Answering Questions via Expert

Answers are long-lived. To ensure your answers age well, follow these best practices.

Make Answers General

When answering a question as an expert, remember that anybody else might ask the same question in the future. Answers that are too specific might not be relevant to someone else asking the same question. Answers should be as general as possible and take into consideration that

  • the person asking the question may be different - avoid using information specific to one individual;
  • the time or date may be different - avoid answers that assume a specific time or date;
  • the place or location may be different - avoid answers that assume a specific location.

For example, in the below conversation, John responded with an answer that assumed a specific date. If someone asks the same question tomorrow, the answer will no longer be valid.

An answer that assumes a date

Use Self-Contained Sentences to Account for Language Variation

People ask questions differently. Account for this in your responses. Avoid starting answers with “yes” or “no”, since the answer depends on specific language structure. It will not make sense in other formulations. For example, the following sentences can all be answered with “yes you can”:

  • Can I work from home today?
  • Is it okay to take an extended lunch?
  • Can I expense my mobile phone bill?

But this same response sounds strange if someone asks the question differently:

  • Any issue with me working from home today?
  • I need to take an extended lunch
  • Is there a policy for expensing mobile phone service?

A better approach is to use complete sentences that are self-contained. These sentences stand on their own and make sense regardless of the context. Here are some self-contained sentences that work for either sentence variation:

  • You can work from home assuming you have cleared it with your manager in advance.
  • If you need to take an extended lunch, please ensure others know when to expect you back.
  • If more than half of your phone usage is work related, the company will expense it.

Account for Incorrect Responses

Expert isn’t perfect. Sometimes it will respond incorrectly. Responses that are too specific and assume perfect classification look funny when they are incorrect. Self-contained sentences are less obviously wrong. Some users may not even realize there was a misclassification.

In the below example, the response starts with “I understand.” This clearly misclassified the question. However, if the response omitted the “I understand” initial text, the response is passable since it is related to music. So the initial text prevents us from suspending disbelief because it is too obviously out of context.

Awkward answer due to misclassification

If you treat Expert as a content management system, it will be difficult to manage the content. Expert works best as a concierge or librarian. These people don’t provide comprehensive information. Rather, they point askers to a source of information. Where possible, answers should point to existing informtion in your portal, shared drive, etc. Doing so ensures Expert’s answer remains valid for a long time.

Expert understands HTML, so you can add HTML links in your responses. Do this in favor of raw URLs.

Use Data Where Possible

If you want to provide personalized answers, use a data integration. Expert integrates easily with Google Sheets. With data integrations, Expert can use the user ID of the asker as well as named entities to lookup information using our powerful Pez formula syntax.

If you don’t have access to data integrations, you can use workflows to provide personalized responses. Go to the Train page in the dashboard and edit the interaction directly.

Question Queues

The behavior of question queues is changing. Instead of per user, they will be per topic. This change will improve response time for askers.

Unanswered questions are called open questions. New questions are routed to the appropriate SME or fallback expert and added to their question queue. Questions are answered according to first in, first out (FIFO) priority.

Expert expects that the next statement given by the human expert is the answer to the open question. When an expert answers the question, Expert removes the open question and records the QA pair in its lexicon.

New questions are added to a topic question queue

Fallback Selection

When there are multiple fallbacks, Expert looks for all fallbacks without open questions. If there is just one, Expert will notify the available fallback. If more than one fallback is available, Expert will pick one randomly.