Skip to content

Did the bot make a mistake? Where to start

First, let's agree on what an error is. A conversational AI doesn't follow a script, and not everything that surprises you is a mistake.

What is NOT an error

  • Different words for the same answer: the same question can get different phrasings. If the content is correct, it's fine.
  • An extra confirmation on ambiguous data: this is prudence, not confusion.
  • A reformulation when the question was vague: it's trying to understand, just like a person would.

What is an error

  • Wrong or invented information, or a promise it shouldn't have made.
  • An "I don't know" about something that is in your sources.
  • Data already provided requested again in the same conversation.
  • Two identical requests, different outcomes: same intent, same information, once correct and once not.
  • A tone that is off-brand compared to how you want to speak to your customers.
  • A malfunction: it doesn't respond, it cuts off, it doesn't connect the call.

Everything in the second list can be fixed. The pages in this section will lead you to the right correction based on three questions.

A wrong answer almost always has a simple cause. Three questions are enough to find it and understand who needs to fix it: you or glacom®.

The three questions

1. Did the bot respond?

  • No, it stayed silent or gave an error → it's a malfunction: report it to glacom®.
  • It said "I don't know" → the information is probably missing: go to question 2.
  • It responded, but incorrectly → go to question 2.

2. Was the information in your content?

3. Was the source updated and well-written?

What if the problem is the tone?

Correct content but wrong form (too cold, too long, wrong language): see Correcting the tone. On the phone, it has its own rules: Correcting the callbot.

Limits

  • This tree covers common cases. Strange cases (happens once in a hundred, only on one channel, only at a certain time) should go straight to Report it to glacom®, with as much detail as possible.

From the app

When you read a conversation from the app, you can rate the wrong answer: the report starts from there.