Blog

ProjectBlog: Voice AI in high-touch sales

Blog: Voice AI in high-touch sales


An Alexa skill that does something useful

This is the story of finding where and how voice AI could work in the sales funnel of a high-touch business.

Air Charter Amy

This April we published Air Charter Amy, an Alexa skill for use in a sales role. It’s no secret that voice assistants are used mostly to play songs, turn on lights, and set alarms.

Conversation AI is today where ecommerce was 20 years ago and today’s innovators are tomorrow’s market share leaders.

Amy is a combination of my interest in conversation AI and my ownership of Planeviz, the private jet charter broker for which she was made. Being your own executive sponsor can expedite a project but doesn’t assure success.

An opportunity?

Using voice AI for sales requires slotting it into the sales funnel where it makes sense. And where it makes sense is where it adds value to the both customer and to the business.

Automation routinely adds value to a business in terms of efficiency metrics but not necessarily in terms of the customer experience. Telephone IVR systems are the poster child of this phenomenon.

Voice AI has to truly improve the CX if you want people to try it and come back.

With customers spending 5-figures for a few hours of flight time, private jet charter is a high-touch business. The first question to be answered then is where in the jet charter business model does voice AI makes sense? Where can it add value to both the CX and the business?

Yes, a red carpet.

One common private jet use case is the “on demand” flight in which a person rents the jet for a specific trip. Unlike booking an airline flight, there is significant manual work involved in reserving a jet for an on-demand trip.

The process starts with the traveler’s request for a quote, typically made through a website form or a phone call. There are many jet charter companies and it’s in the customer’s best interest to shop around and negotiate the desired combination of plane, price, and confidence.

On the other side of this shopping experience are sales agents who field the quote requests. It’s a very competitive market and many quotes produce no further action by the shopper. In other words, there’s plenty of low value busywork.

The opportunity to use voice AI here lies at the very top of the sales funnel where it can potentially improve the user experience and augment the existing business process–with minimal impact on the high-touch aspects of the business.

Business process integration

Amy does one thing for now: engage in conversation to gather trip information and contact details, then send this information to the sales agent and to the prospect.

To get the prospect’s contact details we had the choice of requesting Alexa profile information or connecting to a credentialed, external account, such as Google. We chose to ask for the Alexa first name and email contact details.

Accessing a user’s contact details (with their permission) enables us to use the person’s first name in the dialog and to welcome that person back by name if they return in the future.

With the trip details in hand, we package it in an email that is sent to the sales agent and to the prospect. This information can be sent to a CRM system too.

For bigger companies it would make sense to send the quote request to a Slack channel or another team communication tool to leverage the advantages they offer over email.

Voice is not the right engagement method for evaluating and deciding among many choices.

We stayed away from making Amy a transaction-capable voice assistant for two reasons

  • Actually getting the quotes (instead of fielding a request for quotes) moves automation farther down the sales funnel into the high-touch part of the sales process.
  • Voice AI is not the right engagement method for evaluating multiple charter quotes.

Branding

The Amy graphic has been the face of the Planeviz brand for years and adds some personality to the confirmation email.

For Amy’s voice we are using the stock Alexa voice. Alexa has a voice markup language that can be used to make her sound more human-like, maybe we’ll give her a verbal makeover in a future update.

Other voice options were hiring a voice actor to record the dialog or generating a synthetic voice based on a human voice. Being version 1.0, those two didn’t make sense.

Dialog

I modeled the dialog after a typical request-for-quote phone call with a human.

People respond to questions in various ways and we added some of those possibilities to her intelligence. For example, Amy asks “when are you leaving?” and a prospect might reply “tomorrow afternoon” or “May 1” or “April 30 around 7am”.

We also rotate through multiple variations of the questions she asks to keep the dialog fresh every time someone speaks with her. She might say “where are going?” or “flying to?”

There is much talk in voice experience circles about building an interaction model that can handle conversational diversions. A person gets to a point in the dialog then goes off topic or back to an earlier point to change something or to ask a question related to that earlier point in the conversation.

We did that in an earlier IBM Watson chatbot and learned it adds a level of design and development complexity that I don’t feel is necessary for Amy, at least as she now is.

If the user makes a mistake or wants to make a change then he or she can just say “start over.” The customer experience cost of starting over is not that high with Amy’s fairly quick conversation.

At first I was writing the dialog in a Google Doc. That was fine for me but hard for my developer to keep straight while coding it. We adopted a tool called draw.io (nice tool, integrates with Google Drive) to map the interaction model. The blue and gray are dialog, the rest is logic and functions.

Developing on Alexa

Why Alexa? To expand our development repertoire and to leverage Amazon’s continuous promotion of Alexa to the public.

Despite the Alexa PR bombardment I just read a Microsoft study that broke down the percentage of respondents who have used a voice assistant as Siri 36%, Google Assistant 36% and Alexa 25% and Cortana 19%.

We plan to get to Google Assistant soon but if I could turn back the clock I would start with that instead of Alexa. It’s on 1 billion Android phones which are always with their owners, versus Alexa’s 100 million stay-at-home smart speakers.

In practice, if Watson didn’t understand something we just trained it to be smarter.

Alexa is a predominately B2C voice platform. By way of comparison, IBM Watson is a B2B suite of cognitive services. A fundamental difference between them is access to the natural language model.

With Watson we built our own NLU model using our own data. With Alexa, Amazon owns and controls the natural language model.

In practice, if Watson didn’t understand something we just trained it to be smarter. With Alexa, all we can do is add the expressions people might use in conversation and leave the rest to Alexa.

This helplessness manifested with our invocation name Air Charter Amy, which Alexa would routinely reject in favor of Air Cheddar Amy or Air Trotter Amy followed by the inevitable “sorry I don’t know that one.” Not the best outcome for demoing your cool Alexa skill.

Mangling the name wasn’t a simple matter of not speaking clearly or a thick accent; it had to do with the algorithm Alexa uses to surface skills. It seems a cousin of “you might like” from the Amazon store is used for Alexa.

Fortunately, the Alexa developer support team was able to fix this on their end. Coincidence or not, shortly after this hiccup they changed the skill certification requirement to only one invocation (“open air charter Amy”) instead of three.

One advantage Alexa and its B2C competitors has over Watson is that we didn’t have to make or connect to a UI.

Voice First?

Just 20 years ago ecommerce was a novelty, no enterprise would entrust its data to “the cloud”, and streaming media was, well, it wasn’t. Early adopters Amazon, SalesForce, and NetFlix jumped in and are today’s market leaders.

Publishing a Air Charter Amy is a first step. While we made her for the jet charter business, her information gathering capabilities could be used in other business applications that use calendar and contact information too.

Now comes the fun of introducing it to organizations that might benefit from what it does and seeing where it goes from there.

Resources

Source: Artificial Intelligence on Medium

Leave a Reply

Your email address will not be published. Required fields are marked *

Back To Top
a

Display your work in a bold & confident manner. Sometimes it’s easy for your creativity to stand out from the crowd.

Social