Webhook Triggering Issue Help Needed

[olive breach]

Posting again (2nd day) with the hopes that someone from 11 Labs will jump in. I have a client project that needs to be resolved and I am STUCK!

I have a Webhook that isn’t triggering at the intended time. Here’s the situation:

Initially, the tool to check calendar availability was being called only after the customer mentioned their preferred day. This led to awkward pauses, as the AI would stop talking for 3 to 9 seconds while waiting for the tool to fetch results.

To resolve this, I re-engineered the prompt so the tool would trigger at the very start of the conversation, running in the background. The goal was to preload the calendar data, allowing the AI to have knowledge of available days and times before the customer shared their preference. This should have eliminated the delay and improved conversational flow.

However, despite these changes, the tool still doesn’t trigger at the start of the call. It continues to wait until the customer mentions a preferred day, which reintroduces the awkward pauses.

I’d greatly appreciate help in diagnosing and re-engineering the prompt or Webhook to ensure the tool triggers as soon as the call begins. Any guidance on how to implement this effectively would be incredibly helpful.

[zenith dew]

Have you tried updating the tool description?

[olive breach]

Yes. The problem is when the Webhook triggers the AI simply stops speaking. It's doing the check of the calendar availability, but cannot continue to have a conversation while it's doing it. So if I could have it loaded behind the scenes while the callar is talking that solves the problem. I've tried to make it execute as a first event, tried changing the prompt, it triggers when it's called for. I can try re-engineering my prompt again, but I'm on iteration 39 and no matter what I try it fails.

Wait @zenith dew Are you saying the description itself adds functionality? It's not just for our internal notes?

[zenith dew]

No idea 100% i'm going off the wording..

But I have not used it myself.

[olive breach]

I just tried. It doesn't change it.

Here's what I cahnged it to: Proactively fetch available appointment times via webhook at call initiation. The webhook retrieves the next 10 days of availability, returned in a comma-separated format like "25 Jan 9:30", "26 Jan 10:00". This pre-fetched data enables faster response times when users specify their preferred appointment date.

[zenith dew]

How about "Listen for {callDateTime}" at the beginning of the call.. or something along those lines
Then have that in your prompt {callDateTime}

[olive breach]

so when you say "listen for" do you mean one prompt is listening or sensing the trigger of the other?

[zenith dew]

something along those line, again just my assumption.. In your LLM prompt you are asking it to call this webhook right? So I would try making it is explicit as possible.

[olive breach]

Can I DM you my exact prompt @zenith dew ? Perhaps I’m missing something.

[zenith dew]

I've seen it above, I really dont have any idea how this product works i'm just spitballing ideas at you..

[olive breach]

Are there any 11 labs coders monitoring this group? Are there any company admin who can direct me to someone who can help?

[olive breach]

I’m trying to figure out where to place my temporary fix. Because it immediately triggers the calendar it goes silent, but having the AI say while I checked the calendar, I may be quiet for a moment or two doesn’t work because it won’t speak after the person says the day, ha but that just gave me idea. What if I change the trigger language? I can have the trigger language happen earlier in the conversation when the AI asks what the issue is. It can then trigger it because it stops talking at that point and while the person is talking, it will do the webhook call maybe

[zenith dew]

There are some devs who pop in when they can but they have their hands full with new releases etc, so they don't have a lot of time to be able to come here and help unfortunately.

I think it all comes down to experimentation, as everyone has a different set up and use case so there really isn't any one stop solutions..

but there will be an event today hosted my one of the devs wh is heavily involved in this product https://discord.com/events/1066739690436313158/1315994125828624504