Sam Liang

by itself.

But it's still early.

There's still a lot of work to do.

That's where you asked me earlier about challenges.

A big challenge to overcome is still how do you build the voice agent so that it can conduct a call effectively so that it can achieve certain objectives.

ask questions, answer questions, and maybe lead the user to convince the user to do something.

So how do you

build that agent.

Part of that is to really model human conversation.

We can train that agent by giving it tens of thousands of sales calls.

If it's a sales agent, for example, it's how a good sales rep convince the customer to buy their product

then if you give it enough data, the AI agent could do a pretty good job too, or can even outperform a human agent.

Yeah.

We actually already, another project we have, it's a real-time sales coach.

It's actually just basically what you said.

If the customer asks some tough questions,

If the sales rep is relatively junior, he or she cannot answer a lot of questions effectively yet.

We have this real-time sales coach that actually listens to a customer, hears the objection, and then the sales coach will pop up an answer on the screen in real time.

So if you're doing it on Zoom, the sales rep doesn't read the answer back to the customer.

So it's almost like cheating in an exam.