Workflow by Example (WbE)

Overview

Workflow engineering is a complex and expensive activity. Adding a new workflow to a system often requires lengthy and repeated rounds of software engineering. The time and cost of this work results in a barrier between users and new system functionality. End users can be given the power to write workflow scripts on the desktop, but the large amount of expertise required by the user to implement even the smallest result makes this solution viable only for a small class of users.

Instead, WbE allows users to create workflows by providing an example to the system. Using machine learning, the system constructs a workflow corresponding to the example automatically. The user then provides a set of examples. WbE executes these examples against the workflow and notes any exceptions. The user then repairs these exceptions by providing additional examples. Eventually, using this process, the user creates a complete workflow without understanding arcane scripting languages. In controlled behavioral studies, WbE has been shown to be fast and less error prone than the manual execution of updates.