Genesys Cloud - Main

Sign Up

Expand all | Collapse all

Bot flow - Best practice

  • 1.  Bot flow - Best practice

    Posted 11 days ago

    Good Day Community

    I would like to ask for your assistance with the following.  I have been using the best practices and recommendations to build my Voicebot used for testing.

    Best practice recommendations for building bots in Architect - Genesys Cloud Resource Center

    Wanted to ask if you could perhaps share some ideas or recommendations from your experience in building bots.  How do you treat the conversation if the customer mentions something that is outside of the provide intent?

    Without just providing the no match or no input messaging, any tips or tricks to improve customer experience.

    Thanking you in advance for your support.

    Regards


    #ArchitectandDesign
    #ConversationalAI(Bots,VirtualAgent,etc.)

    ------------------------------
    Stephan Taljaard
    EMBEDIT s.r.o
    ------------------------------


  • 2.  RE: Bot flow - Best practice

    Posted 10 days ago

    Great question Stephan.  Hopefully others can provide some useful tips and tricks



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



  • 3.  RE: Bot flow - Best practice

    Posted 10 days ago

    Do you have an example of a comment that would be "outside of the provide intent"?



    ------------------------------
    James Bonnevie
    Technology Engineer Consultant
    ------------------------------



  • 4.  RE: Bot flow - Best practice

    Posted 9 days ago

    Hi James

    Example would be where the customer enquires about account status but then asks something outside of the account intent that is not currently available.  The Bot will then respond with the no match response and action according to the user input and error handling config. 

    So I'm looking into some of those scenarios and how best to treat them in a production environment.  I understand that you wont be able to cover all topics or questions, just interested in how you treat those to ensure customer satisfaction.

    Regards



    ------------------------------
    Stephan Taljaard
    EMBEDIT s.r.o
    ------------------------------



  • 5.  RE: Bot flow - Best practice

    Posted 10 days ago
    Edited by Luiz Rosa 10 days ago

    Hi Stephan Taljaard

    The bot learns from examples rather than strict rules. Because of that, it's normal that some user inputs fall outside the defined intents.

    In practice, a few things usually help:

    • Provide enough training examples for each intent (Genesys typically recommends around 20–30).

    • Make sure intents don't overlap too much, otherwise the model may get confused.

    • Use slot filling or clarification prompts when the bot is not confident.

    • Continuously review real conversations and refine the model over time.

    You can also use the Optimization Dashboard to analyze metrics like fallback rate and intent confidence. This helps identify where the model needs improvement.

    https://help.genesys.cloud/articles/view-bot-and-digital-bot-metrics-in-the-optimization-dashboard/

    Hope this helps.

    ------------------------------
    Luiz Rosa
    Full stack developer
    ------------------------------



  • 6.  RE: Bot flow - Best practice

    Posted 10 days ago

    Hi Luiz

    Hope you are doing well. 

    Thank you so much for the insight and information shared, much appreciated. 

    Regards

    Stephan



    ------------------------------
    Stephan Taljaard
    EMBEDIT s.r.o
    ------------------------------



  • 7.  RE: Bot flow - Best practice

    Posted 9 days ago

    One best practice I always use is to create dedicated confirmation and denial intents instead of relying only on the standard Ask for Yes/No behavior.

    In many cases, especially in chat, customers do not answer with a clean "yes" or "no." They may use variations, abbreviations, or informal language such as "aham," "sure," "ok," "s," "n," or other slang expressions. The default yes/no handling usually does not capture these well.

    By creating intents for affirmation and negation, the bot becomes much more flexible and natural, because it can recognize the way people actually respond instead of forcing them into one exact format.

    This helps a lot with customer experience, especially when the user says something slightly outside the expected flow. Instead of falling too quickly into no-match behavior, the bot has a better chance of understanding the real meaning and keeping the conversation moving.



    ------------------------------
    Mateus Nunes
    Tech Leader Of CX at Solve4ME
    Brazil
    ------------------------------



  • 8.  RE: Bot flow - Best practice

    Posted 9 days ago

    Hi Mateus

    Thank for this information, much appreciated.  As you mentioned, it can make a big difference with regards to continuing the conversation, rather then just terminating the call.

    Regards



    ------------------------------
    Stephan Taljaard
    EMBEDIT s.r.o
    ------------------------------



  • 9.  RE: Bot flow - Best practice
    Best Answer

    Posted 9 days ago

    We have been using a pre-Virtual Agent bot flow for our inbound call processing for a year now. Here's a breakdown of the flows we use on a typical inbound call.

    Inbound Call Flow: Does the greeting, bunch of customer info collection via data actions to our CMR and other internal platforms that are set as flow variables. Yes/No questions sprinkled here and there that go to a bot flow just for handling Yes/No. Depending on the NL Intent resolved, we either send to queue or other automated systems we have built.

    Inbound Bot Flow Ask for Intent - configured as Natural Language (NLU) asks the callers "In a few words, how can I assist you?" The utterances are collected and matched to their respective intent. In cases of people that go non-compliant with the bot, such as saying agent over and over, I have an Agent Intent built to catch those calls to make another pass at collecting an Intent from the caller via a 2nd Ask for Intent reusable task. This second pass is a copy of the first pass at getting their Intent but gives them some different verbiage to help guide them along, such as informing them they can say membership, insurance, etc.. If they still do not comply, as a last-ditch effort to still serve the caller, we have a stripped down DTMF menu in the Inbound Call Flow that if used, sends the caller to a generic queue for that specific Line of Business. This DTMF has no further clarifying questions on it as the caller is already annoyed at having to try and defeat the Intent collection process in NL.

    Sending to the DTMF menu is done by having the NL Bot flow set a variable of Flow.RecogFailure in a reusable task that Event Handling for the flow points to. From there, the call is handed back to the inbound call flow where Flow.RecogFailure is on a decision node to send it to the stripped down DTMF menu if set to True.

    Inbound Bot Flow Yes/No - a simple bot whose sole purpose is to serve butter ask the caller a Yes/No question. The question is passed as a variable from the Inbound Call Flow to the Bot Flow and shoots an output back of their Yes/No answer. Definitely helps to have Yes and No built as Slots so you can add in synonyms for Yes; ex: affirmative, correct.

    Hopefully this pre-caffeinated, eyes half open wall of text helps.



    ------------------------------
    Brad Carroll
    Auto Club Group
    ------------------------------



  • 10.  RE: Bot flow - Best practice

    Posted 9 days ago

    Hi Brad

    Thank you so much for sharing your ideas and insights regarding this topic, much appreciated.

    Regards



    ------------------------------
    Stephan Taljaard
    EMBEDIT s.r.o
    ------------------------------



  • 11.  RE: Bot flow - Best practice

    Posted 19 hours ago

    Great insights, Brad and Luiz! As a Technical Account Manager, I've seen customers struggle with this exact challenge. One additional recommendation: work with your CSM to schedule regular bot performance reviews using the Optimization Dashboard metrics.

    In my experience, customers who review their fallback rates monthly and continuously refine their training utterances see 30-40% improvement in intent recognition within the first quarter.

    Also, consider implementing conversation transcription analysis to identify patterns in 'no match' scenarios - this helps you discover missing intents you didn't know customers needed.



    ------------------------------
    Minhaj Mubashir
    Technical Account Manager
    ------------------------------



  • 12.  RE: Bot flow - Best practice

    Posted 16 hours ago

    From my experience, one approach that has worked very well is to design a safe fallback path for unexpected situations.

    For example, whenever there is any type of failure - API errors, recognition issues, unexpected responses, or flow exceptions - I like to route the interaction to an emergency queue. This guarantees that the customer will still be assisted by an agent instead of getting stuck in the bot experience.

    Another advantage of this approach is that all interactions routed to that queue become great candidates for analysis. By reviewing those conversations, we can understand what failed (NLU gaps, API errors, unclear prompts, etc.) and continuously improve the bot.

    Another best practice I often apply is creating custom affirmation and negation intents. The built-in Ask for Yes/No works well but only accepts very strict responses like "yes" or "no". In real conversations, customers may say things like "yeah", "of course", "not really", "I guess so", etc. Creating intents for affirmation and negation helps capture those variations and makes the experience much more natural.

    For voice bots specifically, it is also worth testing the different speech recognition engines (V1, V2, and V3). There isn't a universally "best" engine - in my experience, performance can vary depending on the organization, language usage, and audio characteristics. Running controlled tests helps determine which engine performs better for your scenario.

    These practices have helped me make bots more resilient while also creating a feedback loop to continuously improve the experience.



    ------------------------------
    Mateus Nunes
    Tech Leader Of CX at Solve4ME
    Brazil
    ------------------------------