#Einstein Bots1 discussing

Hi community,

I'm getting a persistent generic error when trying to activate Build Model on an Enhanced Einstein Bot (pt-BR), and I can't pinpoint the cause. 

Setup:

• 61 intents total
• 7 intents with 20 utterances each and mlIntentTrainingEnabled = true
• 54 intents with 3–10 utterances and mlIntentTrainingEnabled = false
• No Intent Sets, Cross-Lingual disabled, all utterances locally defined

One additional note on utterance quality: There are no fully duplicate utterances (identical punctuation and casing), but some phrases appear more than once across intents with case variations only

 (e.g., the same sentence in all-lowercase vs. title case). It's unclear whether the NLP pipeline treats these as duplicates during the build process.  

The issue: Build Model always fails with "Couldn't build model." — no further detail.

Interestingly.  

if I drop any trained intent below 20 utterances, the error 

changes

 to the descriptive validation message listing the affected intent. This confirms the system recognizes my trained intents and their counts — yet the build still fails silently when all requirements appear to be met. 

My suspicion: There may be orphaned MLIntent records in the org — intents previously deployed and later removed from metadata without

 explicit destructive deletion — that are corrupting the ML Domain state without surfacing a clear error. 

Has anyone seen this before? Any way to get more detail beyond the generic UI message?

Thanks! 

Hi, Trailblazers!

I’ve successfully configured and deployed my Einstein Bot for a Web Chat channel and added it to my Experience Cloud site. While the chat icon appears on the portal, customers can’t click on it to start a conversation—it’s unresponsive.

What I’ve checked so far:

The bot is active and tested successfully within Salesforce.

The Web Chat channel is set up in Omni-Channel, and the bot is assigned as the agent.

The Embedded Service code is correctly embedded in the portal.

The chat component is added in Experience Builder, and the icon displays on the site.

The Problem:

The chat icon is visible but not functional. Customers can’t click it to initiate a chat.

Has anyone faced a similar issue and resolved it?

Any help or guidance would be greatly appreciated!

#Einstein Bots  #Trailhead

#TrailblazerCommunity  #TrailblazerCommunityHelp  #Trailblazercommunitygroup

#EmbeddedServices  #Embedded Service Chat  #Embedded Chat

Still keeps loading when i click chat with an expert button eventhough i have added to the CORS

#Einstein Bots

Hi everyone,   

I’m looking for some clarity and real-world experience around Flex Credit consumption, particularly related to Einstein summaries and Prompt Builder.   

We currently have 3,000,000 Flex Credits as part of our Salesforce contract, and we’ve already consumed over 50% of them, with around 1,450,000 credits remaining. This level of consumption is happening much faster than we expected, and it’s not very clear to us what exactly is driving the usage. our contract is till Sept 2028   

At the moment, we’re struggling to find detailed diagnostics within Salesforce that explain where and how these credits are being used.

Specifically, I’m hoping the community can help with the following:

1. Is there a native way to view a detailed breakdown of Flex Credit usage, for example by feature, prompt, automation, or Prompt Builder?
2. How are Flex Credits typically consumed for Einstein-generated summaries? we use Einstin to summarize our inbound and outbount ticket calls. 
   • Is usage calculated per summary generated?
   • Does transcript length or prompt complexity affect consumption?
3. Do automatic or programmatic summaries consume Flex Credits differently compared to user-initiated summaries?
4. Are Flex Credits allocated for the entire contract duration, or do they reset annually?
5. For customers using Einstein summaries at scale, what would be considered a normal or expected burn rate?

We want to ensure we’re using Einstein and Prompt Builder efficiently and architecting this correctly, rather than over-consuming credits unnecessarily.   

Any insights, best practices, or lessons learned would be greatly appreciated.

Thanks in advance!