Genesys Cloud - Developer Community!

Hi Team,

We are trying to understand how AI token consumption works in Genesys Cloud.

Our organization receives 250 AI tokens per month for free, and our subscription also includes 1,600 committed AI tokens.

Could you please clarify the following?

1. What exactly does the 1,600 committed AI tokens mean? Is it a monthly committed allocation or a prepaid quantity?
2. How are AI tokens consumed during voice interactions?
3. For a voice bot, is 1 token consumed per call, or is it consumed based on the duration of the call?
4. If token consumption is time-based, how is it calculated (for example, per minute or per second)?
5. If a single interaction uses multiple AI features (such as Voice Bot, Virtual Agent, AI Summary, etc.), how are the tokens calculated? Are they charged separately or based on the highest AI feature used?
6. Is there any report available in Genesys Cloud where we can monitor AI token consumption by feature or interaction?

Thanks in advance for your clarification.

#Architect

------------------------------
Rajasekhar Reddy
Developer
------------------------------

Hi Rajasekhar,

These are excellent questions. Based on the current Genesys documentation, my understanding is as follows. I would also appreciate confirmation from Genesys on the interpretation of prepaid commitments and Voice Bot metering boundaries.

First, AI Experience tokens should be understood as Genesys commercial metering units for AI capabilities. They are not equivalent to raw LLM input or output tokens.

1. Meaning of the 1,600 committed AI tokens

The 250 AI tokens appear to be the monthly fair-use allocation for a Named organization. This allocation renews monthly and does not carry forward.

However, the 1,600 committed tokens require confirmation against the Service Order or invoice. Genesys allows AI Experience tokens to be prepaid on either a monthly or annual basis. Therefore, the commitment could represent a monthly quantity, an annual prepaid quantity, or another contractual structure.

I would recommend confirming:

• Whether the 1,600 tokens renew monthly or are consumed across an annual term.

• Whether unused prepaid tokens carry forward within the contract period.

• How overages are calculated after the prepaid commitment is exhausted.

2. Voice Bot consumption

For a traditional Genesys Voice Bot, token consumption is based on runtime usage, not one token per call.

The published conversion is:

17 minutes of Voice Bot runtime = 1 AI Experience token

Therefore, the relevant metric is the total runtime accumulated in the Voice Bot, aggregated across interactions.

3. Time calculation and rounding

For planning purposes, the calculation appears to be:

Total Voice Bot runtime minutes ÷ 17 = estimated tokens consumed

For example, 1,700 cumulative minutes of Voice Bot runtime would represent approximately 100 tokens.

However, I could not find a public statement that explains the precise rounding rule at an individual interaction level. For example, it is not clear whether runtime is rounded per second, per bot invocation, per conversation, or only after usage is aggregated for billing.

4. Hybrid Inbound Call Flow and Voice Bot scenario

One scenario that is especially relevant for architecture design is a hybrid model:

• The Inbound Call Flow manages deterministic steps such as ANI identification, authentication, API calls, business rules, routing, and TTS.

• The flow invokes a Voice Bot only for a specific conversational task, such as identifying the customer's intent through natural language.

• The Voice Bot returns an intent, slots, output variables, and an exit reason.

• The Inbound Call Flow resumes and continues the journey using deterministic logic.

For example:

Inbound Call Flow
→ identify customer
→ retrieve order information
→ call Voice Bot only to capture the open request
→ receive intent and slots from the bot
→ return to the Inbound Call Flow
→ apply business rules, call APIs, provide the response, route, or end the call.

Architect supports this pattern through the Call Bot Flow action, including inputs, outputs, intent, and exit reason.

My interpretation is that, when only a traditional Voice Bot is used, the relevant AI token consumption should be the time spent inside the Voice Bot portion of the interaction, rather than the entire call duration.

For example:

• 1 minute in the Inbound Call Flow

• 40 seconds in the Voice Bot

• 2 minutes back in the Inbound Call Flow

Would only the 40 seconds of Voice Bot runtime be included in the 17-minutes-per-token calculation?

It would be helpful if Genesys could confirm the exact metering boundary between the Inbound Call Flow and the Voice Bot, as well as the rounding method for these shorter bot invocations.

5. Multiple AI resources in the same interaction

The current billing scenarios indicate that, when Bot Flows, Virtual Agents, and Agentic Virtual Agents are used in the same interaction, Genesys applies the highest applicable AI resource rate rather than charging each of these resources separately.

The current published rates are:

• Traditional Voice Bot: 17 minutes per token

• Virtual Agent: 0.5 token per interaction

• Agentic Virtual Agent with LAM: 1.2 tokens per interaction

For example:

• Voice Bot only: charged based on runtime.

• Voice Bot followed by Virtual Agent: charged at 0.5 token per interaction.

• Voice Bot followed by Agentic Virtual Agent: charged at 1.2 tokens per interaction.

However, I would not automatically apply this "highest-tier resource" rule to every AI capability in the platform without confirmation. Other capabilities, such as AI Summary and Insights, Agent Copilot, AI Scoring, and AI Translate, have their own published meters.

6. Architecture consideration for Agentic Virtual Agent

There is an important cost-design consideration for Agentic Virtual Agent. Genesys indicates that an interaction can be charged at the Agentic Virtual Agent rate when it visits a flow that contains an Agentic Virtual Agent block, even if that block is not ultimately executed.

For that reason, it may be safer to keep Agentic Virtual Agent capabilities in dedicated bot flows and invoke them only when truly needed, instead of placing them inside a shared flow that every interaction visits.

This also supports a hybrid architecture in which the Inbound Call Flow remains the main orchestrator and selectively invokes the appropriate conversational capability.

7. Monitoring and reporting

For usage monitoring, the Resources Usage report under:

Admin > Account > Subscription > Download Usage Report

includes AI token consumption for billable resources such as bots, Predictive Engagement, and predictive routing.

The Virtual Agent Performance dashboard can also help operationally monitor containment, escalation, transfer, recognition failures, and errors.

However, I have not found a native report that allocates the exact AI token charge to each individual conversation or interaction. For internal cost attribution, it may be necessary to combine usage reports with Architect flow data, bot analytics, and conversation analytics.

Confirmation on the hybrid Inbound Call Flow + Voice Bot scenario would be especially valuable for teams designing controlled, cost-efficient AI-assisted IVR journeys.

------------------------------
Fernando Sotto dos Santos
Consultor de Atendimento Senior Grupo Casas Bahia
------------------------------

Good explanation, but here are some caveats:

1. Committed and free tokens are "use it or lose it",  so there is no carryover or yearly bucket.   Also, all token calculations are rounded up to whole tokens, so if you use AVA on 2 calls, that will be 3 tokens total.  
2. Bot minutes:  Bot minutes are billed in 15-second increments.  So 16 seconds will equate to 30 seconds.  Total is rounded up to the minute at the end of the month.
3. Inbound flows are part of the IVR Fair Use on a per-minute basis.  Also, TTS is charged per million characters in Call Flows, but are included in Bot Flow charges.  
4. When it comes to Bot Flows, Virtual Agents, Guides, and Agentic Virtual Agents, a billing ID is applied to the conversation whenever a more sophisticated technology is used.  Also, only one billing ID will be used per conversation no matter how many times you move in and out of them.  For Bot Flows, the Billing ID is 17 minutes per token.  For Virtual Agents and Guides, that billing ID becomes 2 interactions (voice or digital) per token.  For AVA calls, that moves the billing ID to 1.2 tokens per interaction (voice or digital)
5. When calculating different billing IDs, you will have a tough time if you mix different flow types.  For instance, if you have a bot flow of 10 minutes, you will be charged a token, as billing is always in whole amounts. But if you flow in and out of different technologies, that billing goes away, and the highest billing ID for the conversation trumps all others. You can't simply look at total bot minutes or flow entries to understand fully what you will be charged. Likewise, it is difficult to ascertain which billing ID is applied to each conversation.  

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

Original Message:
Sent: 06-27-2026 12:56
From: Fernando Sotto dos Santos
Subject: Clarification on Genesys Cloud AI Token Consumption

Hi Rajasekhar,

These are excellent questions. Based on the current Genesys documentation, my understanding is as follows. I would also appreciate confirmation from Genesys on the interpretation of prepaid commitments and Voice Bot metering boundaries.

First, AI Experience tokens should be understood as Genesys commercial metering units for AI capabilities. They are not equivalent to raw LLM input or output tokens.

1. Meaning of the 1,600 committed AI tokens

The 250 AI tokens appear to be the monthly fair-use allocation for a Named organization. This allocation renews monthly and does not carry forward.

However, the 1,600 committed tokens require confirmation against the Service Order or invoice. Genesys allows AI Experience tokens to be prepaid on either a monthly or annual basis. Therefore, the commitment could represent a monthly quantity, an annual prepaid quantity, or another contractual structure.

I would recommend confirming:

• Whether the 1,600 tokens renew monthly or are consumed across an annual term.

• Whether unused prepaid tokens carry forward within the contract period.

• How overages are calculated after the prepaid commitment is exhausted.

2. Voice Bot consumption

For a traditional Genesys Voice Bot, token consumption is based on runtime usage, not one token per call.

The published conversion is:

17 minutes of Voice Bot runtime = 1 AI Experience token

Therefore, the relevant metric is the total runtime accumulated in the Voice Bot, aggregated across interactions.

3. Time calculation and rounding

For planning purposes, the calculation appears to be:

Total Voice Bot runtime minutes ÷ 17 = estimated tokens consumed

For example, 1,700 cumulative minutes of Voice Bot runtime would represent approximately 100 tokens.

However, I could not find a public statement that explains the precise rounding rule at an individual interaction level. For example, it is not clear whether runtime is rounded per second, per bot invocation, per conversation, or only after usage is aggregated for billing.

4. Hybrid Inbound Call Flow and Voice Bot scenario

One scenario that is especially relevant for architecture design is a hybrid model:

• The Inbound Call Flow manages deterministic steps such as ANI identification, authentication, API calls, business rules, routing, and TTS.

• The flow invokes a Voice Bot only for a specific conversational task, such as identifying the customer's intent through natural language.

• The Voice Bot returns an intent, slots, output variables, and an exit reason.

• The Inbound Call Flow resumes and continues the journey using deterministic logic.

For example:

Inbound Call Flow
→ identify customer
→ retrieve order information
→ call Voice Bot only to capture the open request
→ receive intent and slots from the bot
→ return to the Inbound Call Flow
→ apply business rules, call APIs, provide the response, route, or end the call.

Architect supports this pattern through the Call Bot Flow action, including inputs, outputs, intent, and exit reason.

My interpretation is that, when only a traditional Voice Bot is used, the relevant AI token consumption should be the time spent inside the Voice Bot portion of the interaction, rather than the entire call duration.

For example:

• 1 minute in the Inbound Call Flow

• 40 seconds in the Voice Bot

• 2 minutes back in the Inbound Call Flow

Would only the 40 seconds of Voice Bot runtime be included in the 17-minutes-per-token calculation?

It would be helpful if Genesys could confirm the exact metering boundary between the Inbound Call Flow and the Voice Bot, as well as the rounding method for these shorter bot invocations.

5. Multiple AI resources in the same interaction

The current billing scenarios indicate that, when Bot Flows, Virtual Agents, and Agentic Virtual Agents are used in the same interaction, Genesys applies the highest applicable AI resource rate rather than charging each of these resources separately.

The current published rates are:

• Traditional Voice Bot: 17 minutes per token

• Virtual Agent: 0.5 token per interaction

• Agentic Virtual Agent with LAM: 1.2 tokens per interaction

For example:

• Voice Bot only: charged based on runtime.

• Voice Bot followed by Virtual Agent: charged at 0.5 token per interaction.

• Voice Bot followed by Agentic Virtual Agent: charged at 1.2 tokens per interaction.

However, I would not automatically apply this "highest-tier resource" rule to every AI capability in the platform without confirmation. Other capabilities, such as AI Summary and Insights, Agent Copilot, AI Scoring, and AI Translate, have their own published meters.

6. Architecture consideration for Agentic Virtual Agent

There is an important cost-design consideration for Agentic Virtual Agent. Genesys indicates that an interaction can be charged at the Agentic Virtual Agent rate when it visits a flow that contains an Agentic Virtual Agent block, even if that block is not ultimately executed.

For that reason, it may be safer to keep Agentic Virtual Agent capabilities in dedicated bot flows and invoke them only when truly needed, instead of placing them inside a shared flow that every interaction visits.

This also supports a hybrid architecture in which the Inbound Call Flow remains the main orchestrator and selectively invokes the appropriate conversational capability.

7. Monitoring and reporting

For usage monitoring, the Resources Usage report under:

Admin > Account > Subscription > Download Usage Report

includes AI token consumption for billable resources such as bots, Predictive Engagement, and predictive routing.

The Virtual Agent Performance dashboard can also help operationally monitor containment, escalation, transfer, recognition failures, and errors.

However, I have not found a native report that allocates the exact AI token charge to each individual conversation or interaction. For internal cost attribution, it may be necessary to combine usage reports with Architect flow data, bot analytics, and conversation analytics.

Confirmation on the hybrid Inbound Call Flow + Voice Bot scenario would be especially valuable for teams designing controlled, cost-efficient AI-assisted IVR journeys.

------------------------------
Fernando Sotto dos Santos
Consultor de Atendimento Senior Grupo Casas Bahia
------------------------------

Nice explanation 👍

------------------------------
Rohit Yadav
------------------------------