Current Costs & Options With OpenAI Using AI Agent

[ROLM]

We've been testing the AI Agent on our internal demo system and have noticed that price in OpenAI per call, seems to be higher than expected. Are there any options available to try and minimize this cost?

Additionally, we've had a few instances while testing, where the AI Agent will take a minute or two to transfer to the desired person or department. In one instance, the AI Agent hung up after it was relayed information from the caller. With call screening, the AI Agent will prompt call screening questions however, the call does not get screened to the recipient. Essentially, we are looking for any tips on to configure the AI Agent to perform flawlessly. Thanks!

 

[KyriacosS_3CX]

Hi,

To get a better picture, some extra info might come in useful.

Cost question, what's the agent's config? Anything outside defaults? What languages?

Transfer delays and other glitches, anything happened while system was in verbose? General and AI logs might give some hint if anything was off on the PBX side - also network load/server load at those point in time would be interesting to know.

Call screening, do you mean it does not get transferred, or the recipient doesn't get the screening info?

 

[ROLM]

What's the agent's config? - Receptionist, no custom instructions, one knowledge source, no language switching.

Anything outside defaults? - No custom instructions. One ring group set in routing behavior with basic description. Allows extension routing by name. No busy check before transfer.

What languages? - English

Call screening, do you mean it does not get transferred, or the recipient doesn't get the screening info? - Seems to just do nothing intermittently after saying it is going to transfer the call. The caller just hangs in silence.

Anything happened while system was in verbose? - Current logging is set to "All", but we did notice this while running some tests: 'C:273 Failed to build divert target EndCall.Unknown; fail reason = TerminatedByRule'

 

[nikolascx]

We've been testing the AI Agent on our internal demo system and have noticed that price in OpenAI per call, seems to be higher than expected. Are there any options available to try and minimize this cost?

Thanks for your feedback
What is your cost per call?
Look at the log and the tokens should be printed out.

I think best is to set the log to verbose, stop the 3CX AI Service, Delete the log file, Restart the service again and replicate the one call that fails. Once you do that, send us the log in PM.

C:273 Failed to build divert target EndCall.Unknown; fail reason = TerminatedByRule -> This means that the server intended to divert a call but before it could start, another event occured that tore down the call. Something unpredictable caused by your environment (Either config, or Phones, or trunk or setup)

For now, to measure the costs, let's make sure that the test you make, transfers to an extension or an IVR that is local and fully available. Like this we simplify the scenario and get a base cost for a simple call. That will give you a solid baseline.

The log will tell us if you have something that can be optimized.

At your disposal