Skip to main content

Auto-Reply Escape/Escalation Feature

Problem: Auto-reply isn’t going to detect every question correctly. I would still prefer to let it try to answers questions and miss sometimes, rather than have to review&approve messages or review after the fact to see if the guest’s question/issue was resolved.

Solution: At the bottom of auto-generated messages, there should be a mechanism for a guest to indicate if the question was not resolved and if they need additional help from a human, the guest should be able to reply to messages with a specified keyword, (For example, ‘HELP’) that could then trigger an alert/email/notification so that I would know I need to jump in and answer the question or address the issue

Status: Rejected5 comments

Log in to comment and vote

Comments5

  • Hospitabase

    Team•

    Nov 22, 2024

    Hey @Rick Stevens ! This is a cool idea! I’ve seen some other hosts simply add a line in all of their messaging rules which are in response to Questions, something like this. This would essentially lead to the same end result you’re hoping for, where you’ll get the message that they still need help and be able to reply :)

    I hope that helps in the meantime!

    • Rose Stapler

      •

      Nov 22, 2024

      Thanks for the suggestion!

      In the example above, is there a way to build a notification or workflow off of the “I still need help” response to alert the host to an urgent request?

      The goal here is to separate signal from noise in the communication flow and to ping/notify the host when there is an urgent need from an active guest rather than just a regular message from a prospect or other non-urgent question.

      • Hospitabase

        Team•

        Nov 22, 2024

        Not today, but we want to be able to leverage AI to detect urgency / negative sentiment in conversations to help you prioritize conversations, so that is absolutely something we’ll continue looking into with your use case in mind!

        • Rose Stapler

          •

          Nov 22, 2024

          AI would be a nice to have, but I’d still position that having an explicit user action would would be a more robust solution.

          An explicit action could be complementary to the AI sentiment analysis as a fallback option when the sentiment scoring is below the confidence threshold.

          Having that explicit user action will also give you training data on if the AI correctly identified & routed the user question. Rather than “Help”, you could ask the guest to reply Yes/No on if their question was answered.

        • Rose Stapler

          •

          Nov 22, 2024

          Just saw you marked this as ‘Rejected’

          Super disappointing. That definitely colors my desire to license the platform.