Talk Like a Human

680,000 support tickets.

Not a technology problem.

A language problem.

Insight

AI assistants fail the moment people stop feeling understood. When support tickets stack up, efficiency isn't the problem, empathy is. Users don't escalate to human agents because automation can't solve their issue. They escalate because they don't feel heard.

The business built automation. Users built frustration.

Truth: AI that sounds human earns human trust. The real risk wasn't sounding robotic. It was losing trust entirely.

The Creative Platform

Meet Iris: Automation with empathy built in

Rather than chase efficiency like every other AI assistant, we chased empathy. I built the voice, conversation patterns, and emotional logic that transformed Iris from a technical solution into a trusted brand character.

The challenge wasn't rewriting prompts. It was building a scalable voice system that could evolve with the product while maintaining humanity at enterprise scale.

Testing the Voice Before Writing the Code

To validate that empathy actually worked, we ran Wizard of Oz testing for the voice system. Our team manually responded as Iris while users believed they were interacting with automation.

This let us pressure-test tone, timing, and trust before a single line of code shipped. The results didn't just shape the voice. They shaped the product roadmap itself.

The Voice in Practice

Onboarding

Before: "Welcome. Please configure your account settings to continue."
After: "Hey, I'm Iris. Let's get you set up. It'll only take a minute."

Troubleshooting

Before: "Error: Unable to process your request. Please try again later." After: "That didn't work. Let's try again, or I can connect you with someone who can help."

Escalation

Before: "Your request has been escalated. A support agent will respond within 24-48 hours."
After: "I'm connecting you with our team. They'll have your full context so you won't need to repeat anything."