Jump to: navigation, search

Design

This topic describes part of the functionality of Genesys Content Analyzer.

In designing your category trees and standard responses, remember that they will have two very different groups of “users”—that is, agents and training.

  • Training uses the categories plus categorized e-mails to generate models.
  • Agents use categories in two quite different ways:
      • They use the categories to find standard responses.
      • They give feedback on the category/standard response system, essentially indicating, “Yes (no), the standard response of this category is (is not) a good match with this e-mail,” affirming that this e-mail should/should not be tagged with this category. This tagging becomes one of the attributes of the interaction as it is stored in the Universal Contact Server database.

Important
Agents can use standard responses without giving feedback, but if they do not give feedback you cannot collect enough categorized emails to be useful for training. You then have to create e-mail manually from Knowledge Manager’s Training tab.

Given the importance of high-quality feedback, you may want to designate a special group of agents for this purpose: define the categorizing of interactions as one of their main duties. Remember: the more categorized e-mails you have and the more accurate the categorization is, the more likely the system is to produce accurate models

In designing your category trees and standard responses, keep in mind the following:

  • Do not create too many categories. Many categories allows for many standard responses, and if there are large numbers of standard responses agents are likely to use some responses very little or not at all. This creates the following chain of causation:
    1. There are very few e-mails tagged with a particular category.
    2. The system cannot train for that category.
    3. The system cannot suggest that response.
    4. That category and its response continue to be used very little.
      In short, excess categories are likely to not be used.
  • Try to make categories sufficiently distinct. If two or more standard responses apply to very similar situations, training has difficulty producing a model that can tell them apart.
  • Avoid categories/responses that are too general, like “Not enough information.” Agents will use only one or two such general responses and ignore any others, with two undesirable results:
    • Training has a hard time producing a good model because the e-mails it uses have a huge variety of content.
    • The system is unable to include the unused categories/responses in training, because there are very few e-mails tagged with those categories in the database.
This page was last modified on December 17, 2013, at 11:54.

Feedback

Comment on this article:

blog comments powered by Disqus