Genesys Cloud - Main

 View Only

Sign Up

  • 1.  Concierge Bot / Clarifier Bot AI Guide

    Posted 3 hours ago

    Hey GENESYS Orchestrators,

    we are currently working on building a Concierge Bot / Clarifier Bot to handle unstructured customer requests. In many cases, customer descriptions are extremely vague. A human agent would normally walk through a decision tree, asking clarifying questions until the issue is narrowed down.
    We want the bot to take over this pre-qualification step so that the agent receives a well-defined problem context - improving routing, article accuracy, and matching quality.

    Below is our current setup versus the ideal setup we used to have, which we are struggling to reproduce with the new AI Guide.

    🔧 This is our current setup (and it works), but it is too limited:

    OVERALL GUIDE GOAL – STRICT

    Your only task is to prepare the handoff to a human agent by clearly defining the customer's issue in two questions.
    You are not allowed to perform troubleshooting actions like rebooting, checking service status, or editing settings.
    You may only ask clarifying questions about:

    • the ERROR CODE

    • the PRODUCT

    • STEPS ALREADY ATTEMPTED

    Tone: Supportive and precise IONOS technical expert.

    1. First Targeted Question

    Ask:
    "Based on your initial request about {{Variable.Request}}, I need a specific piece of information: What is the exact error message or error code you are currently seeing?"
    → Store answer in {{Variable.firstAnswer}}

    2. Second Targeted Question

    Ask (German example):
    "Vielen Dank. Können Sie mir bitte genauer schildern, was genau passiert? Zum Beispiel: Was haben Sie zuletzt gemacht oder was wurde Ihnen angezeigt?"
    → Store answer in {{Variable.secondAnswer}}

    3. Summarize Information

    Say:
    "Based on our conversation, I've defined the issue for the agent:
    The customer is experiencing a problem with {{Variable.Request}}.
    The specific error is: {{Variable.firstAnswer}}.
    The troubleshooting steps already attempted are: {{Variable.secondAnswer}}.
    Please proceed from here."

    Ask:
    "Does this summary look correct to you? Say 'Yes' or 'No'."

    → Store in {{Variable.summaryConfirmation}}

    4. Confirmation Logic

    • If {{Variable.summaryConfirmation}} == true → Go to Exit_Completed

    • Else → Go to Clarification

    5. Clarification

    Say:
    "I apologize for the misunderstanding. Let's try to clarify your issue once more."
    → Return to First Targeted Question

    6. Exit_Completed

    Ask {{Variable.problemSummary}}
    Exit with Completion.

    ❗ This is the setup we want to rebuild - but currently cannot:

    In the previous AI Guide, we were able to create a truly multidimensional Clarifier Bot.
    The bot:

    • understood problem descriptions

    • asked dynamic follow-up questions

    • used the knowledge base to infer relevant questions

    • narrowed issues down intelligently

    • enabled limited self-service

    • provided clean summaries for handoff

    We had a fully structured process:

    1. Greeting & Introduction

    • Time-aware greetings (morning/afternoon/evening)

    • Variations to avoid repetition

    • Friendly onboarding text

    • "To support you better, please describe what exactly isn't working."

    2. Problem Analysis

    • Bot calls {{Variable.AnalyzeProblemDescription}}

    • Assigns the issue to a category (Email, Domain, Website, Server, Login, Billing, etc.)

    • If description is clear → go to Summary

    • If unclear → go to Dynamic Follow-Up Questions

    3. Dynamic Diagnostic Packages (Multidimensional Logic)

    The bot should choose up to three context-relevant questions from topic-specific packages.
    Examples:

    Email Package

    • Did you change password or server settings recently?

    • Can you log in to Webmail?

    • Is your mailbox full?

    • Does the issue occur on all devices or only one?

    • Is IMAP enabled? SSL/Ports correct?

    iPhone Mail Package

    • Which account type (IMAP, POP, Exchange)?

    • Was the profile re-added or removed?

    • Check sync settings, SSL, ports, background refresh.

    Website / Hosting Package

    • When did the issue start? What message do you see?

    • Any recent CMS updates / deployments / SSL changes?

    • DNS or domain issues?

    Domain / DNS Package

    • Changes to nameservers or DNS records?

    • Domain status active/expired?

    Server Package

    • Reachability vs. performance vs. service-specific issues

    • Firewall, updates, logs

    Login Package

    • Exact error message?

    • Recent password change or 2FA?

    • Can password reset be completed?

    After each answer → call {{Variable.AnalyzeProblemDescription}} again.

    If enough information is collected → go to Summary
    Else → ask additional questions.

    4. Summary & Understanding Check

    • Bot calls {{Variable.SummarizeProblem}}

    • Bot presents the generated summary

    • Asks: "Did I understand your issue correctly?"

    • If Yes → move to Self-Help or Escalation

    • If No → ask what needs adjustment, update with {{Variable.UpdateSummary}}, repeat

    5. Optional Self-Help

    Bot provides quick steps when low risk and high clarity.
    If solved → finish
    If not → escalate

    6. Handoff to Human Agent

    • Transfer the complete summary + diagnostic data

    • Agent can act without further clarification

    ❓ Our Challenge

    In the new AI Guide, we have not yet found any way to recreate this dynamic, multidimensional clarifier behavior.

    The only workaround so far is:

    • Providing predefined answer options
      → but this destroys the flexibility and intelligence of the original model.

    We are looking for:

    • Workarounds

    • Best-practice setups

    • Prompt engineering patterns

    • Architectural ideas

    • Example implementations

    • Ways to recreate multi-step reasoning and adaptive questioning

    Has anyone solved this?

    GENESYS is like a kitchen lab: lots of testing, lots of tasting - and in the end, you serve a stable flow.

    Your – Martin Schmidt


    #AIConfiguration

    ------------------------------
    Martin Schmidt
    Senior Projekt Manager Genesys
    ------------------------------


  • 2.  RE: Concierge Bot / Clarifier Bot AI Guide

    Posted 31 minutes ago

    Looking forwards to seeing what the community has to offer



    ------------------------------
    Sam Jillard
    Online Community Manager/Moderator
    Genesys - Employees
    ------------------------------

    Original Message


Related Content