nexpo'96 POST-PRESS
Machine Manages Mailroom

For an average Sunday edition, 30 million free-standing inserts of up to 70 different sizes, weights, thicknesses and paper types arrive at The Washington Post. The job of collecting, counting, checking, storing and retrieving these preprints poses a massive challenge. Drawing up a plan for sorting them into bundles tailored for 180 zones, using a wide range of equipment and a variety of workers, and then subdividing them into carrier routes and delivering them to the right people taxes the limits of human ingenuity. Handling five million to eight million daily pieces requires a great deal of analysis, too.

Post managers wondered if they could assign this mind-boggling process to a computer. So last year the newspaper joined The Boston Globe and IBM's Integrated Solutions Systems Corp. (White Plains, N.Y.) to apply artificial intelligence to setting up and managing post-press operations.

During a NEXPO session, "Post-Press Packaging and Distribution," William J. Gard, director of production engineering at the Post, explained how the quest for an Integrated Mailroom Management System is proceeding.

First, managers identified six key steps in the preprint process. Next managers located "domain experts" who understand each step. These experienced hands helped establish rules, or knowledge-based ASCII statements used by the AI program to approximate human decisions. Each rule carries a peg, or priority, assigned in descending order.

With these factors entered into the program, one rule, "assign hopper for odd-page combo," and its peg, "try to assign FSIs with odd (strange) page combinations to end of batting order," ranks lower than another rule, "assign hopper to TV," and its peg, "try to assign TV book to hopper 20."

Gard explained that applying math models and heuristic algorithms to the rules and pegs can produce useful schedules that assign jobs to shifts, determine the type of machines and labor for each shift and present the plan to users for review and modification. If part of the plan is unworkable, a human expert may override it. When overrides become continuous, Gard noted, that defines a new rule.

As of June, the Post was testing computer-generated schedules and looking forward to the next step of integrating with other newspaper software.


TechNews Volume 2, Number 4: July/August 1996
Return to July/August Home Page
Return to TechNews Topic Index

©1997 Newspaper Association of America. All rights reserved.