Cisco Systems CEO John Chambers were feverishly expanding.

Don't hail Claflin as the new Nostradamus. Though 3Com was among the early batch of tech vendors to call the economic slowdown, the networking company has missed even its most conservative sales estimates, which forced it to write off almost $19 million in inventory last quarter and cut staff by half since last year. What Claflin got right was understanding just how bad companies are at forecasting--and not betting his business on it. "Businesses have largely discarded the idea that they can accurately forecast," Claflin says. "They still do it, but they know they can't rely on it."

That may be too harsh, but poor forecasting is one factor behind efforts at better visibility and quicker reaction in supply chains (see story, "In Search Of The Big Picture: Supply Chains"). Still, the advantages of even a slightly better glimpse into the future are so compelling that managers haven't given up trying. Of course, no one expects to be able to predict horrors such as the terrorist attacks of Sept. 11 or natural disasters such as hurricanes and earthquakes. Likewise, picking broad business-cycle and stock-market trends is a brand of black magic best left to economists and statisticians. Instead, what businesses and IT managers deal with is referred to as demand forecasting--collecting information from customers, suppliers, and employees and combining it with historical sales and economic data to predict near-term demand.

Managers confront the problem with a quiver of applications and systems, eyeing the goals of pulling the right information from the reams of data available inside and outside the company, putting it into the right historical context, helping managers use the results to determine where business is headed, then showing what future scenarios might look like. Desktop software such as ForecastPro from Business Forecast Systems Inc. applies various forecasting methods to historical data. For more analysis, managers can turn to business-intelligence tools from vendors such as Business Objects, Cognos, Hyperion, and SAS Institute, and applications or databases from companies such as Oracle to help sort and present data. Sales-force and customer-relationship management software also plays a role, letting managers draw forecasts out of employees who are closest to customers. There are built-in demand-management apps in enterprise resource planning systems such as SAP's and supply-chain management tools from companies such as i2 Technologies and Manugistics. Of course, there's also the humble spreadsheet, which, despite its limitations, remains a widely used forecasting platform.

Ask Rose Melillo, financial systems manager for Murray Inc., a Brentwood, Tenn., manufacturer of lawn mowers and snowblowers. Last year, forecasting at the more than $700-million-a-year company amounted to weekly meetings of managers of six divisions, each with his or her own report and forecast. "Everyone came to the meeting with their spreadsheets and talked about what's going to happen," Melillo says. "And everyone's spreadsheet had a different answer."

Now Murray can consolidate the managers' sales and production forecasts as often as every week, thanks to a new 18-month rolling-forecast system. For the past few months, the company has run the process on a business-intelligence program from Cognos Inc. called Cognos Finance. Every division can access and work from it, giving the company one set of data with which to generate forecasts. The system lets Melillo load sales data directly from a Coda Financials general ledger system and production data from a Cincom manufacturing software package into an Oracle database on which Cognos runs. Murray can bring in up-to-date results on sales and manufacturing costs as well as outside data on factors such as the weather and consumer confidence. That helps the company decide, for instance, whether to continue building consumer products with more volatile demand, such as minibikes and all-terrain vehicles, in this slower economy. "It's a way to put our managers' ability to forecast together in one place and see the results," Melillo says.

Forecasting often suffers from poor data: not enough, not the right stuff, too old, or not presented in a relevant way. IT managers and vendors are working to improve this, but even the best data is subject to human judgment, the irreplaceable process of building an assumption about where the business is headed. "Forecasting is highly dependent on initial assumptions and data collection, and both of those are fraught with problems," says Joshua Greenbaum, a principal with Enterprise Applications Consulting. "But we're getting much, much better." Greenbaum points to improvements such as templates for analysis built into software, which reduce the formula errors that come from building massive spreadsheet models in-house and also make clearer the assumptions on which a prediction is based.

Carl McKenna knows how hard it is to collect the right data for an accurate forecast. The manager of marketing IS for Kennametal Inc., a $1.8-billion-a-year maker of metal-cutting tools and machines, picks sales information from multiple ERP and financial-reporting systems and draws historical-comparison data from Kennametal's data warehouse, which runs on an Oracle database and uses SAS Institute's business-intelligence tools. The Latrobe, Pa., manufacturer recently started using Ockham Technologies Inc.'s Sales Razor sales-force management tool to get forecasts from sales staff more quickly and more often. Putting all the pieces together is critical to a solid forecast. "Without one single data picture, your level of success will be affected," McKenna says.

Collaboration and extended supply chains make forecasting more difficult and increase the potential for miscommunication, excess inventory, and overcapacity. That's because suppliers often don't have the information they need to make accurate forecasts. An InformationWeek Research study earlier this year found that six out of 10 companies don't share sales forecasts with suppliers, and just over half do any collaborative planning. Tim Cuny, a principal for Lante Consulting, says that cagey approach to sharing information helps explain the multiplying errors created in collaborative forecasting. "It's rarely a two-way street, because business isn't a very trusting process," Cuny says. "A lot of this is relationship management more than technology management."

For forecasting flops, it's tough to top the IT and telecom-supply industries. When the sudden spending slowdown hit late last year, manufacturers' forecasts went out the window, leading to huge costs such as Cisco Systems' $2.2 billion write-off of excess inventory in April. Intel repeatedly revised forecasts--in March, about six weeks after it predicted a 15% sales drop from the prior quarter, it concluded sales would actually fall 25%. But since spring, it's largely met its lowered sales targets.

Howard Woolf, sales and services president for Intel's computer-telephony products group in the Americas, doesn't work on companywide forecasts. Instead, he keeps his eye on one piece of information that's part of every forecast: the salesperson's estimate of what deals are in the pipeline. Woolf says his group has sharpened its skills to where forecasts vary less than 10% from reality--barring unforeseeable events. One reason: a better IT tool. Less than a year ago, Intel used spreadsheets to manage its sales staff's forecasting data. By switching to sales-force management capabilities from Ockham Technologies, Woolf can look at how a forecast--for a single deal, one salesperson, a region, or the whole group--changes over time. That lets him spot people or areas that are consistently inaccurate and figure out whether they're overestimating the potential or having trouble closing good prospects.

When companies ask their sales forces about deals in the pipeline, the information is filtered through each salesperson's sense of optimism and judgment. "One man's 70% is another man's 30%," Woolf says. So Intel defined the terms for salespeople: A 10% chance for a sale is a funded opportunity with no idea whether Intel's in the running; 50% is certainty that a purchase will be made and Intel's a front-runner with a 50-50 shot; 90%, you've got the deal and are waiting on paperwork. The Ockham tool lets Woolf identify deals that aren't progressing before an opportunity is lost. "Forecasting isn't my goal," Woolf says. "It's a means to increase sales."

Technology isn't going to replace judgment when it comes to forecasting. What managers need from forecasting tools is better information that gets them closer to understanding what's happening in real time and lets them quickly cast scenarios into the future. Then there will always be that mo-ment when managers put all the data together and run it through their experience and gut-level instincts to make the call about whether a trend is headed north or south.

Illustration by Richard Downs