Certification and Training

Good Insights here. We know, we can output the data linked with tools or start contexts from Agentic, but we would also like to see the ability to bring some dynamically generated data from Agentic to Architect flow/scripts.

Good Insights here. We know, we can output the data linked with tools or start contexts from Agentic, but we would also like to see the ability to bring some dynamically generated data from Agentic to Architect flow/scripts.

We can look into adding direct support for data tables as a roadmap item, but its actually really easy to call a data table from a data action. In fact, its easy for all Genesys APIs. For data tables specifically, you would create a data action, no auth needed assuming its in the same org, and then this becomes your Request URL:
/api/v2/flows/datatables/<DATA_TABLE_ID>/rows/${input.key}?showbrief=false

You would also update input.key if you're using a different string for the key reference of your data table. Make sure to include the showbrief=false to return the full data table row. 

I use this all the time!

Good Insights here. We know, we can output the data linked with tools or start contexts from Agentic, but we would also like to see the ability to bring some dynamically generated data from Agentic to Architect flow/scripts.

@Mitchell Mason - Hope you've been well. We're at the point where we're throwing all types of edge cases and unscripted scenarios at our AVA, and ran into something interesting. Before I explain what we've noticed, my question to you is "how would you get a conversation that is clearly going nowhere out of the hands of the AVA?" I'm thinking of something like a "No Match" path it can take after x attempts.

For example, despite [what we believe are] clear guidelines to escalate the interaction to a live agent IF the context is unclear or the conversation is not progressing, most, if not all, times the AVA holds onto the conversation asking for more clarity:

Here are the two primary Guidelines it should've been using, which ironically are the first and last Guidelines to help emphasize the importance. I wouldn't be surprised if there's a better way to write these guidelines, but figured I'd be as verbose as possible just to see what would happen.

------------------------------
Brian T. Jones | Ascension | Senior Specialist - Technology | Colossians 3:23-24
------------------------------

On your first prompt, try reversing the language.  Have it let the caller know they need to escalate to an agent, then tell the AVA to run the function and exit the agent with escalation.   

For the second, did you enable the Wait Experience?  That might conflict with that guideline.

------------------------------
Robert Wakefield-Carl
ttec Digital
Sr. Director - Innovation Architects
Robert.WC@ttecdigital.com
https://www.ttecDigital.com
https://RobertWC.Blogspot.com
------------------------------


Original Message:
Sent: 05-18-2026 14:57
From: Brian Jones
Subject: Ask the Expert: LAM Powered Agentic Virtual Agents

@Mitchell Mason - Hope you've been well. We're at the point where we're throwing all types of edge cases and unscripted scenarios at our AVA, and ran into something interesting. Before I explain what we've noticed, my question to you is "how would you get a conversation that is clearly going nowhere out of the hands of the AVA?" I'm thinking of something like a "No Match" path it can take after x attempts.

For example, despite [what we believe are] clear guidelines to escalate the interaction to a live agent IF the context is unclear or the conversation is not progressing, most, if not all, times the AVA holds onto the conversation asking for more clarity:

Here are the two primary Guidelines it should've been using, which ironically are the first and last Guidelines to help emphasize the importance. I wouldn't be surprised if there's a better way to write these guidelines, but figured I'd be as verbose as possible just to see what would happen.

------------------------------
Brian T. Jones | Ascension | Senior Specialist - Technology | Colossians 3:23-24

Thanks for responding RWC. Do you have an example of how you would write the first guideline? I've rephrased that thing a dozen times, yet still get stuck in a loop with the AVA. Also, is what you're proposing reliant on the Agent Escalation feature being enabled in Architect? Asking because we don't have that enabled for a reason (i.e. agent escalations are gracefully handled by other guidelines), and in this test case an agent is not explicitly being requested--the patient is just being difficult 😉

As for your second point, yes, we have the Wait Experience enabled, but I'm unsure how/why that would conflict with the last guideline given the Comfort Statements only come into play when the AVA is using a Tool (which is not the case in my example given it never gets more than a step out of the starting blocks, haha).

------------------------------
Brian T. Jones | Ascension | Senior Specialist - Technology | Colossians 3:23-24
------------------------------


Original Message:
Sent: 05-19-2026 09:53
From: Robert Wakefield-Carl
Subject: Ask the Expert: LAM Powered Agentic Virtual Agents

On your first prompt, try reversing the language.  Have it let the caller know they need to escalate to an agent, then tell the AVA to run the function and exit the agent with escalation.   

For the second, did you enable the Wait Experience?  That might conflict with that guideline.

------------------------------
Robert Wakefield-Carl
ttec Digital
Sr. Director - Innovation Architects
Robert.WC@ttecdigital.com
https://www.ttecDigital.com
https://RobertWC.Blogspot.com


Original Message:
Sent: 05-18-2026 14:57
From: Brian Jones
Subject: Ask the Expert: LAM Powered Agentic Virtual Agents

@Mitchell Mason - Hope you've been well. We're at the point where we're throwing all types of edge cases and unscripted scenarios at our AVA, and ran into something interesting. Before I explain what we've noticed, my question to you is "how would you get a conversation that is clearly going nowhere out of the hands of the AVA?" I'm thinking of something like a "No Match" path it can take after x attempts.

For example, despite [what we believe are] clear guidelines to escalate the interaction to a live agent IF the context is unclear or the conversation is not progressing, most, if not all, times the AVA holds onto the conversation asking for more clarity:

Here are the two primary Guidelines it should've been using, which ironically are the first and last Guidelines to help emphasize the importance. I wouldn't be surprised if there's a better way to write these guidelines, but figured I'd be as verbose as possible just to see what would happen.

------------------------------
Brian T. Jones | Ascension | Senior Specialist - Technology | Colossians 3:23-24

For the escalation, best is to exit the AVA into the Bot Flow with a variable that will let the Bot Flow to follow the transfer to ACD path.   I would word it like this:
Say "Sorry-I'm not able to help with that right now. I've tried to clarify twice but still don't have enough information. I'm going to transfer you to a live agent who can help. Please hold while I connect you." only after the caller has failed to provide enough information in two separate clarification attempts, or when the caller cannot describe the issue clearly enough for a third attempt to be useful. Use the set_participantdata with the input of "Agent" and escalate to a live agent.

The Comfort stuff is brand new, but I did see a difference in responses when the AVA repeats itself when turned on.  I am not for sure that is the problem.  Try this for the second guideline: "Ask at most two clarifying questions, but rephrase each one differently and include an example if helpful. Do not repeat the same wording or engage in small talk. If the caller still cannot provide enough context after the second attempt, set participant data and transfer to a live agent."

Thanks for responding RWC. Do you have an example of how you would write the first guideline? I've rephrased that thing a dozen times, yet still get stuck in a loop with the AVA. Also, is what you're proposing reliant on the Agent Escalation feature being enabled in Architect? Asking because we don't have that enabled for a reason (i.e. agent escalations are gracefully handled by other guidelines), and in this test case an agent is not explicitly being requested--the patient is just being difficult 😉

As for your second point, yes, we have the Wait Experience enabled, but I'm unsure how/why that would conflict with the last guideline given the Comfort Statements only come into play when the AVA is using a Tool (which is not the case in my example given it never gets more than a step out of the starting blocks, haha).

------------------------------
Brian T. Jones | Ascension | Senior Specialist - Technology | Colossians 3:23-24


Original Message:
Sent: 05-19-2026 09:53
From: Robert Wakefield-Carl
Subject: Ask the Expert: LAM Powered Agentic Virtual Agents

On your first prompt, try reversing the language.  Have it let the caller know they need to escalate to an agent, then tell the AVA to run the function and exit the agent with escalation.   

For the second, did you enable the Wait Experience?  That might conflict with that guideline.

------------------------------
Robert Wakefield-Carl
ttec Digital
Sr. Director - Innovation Architects
Robert.WC@ttecdigital.com
https://www.ttecDigital.com
https://RobertWC.Blogspot.com


Original Message:
Sent: 05-18-2026 14:57
From: Brian Jones
Subject: Ask the Expert: LAM Powered Agentic Virtual Agents

@Mitchell Mason - Hope you've been well. We're at the point where we're throwing all types of edge cases and unscripted scenarios at our AVA, and ran into something interesting. Before I explain what we've noticed, my question to you is "how would you get a conversation that is clearly going nowhere out of the hands of the AVA?" I'm thinking of something like a "No Match" path it can take after x attempts.

For example, despite [what we believe are] clear guidelines to escalate the interaction to a live agent IF the context is unclear or the conversation is not progressing, most, if not all, times the AVA holds onto the conversation asking for more clarity:

Here are the two primary Guidelines it should've been using, which ironically are the first and last Guidelines to help emphasize the importance. I wouldn't be surprised if there's a better way to write these guidelines, but figured I'd be as verbose as possible just to see what would happen.

------------------------------
Brian T. Jones | Ascension | Senior Specialist - Technology | Colossians 3:23-24

Hello, community!

Thank you so much for the discussion! I hope, I'm in the right community with my question.

I'm currently thinking about the best way to use AVA. I understand that it's particularly well-suited for dynamic conversations and FAQ scenarios. I'm wondering what limitations AVA has regarding the size or complexity of the bot. For example, if I have many FAQ cases in the banking sector that I can resolve by integrating Knowledge, many additional use cases involving tool usage could arise. What are AVA's limitations, and what would the architecture look like in that case? Would I build more than one AVA bot that I orchestrate via the Architect or do I have one big AVA bot handling all my scenarios? I would be very interested to hear the community's and experts' thoughts on this.

Many thanks to everyone; I really appreciate the exchange of experiences!

First, remember that this product has been released for less than 2 months. Second, NO other companies offer the same level of technology, ease of use, and production speed that AVA offers. Sure, you can find tools out there that do what AVA does if you are willing to spend the time integrating them and the tokens they require.  

The other matter is the price. Using AVA for strictly and AVA bot is a bit overkill, especially with Gemini and Bing out there scouring the resource center and community pages for answers. Combining FAQs with Intelligent routing is still an expensive proposition. Remember that EACH conversation will cost you 1.2 tokens. That is nearly 20 minutes of standard voice bot, which, if you spend an average of, say, 3 min in the IVR, is 7 calls, totaling $8.40 with AVA. Please make sure that whatever the AVA is doing makes it worth it. Sure, the way the AVA interacts with customers so naturally is cool, but will upper management approve $1.20 per call to sound cool?

Finally, where AVA really shines is combining the routing with FAQ and (at present), some outcome-based self-service items like lookup and paying accounts, filling out forms ahead of transferring to a human agent, or fulfilling the low-hanging fruit that your front-line agents are handling so they can be better utilized for more advanced conversations.  

I feel that at the current iteration of AVA, we are best introducing the agentic conversation capabilities to replace voice-based bots and shooting for back-end data actions and MCP servers to support more sophisticated service features as the product matures, which, if the last 6 months of innovation from Genesy is any indication, could be in just a few weeks!

If you are looking for where AVA is being used now, it's hard to find examples, as few companies are fully up to speed on Genesys development to have a product in production. Happy to discuss what we have shown to clients and would love to hear about upcoming projects from others with an AVA focus.    

Hello, community!

Thank you so much for the discussion! I hope, I'm in the right community with my question.

I'm currently thinking about the best way to use AVA. I understand that it's particularly well-suited for dynamic conversations and FAQ scenarios. I'm wondering what limitations AVA has regarding the size or complexity of the bot. For example, if I have many FAQ cases in the banking sector that I can resolve by integrating Knowledge, many additional use cases involving tool usage could arise. What are AVA's limitations, and what would the architecture look like in that case? Would I build more than one AVA bot that I orchestrate via the Architect or do I have one big AVA bot handling all my scenarios? I would be very interested to hear the community's and experts' thoughts on this.

Many thanks to everyone; I really appreciate the exchange of experiences!

Hi Robert,

thank you very much for your explanations and thoughts. For sure, the product is very new in the market and usage in production will be difficult to find. I asked my question as I'm currently in the situation of discussing with the customer how we are going to set up the new bot. For now the answer might be quite simple and of course the 1.2 tokens per interaction plays also a significant role in all this discussion. As the bot will grow as we are going to add more and more use cases I was wondering if one AVA bot might handle all the complexity using the role and the guidelines – for sure I have not enough experience to answer this question as of now. That's why I wanted to reach out here and check for any thoughts or initial experiences with more complex AVA bots (e.g. also from any PoCs which didn't probably make it to production. Anyways really appreciated your thoughts.

BR,

Uzay   

First, remember that this product has been released for less than 2 months. Second, NO other companies offer the same level of technology, ease of use, and production speed that AVA offers. Sure, you can find tools out there that do what AVA does if you are willing to spend the time integrating them and the tokens they require.  

The other matter is the price. Using AVA for strictly and AVA bot is a bit overkill, especially with Gemini and Bing out there scouring the resource center and community pages for answers. Combining FAQs with Intelligent routing is still an expensive proposition. Remember that EACH conversation will cost you 1.2 tokens. That is nearly 20 minutes of standard voice bot, which, if you spend an average of, say, 3 min in the IVR, is 7 calls, totaling $8.40 with AVA. Please make sure that whatever the AVA is doing makes it worth it. Sure, the way the AVA interacts with customers so naturally is cool, but will upper management approve $1.20 per call to sound cool?

Finally, where AVA really shines is combining the routing with FAQ and (at present), some outcome-based self-service items like lookup and paying accounts, filling out forms ahead of transferring to a human agent, or fulfilling the low-hanging fruit that your front-line agents are handling so they can be better utilized for more advanced conversations.  

I feel that at the current iteration of AVA, we are best introducing the agentic conversation capabilities to replace voice-based bots and shooting for back-end data actions and MCP servers to support more sophisticated service features as the product matures, which, if the last 6 months of innovation from Genesy is any indication, could be in just a few weeks!

If you are looking for where AVA is being used now, it's hard to find examples, as few companies are fully up to speed on Genesys development to have a product in production. Happy to discuss what we have shown to clients and would love to hear about upcoming projects from others with an AVA focus.    

Hello, community!

Thank you so much for the discussion! I hope, I'm in the right community with my question.

I'm currently thinking about the best way to use AVA. I understand that it's particularly well-suited for dynamic conversations and FAQ scenarios. I'm wondering what limitations AVA has regarding the size or complexity of the bot. For example, if I have many FAQ cases in the banking sector that I can resolve by integrating Knowledge, many additional use cases involving tool usage could arise. What are AVA's limitations, and what would the architecture look like in that case? Would I build more than one AVA bot that I orchestrate via the Architect or do I have one big AVA bot handling all my scenarios? I would be very interested to hear the community's and experts' thoughts on this.

Many thanks to everyone; I really appreciate the exchange of experiences!