The changing faces of chatbot avatar design

How to design the right avatar for your corporate chatbot

Selecting the right chatbot avatar
  • AI chatbots will save businesses an estimated $8 billion annually by 2022
  • Companies can choose a range of chatbot avatar forms and likenesses for their chatbots to fit audiences and use cases
  • x and emotional intelligence are the main determinants of chatbot avatar success

Nearly a decade ago, the team that designed Siri had a decision to make. Should they give Siri a humanoid avatar to go along with her voice interface? Or should they render her on screen as she remains today: a small undulating orb, hovering quietly in the background?

Ultimately, they ditched the avatar, but it was a more out of gut instinct than empirical research. As Shawn Carolan, one of Siri’s early investors, told Forbes a few years later, “If Siri looked like me or you, maybe our friends would use it and like it, but somebody in a different part of the world would not. So, we figured if it was just a name and a nice voice, people could put their own face on it.”

Many more companies today are facing the same dilemma, only now the stakes are higher and the choices more complex. AI chatbots and assistants are no longer just the province of giant tech companies like Apple, Amazon, and Microsoft. They’ve evolved into a category of business applications that’s growing 25% annually and estimated to save businesses $8 billion per year in costs by 2022, according to Juniper.

Chatbots powered by AI are already handling phone and even video conversations with customers, helping to reduce the cost of support. In the future, virtual assistants could wind up handling most basic support requests, experts say.

“Two to three years ago I wouldn’t have expected the kind of quality of interaction and the expectation that AI will understand you,” says Austin Tate, professor of knowledge‑based systems at the University of Edinburgh and director of the Artificial Intelligence Applications Institute. “I am surprised now when it doesn’t.”

Because virtual assistants are getting smarter and more capable, companies are assigning them tasks that only humans could do in the past. This raises new questions: How do companies make customers and employees feel comfortable interacting with virtual agents? Should they set them up as faceless tools with customizable voice and gender? Or should they design ethnically ambiguous, 3D human avatars such as Autodesk’s Ava, with thousands of programmed facial expressions designed to respond to every nuance of your interactions?

There isn’t a set playbook, at least not yet. Companies are following their own cues and knowledge of their customers’ needs to figure out the right form factor for each virtual agent.

Chatbot avatar diversity

Today’s AI avatar population are a diverse bunch. When it comes to a defining characteristic such as gender, many companies have moved away from an older standard of deferential female characters or voices toward a broader mix of genders. Others offer genderless avatars or male‑female tandems such as’s virtual assistants Andrew and Amy.

Some AI technology providers, such as New York‑based IPsoft, have adopted a hyper‑realistic approach. In 2014, IPSoft launched Amelia, an AI chatbot platform fronted by a 3D avatar that they produced by digitizing the features and movements of real‑life model Lauren Hayes.

Similarly, the Pasadena, CA‑based startup Oben created PAI, an AI that allows individuals to create detailed virtual replicas of themselves that can interact with their contacts or customers when the flesh‑and‑blood person isn’t around. The company is also developing AI‑driven celebrity avatars that might one day interact with fans.

Other companies use animated avatars for very grown‑up tasks. Baidu’s Melody looks like a character from a Pixar film, but she helps Chinese physicians diagnose patients.

The range of use cases help explain why IPsoft has adopted a mix‑and‑match product strategy, says Jonathan Crane, the New York City tech company’s commercial director. IPsoft sells chatbot technology to businesses in more than a dozen industries. The tech includes a 3D avatar called Amelia along with the cognition technology and algorithms that support customer interactions.

If they wish, IPsoft customers can replace Amelia with simpler avatars that reflect their brands. One medical supply company chose that option to provide IT support for its 45,000 employees. As a result, average call times have plunged from three minutes to 30 seconds with a 95.9% success rate.

For other companies, the hyper‑realistic approach has proven a better fit. AI company Soul Machines designed the ultra life‑like Cora for NatWest, a major retail and commercial bank in the UK. Digitally modeled after a female engineer at Soul Machines, Cora handles over 200 banking queries, freeing call center employees to handle more complex issues. Ironically, the more human‑like avatar proved especially appealing to bank branch customers.

The smarter chatbot avatar, the better

When it comes to chatbots, brains trump beauty. The cognitive abilities of chatbots have advanced significantly in recent years, thanks in part to ever‑cheaper GPUs that can handle the insatiable compute demands of machine learning algorithms.

Users increasingly expect chatbot experiences to match the quality of interaction with human agents. Advanced natural language processing isn’t enough, experts say. To hold their appeal, avatars must be able to interact in context. Chatbots need to be able to switch context—to stop helping a person with one issue and instantly address another—and then go back to the original issue. “Human dialogue is complex and nonlinear,” says Allan Anderson, IPSoft’s director of enterprise solutions.

Context‑switching can help users complete transactions in lengthy processes such as applying for car insurance. Conventional chatbots can only follow a scripted process. If a customer wants to change an earlier option, such as selecting third‑party insurance or adding another driver, a context‑switching AI can handle it without handing off to a human operator, which frustrates customers and spikes costs.

As for those human agents or NatWest’s bank tellers, they needn’t worry, at least not yet. Research shows that when faced with more complex or big‑ticket interactions with chatbots or AIs, people still want people to help them out.