Talk Like a Human

The chatbot was there to help.
Yet, people kept asking for a person.

Role: Conversational Designer · Brand Voice Strategy · Copy
Client: Fortune 500 tech company

Client under NDA. Visuals recreated to reflect strategy, voice, and results.

Problem

Support needed support.
A global tech company's sales and support teams were drowning in tickets. Each party increasingly frustrated with the other. Associates needed answers fast. Support needed the volume to stop.

Insight

A helpful answer starts with a little understanding.
People were escalating requests the chatbot could handle. Its mechanical responses left them feeling unheard, so they turned to a person instead.

Idea

Teach Iris some manners.
I developed Iris’s voice around how a helpful support person would respond. “Your entry does not register” became “I’m sorry, I didn’t catch that. Can you repeat the question?” Iris took responsibility for the misunderstanding and gave the person a way to continue.

Before development, we used Wizard of Oz testing, with a human supplying Iris’s responses behind the scenes. Those sessions shaped more than the wording. We revised the menu options, reordered questions and changed how Iris handled misunderstandings, letting the way people actually asked for help shape the conversation.

See how Iris guides someone through a support request, from the first question to the next step.

Iris introduces itself and gives people two ways to ask for help: choose an option or type a request.

I paired each voice principle with examples of what Iris should and shouldn’t say.

I wrote variations for first-time and returning users, including greetings that acknowledge the time of day.

80%

Fewer tickets sent to support technicians

200+

Conversation touch points written