Field Report: Avnet, Phoenix
Streamlining the bid process brings huge scalablity gains.
Semantic integration may sound like highfalutin theory to some, but by making sense of millions of cryptic requests for bids, semantic integration is adding millions of dollars to the bottom line at electronics giant Avnet. The $10 billion distributor has some 2.5 million items in its master catalog, selling parts and components from more than 280 suppliers to thousands of electronic goods manufacturers. The challenge? Customers submit two million line-item requests for quotes per month, and they do so in a jumble of formats, including XML files, spreadsheets, delimited files and even Word documents. Given the volume, Avnet has to rely on computer interpretation.
"I've been involved in the quote process for eight years, and when we started out, we had a match rate of only 6%," says Beth Scheer, a vice president of development at Avnet. "Over the years, we've built a rules-based approach to automate the matching, but we've been stuck at a 50 to 55% match rate and exhausted our ability to build more rules."
If Avnet had an army of clerks, it could probably match every incoming request for a bid because the electronics business, like most industries, has a language all its own. Manufacturers use consistent part-numbering schemes that the trained eye can break down into manufacturer names, voltage, wattage, case sizes, capacities and so on. A full-time business analyst and a full-time IT developer spent seven and a half years methodically developing rules to interpret bids automatically. But with so many suppliers and bids, Avnet was hitting a wall it couldn't scale with more rules.
When Scheer saw a demonstration of semantic integration technology from Silver Creek Systems, she realized it could solve the matching problem by literally "understanding" incoming bids. "You can train this system on the semantics of the elements within part numbers, and it can figure out the right part number based on a natural-language description," she says.
Silver Creek's Data Refraction platform provides "intelligent" technology that can be trained to understand at a semantic level. The system is used primarily on large volumes of structured and semi-structured information, and it can interpret long, textual descriptions, abbreviations and a mix of descriptive attributes.
Avnet implemented Silver Creek's software last February, and within 90 days, four full-time pricing and parts analysts and six others working part time had trained the system on the distributor's highest-volume product lines. "Our match rate is now at about 62%, but as we do more training and it learns from exceptions, we're going to get incremental increases," Scheer says. "The initial goal was a 75-percent match rate, but we now think we can get to 85%."
Avnet spent about $500,000 on the software and $200,000 on development and implementation, according to Scheer, but the investment has been well worth it. "If we can't match the request, we won't quote, and that means we don't get the business," she says. "My quote-to-win ratio is at 22%, so if we can quote that many more lines, I can win that much more business."
Silver Creek's software has not only increased Avnet's match rate by 7% to 12% thus far (which translates to 140,000 to 240,000 more line-item quotes and 30,000 to 50,000 more orders per month), it was also used to support the acquisition of a rival distributor. "We took [the acquired company's] master part catalog and ran it through Silver Creek's software to support the system integration," Scheer says. "We came up with a 75% hit rate in what otherwise would have been a manual process of mapping part numbers."
Semantic integration is a compliment rather than a replacement for conventional methods such as extract, transform and load (ETL) and enterprise information integration (EII), according to Barbara Mowry, CEO of Silver Creek. But it fills a crucial gap in "making information usable the way people need to understand it," she says. "You'll still need to map, share and move data between systems, but all those activities work better if you can put the information in context."
— Doug Henschen
About the Author
You May Also Like
2024 InformationWeek US IT Salary Report
May 29, 20242022 State of ITOps and SecOps
Jun 21, 2022