GenAI Appointment Booking: from Dogfooding to Production

One morning, I woke up to a surprising calendar entry from a potential client. It’s not the usual “Book an Appointment ” calendar invite, I had no memory of scheduling it, yet there it was, perfectly timed:

At the end of the summary message, it says: “Scheduled by SeaChat at chat.seasalt.ai ”.

Intrigued, I delved into the chat history and discovered that our SeaChat agent, embedded on our homepage, had seamlessly handled the entire interaction:

Here the SeaChat agent functioned as a “pre-sales” agent: it first answered customer’s questions. When there was intent detected:?

Prospect: “How do I reach out to your sales quickly?”

The SeaChat agent captured the request to do this “quickly” and offered the soonest available appointment slot:

SeaChat: “The soonest available appointment slots are on June 13 at 10:00AM, 10:30AM, 1:00PM, or 1:30PM in America Los Angeles time zone.”

This scenario exemplifies the “AIDA Buyer’s journey”:

• Awareness: Buyer discovered Seasalt.ai and its service
• Interest: Buyer started asking questions and the SeaChat agents answered all of them
• Desire: Buyer wants to “reach out to sales quickly” – this is a strong conversion signal!
• Action: the SeaChat agent took action and booked the Buyer!

The prospect returned later with more queries, which SeaChat handled efficiently:

Interest: How did the customer start talking to us?

Here’s how our product SeaChat supports various customer interactions. We installed SeaChat to 4 different places:

Customer Acquisition purpose:

• Homepage Integration: Engages visitors immediately.
• Product Landing Pages: Provides detailed product information.

Customer Service purpose:

• Product Wiki Integration: Answers queries faster than reading through extensive documentation.
• In-Product Chat: Offers real-time support.

SeaChat’s omni-channel capability allows customers to use their preferred messaging apps, enhancing accessibility and convenience.

The web interface is user-friendly and responsive, ensuring continuous support, even outside business hours.

In this way, even if we are asleep, the SeaChat conversation agent can always reply to customers with contextual information that is relevant to our company, and also to the customer’s needs.

Conversion excellence: how did the SeaChat agent book the appointment?

An appointment must be connected to a calendar system, and it must be reflected on both parties’ calendars.?

We built a Google Calendar integration that connected to the CEO’s calendar:

The SeaChat agent can find empty time slots from my calendar, and offer meeting times verbally. It doesn’t disclose any confidential information on the calendar, just available time slots as the user intended.

Once the customer confirms, the SeaChat agent will actually place the appointment to both parties’ calendar via their email addresses. A customer can later come back to modify the appointment too.

As a business owner, the best value I appreciate the SeaChat agent is the booked appointments that happen automatically during my sleep, or when I’m busy, or totally offloading this burden from my daily activities.

The conversational interface is more chatty and friendlier than a cold Calendly link. Some people don’t take it well and found the “here’s my calendar, go ahead and choose a time” rather cold and condescending.

Behind the scenes: long journey to production

There are quite a few challenges when it comes to a conversational interface for appointment booking:

1. Satisfy the Desire: Offer relevant and accurate information first to meet the customers’ need – otherwise why would they want to book a time to talk more?
2. Take the Action: Make appointment booking a smooth experience – with a lot of time and linguistic challenges!

Customer’s desire is all you need to meet first

Customers come to our website for various reasons – how do we make sure that all questions are answered, or when it comes to sensitive information, such as enterprise pricing, lead them to a more in-depth sales call?

We actually built the whole SeaChat system with the claim that anyone can launch a chatbot that knows about your company in 10 minutes. Various data importers are built: it reads 20+ file types of your company knowledge base, and also crawls websites via URLs and sitemaps.

SeaChat even transcribes company internal audios and videos to extract more knowledge!

We dogfooded this product ourselves by feeding SeaChat with our product wiki, websites, and refined our FAQs as customers ask more and more of it. That’s why in the initial chat, when the prospect asks about WhatsApp API, our relationship with Twilio, chatbot, and also customer services, our SeaChat agent successfully answered all of them – like a human being!?

Taking the action is harder than we thought

When it comes down to appointment booking, it’s a lot harder than anyone could imagine, if the appointment is done back-and-force via a language interface. Here are a few things:

1. Calculate time zone: we are headquartered in Seattle and offer appointment times in the Pacific Time Zone. But our users come from all over the world. We had to do the calculations to make sure we didn’t mess up the time.
2. Infer time zone: what if the prospect says “I’m in Sydney”? What timezone is it and how does it get reflected in our own time zone?
3. Infer relative dates: we had to ground SeaChat with the current date, so that it can infer dates like “2 weeks from now”, or “the week before Labor Day”.
4. Infer relative time: this one’s interesting, read on –

Even our own team cannot agree on time of the day

Is 5 pm considered afternoon or evening?

When I hear someone say “Last evening I worked on this pull request …”, I thought, wow, this person’s doing some overtime.?

Then in a week’s time, I kept hearing them working “evening hours” over and over again. I wondered what’s going on with our work arrangement?

Finally I found that in some people’s vocabulary, 5pm is already “evening”. Hell, in Seattle when it goes dark at 4pm in the winter, why wouldn’t it be? But I thought 5pm was still “afternoon”.

Eventually we adopted the weatherman’s standard definition of times :

Here are some interesting pick:

• Late morning is 9am to noon – but for some programmers who go to work at 9am or 930am, it’s actually early morning for them.
• Late at night is actually the next day: midnight to 6am.

I convinced our colleague that when they worked at 5pm yesterday, it’s “late afternoon”, not “evening”, and not “last night”. ??

A challenge that “x.ai” couldn’t solve

Now lots of people know Elon Musk’s xAI and GROK, but does anyone remember [email protected] ? When you wanted to schedule a meeting time, you can just “cc [email protected] ” and say “Amy can help us find a meeting time”.

10 years ago , the former x.ai tried so hard to solve the meeting scheduling problem. But AI, or machine learning at that time, and $23M more , was not powerful enough to overcome all the intrinsic challenges from human’s “I changed my mind again and we need to re-schedule” nature, or linguistic and cultural differences.

10 years later, the solution is still not as simple as just hooking up with a calendar system with a large language model.

SeaChat gave it another go, and it’s doing pretty well for us, because I have sales meetings to go to that were booked during my sleep. ????

Going beyond chat: book customers on the phone

SeaChat’s innovation doesn’t stop at chat. We’ve integrated phone call scheduling, where SeaChat can automatically book appointments from phone interactions, further simplifying the process and addressing a much bigger market.

This is our “one more thing” moment. See the following booked appointment. When someone called me, my SeaChat agent just booked them automatically, then sent me a calendar invite:

Here’s the transcription of the recording for the phone call in case you wonder what was said on the phone:

Simple and powerful, right??

Our journey to make appointment booking effortless and efficient has taken significant engineering effort. But the result is a system that operates 24/7, across multiple channels, and meets customers where they are, whenever they need us.

This is all driven by Seasalt.ai ’s BHAG (Big, Hairy, Audacious Goal):

By leveraging SeaChat, we’ve not only enhanced our customer engagement but also streamlined our sales process, providing a model that GenAI startups and professionals can appreciate and implement.

What is your company’s BHAG and what was your journey to bring dog food to production?