Introduction

Chris here. I'm a director of sales and business development in a remote company located in the U .S.

Why We Tested Voice AI in Sales

and today I'm going to talk about a field test that we did across the funnel.

Essentially it's a fancy way of saying we just experimented with voice AI in sales.

Company Context and AI Adoption

So a little bit of context, I joined Time Doctor back in July and obviously everyone was talking about how to use AI in day -to -day.

Engineering was already obviously using it to ship products faster.

Our product team was trying to figure out how we can use AI within the product itself.

Our chat team, well, our support team had already started using a chat agent and they had seen great gains.

So then we

Early Productivity Experiments

started discussing how can we use AI in sales. So one of the first things that we we tried out, similar to what Cosmas just showed us, was a lot of productivity agents.

We did agents similar for emails, we did agents that would help with coaching, again with the transcripts, and we also tried to use the tools that we already had to present enough information to our sales reps so they wouldn't need to do any research, thus save some time.

Targeting a Specific Revenue Bottleneck

What we're going to talk about today is how we try to use AI to direct it at a specific bottleneck that we had in terms of generating revenue.

Team Structure and the Speed-to-Lead Problem

So a little bit more context, our Time Doctor sales team is completely remote. We have agents across the whole world and they're what we call full sales cycle AEs.

So for instance, the example with Cosmas, he's an SDR and he would hand over the lead to an AE.

we don't have that we have one person doing everything so they're doing demos they're reaching out to leads if someone submits a request they would call them reach out to them to get a meeting booked so as you can understand they

have a lot of things that they need to do at the same time this is this is all great because they have a lot of control but one of the problems that this is

causing is when the lead comes in when they want to have a demo or when they submit a trial request we can reach out to them at the very minute that they submit that and that can be a problem for for a couple of reasons we call this

Why Speed to Lead Matters

metric the speed to lead and and why is it important so there's a lot of research out there to say that if you reach out to someone in the first five minutes after they submit a request you have up to 20 more times 21 more times chance of qualifying them versus calling them within 30 minutes and another very

important metric is that if someone if you're the first vendor that speaks to someone the chances of you closing that customer are 35 to 50 percent again this is when we're talking about multiple vendors not if you're just talking to one vendor so as you can understand this was this was really really important so

Building the Workflow

the first thing that we did was set up a workflow there's multiple tools that you can set up workflows we use the n8n if you're familiar with it but essentially

From Form Submit to Automation

we have a CRM called HubSpot so when someone submitted a form on our website it would trigger an automation flow in this n8n platform so it would take the information from the form fill it would parse it to make it simple enough and

Two Voice Agents: Booking and Qualification

then depending on if the prospect at the time had already booked a call or not it it would send them to one of two different AI voice booking agents.

So one, we called it Booking Beth, and if they hadn't scheduled a call, the goal was for that agent to book a call with them.

And the other one, if they had already booked a call from our calendar, then it would call them and essentially do Cosmaz's work or an SDR's work,

which would be to qualify them and create the notes and send these to the rep so they could be prepared and optimized for the call.

Retries, Failures, and CRM Handoff

now this after the call was made if we connected then it would just send the information into our CRM if it didn't connect it would try you know a specific amount of times and then it would at some point give up you know the number

Live Demo Overview

of times that we assigned it okay so now we're going to just going to go into a

Creating the Agent

little bit of a demo so we showed you the workflow now we're going to dive into how we created the agent itself and it's actually a lot less intimidating that it might sound it's very very similar to creating a GPT itself so here

are some agents that we already created essentially here is how you create a new agent but we'll go into one that we have created already so this is the selling

Purpose, Persona, and Prompt Design

Susan one the one that essentially would qualify the prospect it would ask engaging questions and then pass that information on to the rep and confirm confirm the call, obviously.

So here is where you would actually go in and add the prompt similar to how you create a GBT.

So for instance, here in this agent, we've identified what the purpose needs to be.

So this is Susan. It's a friendly assistant from Time Doctor.

The purpose is to confirm an upcoming demo that a prospect has already scheduled. And the goal is to ask a few clarifying questions to understand the current situation.

Then Then we dictated what we wanted the voice or the persona of it to be.

So we want it to be friendly, upbeat, and helpful, but also professional and enthusiastic.

You can dictate what you want the speech characteristics to be.

Do you want it to be natural, conversational? Do you want it to be more strict, more short? Again, this is something that you can dictate within the prompt.

Branching Logic and Clarity

We went into different kinds of branching logics as well. So if a prospect asks a specific question, then it would go into a different logic of questions or answers

And then yeah, essentially the clearer you are similar to a GPT the better output you would have from the agent itself

So this might be tricky, but I'm gonna try and show you the agent, but we have the microphone And this microphone, so we'll see how it works

Sample Booking Call

Hi, this is Beth calling from time doctor. Am I speaking with Raphael? Hi this is Rafael. Perfect thank you.

I just want to let you know this is an AI call from Time Doctor. You recently tried to find a time with our team but couldn't find an option that worked. I'm calling to make sure we can find a time that fits your schedule.

Do you have a minute to check a couple of options? Yeah happy to. Great thank you.

What specific day and time were you hoping to schedule your 30 -minute demo? I could do Thursday at 3 p .m. and Friday at 1 p .m.

Hello? Got it. Thank you. So you're available on Thursday at 3 in the afternoon and Friday at 1 in the afternoon. Yes.

I'll make sure our team checks those slots right away. Before I let you go, is there anything you'd like me to pass on to our team ahead of your upcoming call? Could I actually just ask a question? Could you remind me what Time Doctor was before I let you go?

Of course. Time Doctor is a tool that helps teams track time, manage productivity, and understand how work hours are spent. It's used by businesses to improve efficiency and get clear reports on how time is used during the workday. Is there anything else you'd like to know or anything else I can help you with?

No, I think that's good. Thank you. Thanks for your time today, Rafael. Have a great day.

Results and Lessons

So if we did the actual agent, the more conversational one, it would have actually been more impressive, but I don't want to take any more of your time. We just did the simpler one, which was the booking one.

Did It Replace Sales Reps?

if it worked really well i would be asking you are sales people going to lose their jobs the the answer to this was obviously no even though this was an experiment that that worked really well the actual agent in the wild when we released it it worked pretty well this was a problem that

Performance in the Wild

happened in real situations the lag and the latency and sometimes it created awkward conversations but for the most part for about 80 to 90 percent of the time it was actually pretty efficient it was pretty quick it asked really thoughtful questions and it passed on really great notes to the AE without needing to have any person there at all

Outcome Metrics

but the reality was that the results even when the call was connected and even when the AI voice was working well still about 70 % of the calls were unsuccessful and when I mean a successful they didn't manage to complete their goal they didn't manage to get the meeting booked or they didn't manage to have a conversation with the real person and grab that information and send it to the rep.

The Core Challenge: Willingness to Engage

The reason, it wasn't because the agent wasn't working, as we saw today. It was because people didn't really want to engage with an AI.

In this situation, the person that we were trying to speak to, they didn't choose to speak to an AI.

So most people, when they actually finally understood that it was an AI, they either requested to speak to a real person or they simply hang up. up. In 42 % of the cases, they hang up and they didn't want to continue on with this.

Human Expectations of Voice

So even though this was a failed experiment, there was quite a few learnings and I think because we as humans tie voice as such an integral part of communication, we're not really ready to engage with AI voice in that kind of sense.

You know, having it sort of of like another person on the other end of the line.

We engage to use AI in our daily productivity, you know, all the time, as we all know, but I think that's a place that we're not there yet.

What Must Change for Adoption

For me, one of two things needs to happen if we're going to engage with a typical agent like this in the future. It either

needs to be unrecognizable from a real human in the sense that it will call us and we're not going to be able to tell if it is a human or not or we need to

choose to engage with that tool ourselves so for people in marketing for instance or for people in sales if we can have the prospect actually choose to engage with that AI most likely it will be okay but for now no if you're not in

Other Practical Uses of Voice AI

sales there's still multiple ways that you can use you know voice AI chat GPT obviously has a conversation model that you can use but you could also use 11

11 labs to maybe practice an interview before you go on to an interview maybe there's a specific subject that you want to be better at in a conversation you know maybe politics maybe art you could use a voice AI to practice live and I

promise it's gonna be much better than the demonstration today with a lot less

Conclusion

technical difficulties so yeah that was it thank you