Voker | Analytics for your AI Agents

Hey, I'm Tyler, Co Founder and CEO of Voker, the Agent Analytics platform for AI product teams.

Voker gives you full visibility into what your users are asking of your agents and whether or not your agents are delivering, all without you having to dig through logs.

To demonstrate, let's pretend I'm an AI Product manager at HotelAI an online booking platform that lets users book hotels through their conversational agent.

I'm responsible for the usage and performance of our agent, so I'm going to take a look at my metrics.

I can see that the correction rate's been increasing over the past few weeks, which is a problem.

These are examples of users being frustrated or having to give the agent additional information to help it complete its task.

This is worth digging into.

I'm looking down here at intent categories, and I see that for version one of our agent, everything looks fine.

Resolution rates are good, and most users are asking about hotels.

But if I filter to version two, which I know was deployed early in April, I can see that there's a problem.

Looks like there's a new user intent that's showing up restaurant reservations, and the agent isn't performing well in those cases.

It has a 43% correction rate and only 52% resolution.

We need to figure out why.

As a PM, I want to put myself in the shoes of my users.

So I'm going to dig into a specific session.

Let's pick this first one where it looks like John didn't get all of his intents resolved.

We can see on the session detail page that John's trying to book a trip for his anniversary.

Over on the session timeline, we see that John booked a hotel and then he asked about booking a restaurant.

I'm going to jump to that part of the conversation.

I can see that Voker automatically detected the resolution that the hotel booking was successful.

That's great.

But the problem starts to arise when the agent offers to help with the rest of the trip.

Looks, like it offers a restaurant reservation.

So John agrees.

yeah, help me book that.

If we scroll through the conversation, we can see that the agent tries to make some tool calls, but it actually doesn't have a tool for booking the reservation after it finds restaurants nearby.

This frustrates John, understandably, because the agent over promised and we can see that was automatically detected by Voker as well as a correction.

I can link this conversation and the previous dashboard to my engineers and ask them to make a fix and add a tool so that the agent can actually book restaurants.

Now that our engineers have added that tool, I'm going to switch to the newest version of the Agent and confirm that restaurant reservations are working as intended.

With Voker.

The whole AI product team, from PMs to engineers, have the tools they need to build better agents.

You can too.

Go to Voker.ai today to sign up for a free account.