Giving Voice to A Virtual Assistant


Develop a virtual assistant to gauge customer interest in engaging with an automated conversational interface for help with common post-purchase issues


  • Using real conversations with live customer service agents to better understand common customer problems and the language used to explain and resolve them
  • Creating natural and intuitive conversational flows that seamlessly connect regardless of the path a customer ends up taking
  • Writing conversational copy to help customers solve their problems, either by providing direct educational content or by redirecting them to the appropriate self-service flows
  • Establishing rules around the virtual assistant’s voice and tone, and leveraging Wayfair’s overarching voice pillars established by the brand team
  • screenshot of a word document showing a potential conversation between a virtual assistant and a customer about changing an address
  • Spreadsheet outlining the copy and sequence of information for a virtual assistant skill
  • green spreadsheet showing potential virtual assistant errors and the associated response copy
  • screenshot of a word document titled "virtual assistant voice and tone brief"



The post-purchase team hypothesized that developing an effective automated help option would bring value to both customers and the business. Wayfair caters to a wide customer base with different preferences for order help. Some prefer a high-touch experience and will almost always call into customer service for assistance, but others want to avoid calling at all costs. Providing guidance that doesn’t involve a live customer service interaction solves a major customer problem for the latter group. From a business perspective, pointing more customers toward an automated solution would result in cost-savings by deflecting customer contacts. 

Role: Content Strategist

Skills & Deliverables

  • Conversational AI Design
  • Voice and Tone Guides
  • UX Copy
  • Stakeholder Management


The team began in the discovery stage, and first obtained transcripts from real customer service call. We used these transcripts to pinpoint the most common customer problems, which were then used to determine the first ten customer intents for the virtual assistant (VA) to recognize.

As a little background, the VA is programed to recognize what the customer is asking and what they’re trying to accomplish. These are considered “intents.” For example, a customer might want to return their item. The VA recognizes this as the “return” intent and can provide the appropriate information to help them start a return. The VA can also perform “skills.” A skill is an interactive conversation between the VA and the user. The VA asks a question and the user inputs text or selects an option via a button. Some example skills might be to greet the customer and ask them how it can help, or to confirm with the customer that it it’s accurately understood their intent.

Using real customer questions to identify intents meant we’d be more likely to recognize and solve most problems. We were also able to see the natural flow of these conversations and track the language and phrases both parties were using. I took these insights and used them to map out example conversations with the VA, and then worked with engineering to determine how to order the skills and intents.

Writing the copy for the VA was a bit like writing a choose your own adventure book. I used a lot of if-then statements and made sure to consider every possibility in order to maintain a continuous conversation. I also identified where we could reuse copy to maintain consistency and better manage our content long term.

It wasn’t in scope for launch to explore whether we to give the VA any kind of persona, but I did believe it was important to establish its overall voice and tone. My final deliverable was a voice and tone brief that serves as a reference point for anyone who needs to write content for the VA.

An early working version of the virtual assistant during development


The first version of the virtual assistant had a higher-than-expected engagement rate, proving that customers are interested in interacting with this type of help option. We also saw that 94.5% of customers who engaged with the virtual assistant went on to have a conversation with it, suggesting the copy was clear, compelling, and helpful. The team is continuing to iterate on the virtual assistant and make the customer experience even better.