Customer support · Production

Wayfair's Wilma LLM copilot helps customer service agents resolve issues 12% faster

The problem

Customer service agents had to navigate complex chat interactions while applying hundreds of Wayfair policies and maintaining empathy, a task that became especially challenging at peak shopping volume.

First attempt

The first version of Wilma used a single long prompt triggered by one button; the LLM frequently followed irrelevant examples, agents felt they lacked control, and during resolution negotiation the model sometimes offered inappropriate resolutions.

Workflow diagram · grounded in source
1
Agent selects action type
trigger
“The agent clicks a button to select what they want Wilma's help with (Discovery, Resolution, Empathy, or Give Me a Minute).”
2
Routing LLM selects prompt template
routing
“A prompt template is selected based on which button the agent clicked, some business logic, and the analysis of a routing LLM.”
3
Template filled with real-time data
integration
“The prompt template is filled in with real-time customer, order, and product information that is pulled from Wayfair's systems.”
4
Multi-LLM conversation analysis
ai_action
“The routing LLM identifies if we are in a negotiation situation. The proposal LLM identifies resolutions that have already been proposed. The suitability LLM decides which resolutions are reasonable to offer. The current resolution LLM i…”
5
LLM generates draft response
ai_action
“The LLM generates a response using the filled in prompt template.”
6
Response validated and delivered
validation
“The response is checked for appropriateness, information is added, and the final output is delivered back to the agent.”
7
Agent reviews and sends
human_review
“The agent is now free to use, edit, or not use the message, as they see fit.”
Reported outcome

Agents using Wilma address customers' needs 12% faster and adhere to Wayfair's customer policies between 2% and 5% more, depending on issue type.

Reported metrics
Customer issue resolution speed12% faster
Policy adherence improvementbetween 2% and 5% more, depending on issue type
Reported stack
GeminiGPTJinja
Source
https://www.aboutwayfair.com/careers/tech-blog/the-evolution-of-wilma-wayfairs-customer-service-agent-copilot
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

Agents using Wilma address customers' needs 12% faster and adhere to Wayfair's customer policies between 2% and 5% more, depending on issue type.

What tools did this team use?

Gemini, GPT, Jinja.

What results were reported?

Customer issue resolution speed: 12% faster; Policy adherence improvement: between 2% and 5% more, depending on issue type (source-reported, not independently verified).

What failed first in this deployment?

The first version of Wilma used a single long prompt triggered by one button; the LLM frequently followed irrelevant examples, agents felt they lacked control, and during resolution negotiation the model sometimes off…

How is this customer support AI workflow structured?

Agent selects action type → Routing LLM selects prompt template → Template filled with real-time data → Multi-LLM conversation analysis → LLM generates draft response → Response validated and delivered → Agent reviews and sends.

Related customer support cases