bland-ai

An AI-powered platform that automates enterprise phone calls with human-like conversational agents.

Bland AI

All right Audrey you're all set for this Sunday at 7 o'clock a.m.

With four players, is there anything else I can help you with? When I came across AI voice agent, the first application that came to mind is to build an AI receptionist for golf courses. They are the businesses I call most often when it comes to appointment booking over the phone. 

So today I'm going to show you how I build an AI voice agent in Bland AI that's able to handle appointment booking, cancellation, and rescheduling. Let's jump right into it. For the AI receptionist that I've created, you can see I have two branches here. One is for the booking and the other side is the cancellation. 

If we go down the cancellation pathway, I'm just collecting the booking information and then connecting that to a web hook where I'm searching for that specific event and proceed to delete that. Once that's completed, I'm asking the user if they like to reschedule. If they do, then I'm merging the remaining action with the booking appointment route. 

So for the booking side, I'm again collecting user preferences in terms of date and time they'd like to come in. Then I'm connecting.

That to a web hook to check the availability from my Google Calendar. Once I obtain that information or knowledge, I can then offer the user three different available time slots. Once the user chooses or did not choose, if the user did not choose, then it's simply a loop. I am then continue to offer more time slots for the user to choose from. If the user chooses the time slot, then I proceed to ask for more booking information. 

One thing about Bland is I find that static text works the best a lot of time. Instead of setting static text, I can also put in a prompt for the AI agent to kind of go off script and that don't always work out for me. A lot of time when I'm collecting multiple variables, I find the agents sometime like to be very creative and leave out one or two pieces of information. So this is when I tend to use static text as much as I can. 

Once that information is collected, all three variable will be sent over to the web hook. From there, we'll be able to create a Google event using those information and that brings us to the end of the pathway. 

Now let's take a look at the make scenario for all three web hooks. For the cancellation web hook, we have collected the user name and date, and those are the two variables we're going to use to search for that particular event on our Google Calendar. 

So we're going to set the day to the start date. The end date is going to be the end of day of that day. Then I'm searching through all the Google events I have on the calendar that matches with the username, and that is going to be the event that we want to delete. 

So this module is going to output that exact user event ID. Then I'm initiating this module to delete that specific event ID. Uh, so this is pretty straightforward, and I'm sending back a web hook response to Bland to let them know that API has completed. 

For the second web hook and finding available time slots from Google Calendar, this is the one that's a little bit tricky, and we need chpt to do a translation because the user can be giving a very different response. A lot of time, the caller is not going to say I like to come in on July 20th; they will likely say I.

Want to come in early morning next week or do you have anything on Saturday? So based on that response we need a translation in place. So the user can say anytime next week. Then we're going to output a start time based on Monday 12:00 till Friday at uh 11:59. Right? If the user says anything on the 19th then I'm outputting the entire duration for the 19th. 

So I'm letting the assist to know what is the current time as well as user preference. From there I should be getting a time range. So from this time to this time then it's going through two text parsers. So this parser is parsing out the start time and this is parsing out the end time. For both of these I also asked GPT help to provide the formula for me. Essentially let it know what's the output I need and what is the input I have and asking it to get me the expression so we can do the same. 

This parser is giving me the start date and this one is giving me the end date. So then in Google Calendar I'm able to set a minimum and maximum time that I'm looking for my free and busy information for. You're going to share the AI Agent Store with friends.

Have an array of results that's being outputed from this module. So the iterator is going to parel one result at a time from the array and then send it send it through this transformation and then come back to this for the next one. So it's going to go bing bing bing bing bing and then jump right back here for the second result and then gets reformat it aggregate it again and then come back for the next one. 

The transformation I'm doing in here is to format that into the preferred date format. Then I'm using the GPT module again. Now that I have the busy times as well as the range, I'm asking this module to output all the available times because I find even if I see to Bland the busy time it still doesn't do a good job of providing available time based on the busy time. So it's best to give your AI voice software the final result rather than having it compute any logic. Therefore, I'm using GPT to provide available time based on the busy time that we have gaed from the Google Calendar. 

One scenario I have to think about is if busy time is empty that means the

Whole time range is available so I generate the entire duration for the available times. Then I'm sending whatever response I'm getting it from here straight back to Bland. Here I can show you the previous operation that I've just done. The user response is I like to come in this Sunday and that is what's being captured in the user preference variable. 

Then I'm getting the time range of the entire Sunday from 12:00 to end of day. These text parser then output my exact start and end day. The Google Calendar is going to get me the all the busy time slots within that day and pushing busy time array through these transformation. The busy time array gets formatted in the date format I want and it's being aggregated back to a single stren of text. 

This GPT assistant is going to give me all the available times based on busy times as well as the time range for the final web hook. This one is also a very short one. I'm simply creating an event based on all the user variable that we've passed from Bland that include the username phone number date time which are what the. 

User have selected in the previous conversation and I'm setting that duration to 15. I think that's how long typically we space out each player or each player group. 

And then I'm also gathering the number of players just to try to make it realistic here for a golf course. Uh, so this is going to create that event on the calendar for me with the name and phone number being the event name. 

Then I'm sending the success response back to Bland so that it knows to proceed to the next note. Hey there, you have reached Toronto Valley Golf Club. How can I help you? 

Yeah, I want to cancel my tea time. What is the name in the date for the booking? It's for Audrey for September 21st. Let me check on the status of that booking for you. That is cancelled. Would you like to Res it for any other day? 

Yes, of course. When would you like to come in? Maybe this Sunday early morning. Let me check. This Sunday we have a few time slots available. We have 7:00 a.m.

10:00 a.m. and 1:00 p.m. Which one of those works best for you? 7 a.m. works. Can I have name, phone number, and number of players for the booking? Yeah, it's Audrey. 6 47 for four players. Got it. 

All right, Audrey, you're all set for this Sunday at 7 o'clock a.m. with four players. Is there anything else I can help you with? No, that's it. Great, see you then. 

I'm really excited to go to nearby golf courses to get feedback for the solution. Think of a Pro Shop that's 24/7, no delays, always pick up. If you like what you saw there, share the AI Agent Store with friends.

Note, that there may be slight differences in text and video. Watch video for the original content.

Bland AI Video