Customer support · Production

Smith.ai launches next-generation generative AI live-staffed chat powered by LLMs

The problem

Previous-generation AI chat was limited to linear, pre-scripted interactions that could not understand conversational context, forcing human agents to patch knowledge gaps rather than handle genuinely complex situations.

First attempt

Earlier AI platforms offered only IVR-like scripted flows because language variability was too complex for the models of that era; human agents were used to patch gaps in AI understanding rather than to add genuine value.

Workflow diagram · grounded in source
1
Customer chat initiated
trigger
“our 24/7 live-staffed web chat service that continues to offer the best of both worlds”
2
Business data ingested
integration
“Our AI ingests data from the website where the chat resides and can consider that information as part of its responses”
3
LLM generates contextual response
ai_action
“infusing the AI with just-in-time context from within the chat conversation along with other external content”
4
Free-form and playbook handling
ai_action
“Smith.ai's updated AI functionality handles free-form communication with structured interfaces like playbooks. They live alongside each other, and we can switch between them seamlessly. This means that our AI chat can engage in free-flow…”
5
Human agent intervention
human_review
“Smith.ai still leverages our North America-based agents to supervise the AI software and intervene when a conversation becomes overly complex or requires the human touch”
6
Chat response delivered
output
“providing coherent, accurate, and business- or product-specific responses — and delivered in a natural way”
Reported outcome

Smith.ai's new generative AI chat can handle more chats with more meaningful conversations, with human agents intervening only when truly necessary rather than filling AI knowledge gaps.

Reported metrics
Chat experience qualityfaster, more natural, and more helpful chat experiences
autonomous AI fulfillmentAI can successfully fulfill customer needs on its own more of the time
Chat volume and conversation qualityhandle more chats, have more meaningful conversations
Chat reply accuracymore accurate chat replies
Reported stack
LLMs
Source
https://smith.ai/blog/how-we-defined-the-next-generation-of-smith-ais-live-staffed-ai-chat
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

Smith.ai's new generative AI chat can handle more chats with more meaningful conversations, with human agents intervening only when truly necessary rather than filling AI knowledge gaps.

What tools did this team use?

LLMs.

What results were reported?

Chat experience quality: faster, more natural, and more helpful chat experiences; autonomous AI fulfillment: AI can successfully fulfill customer needs on its own more of the time; Chat volume and conversation quality: handle more chats, have more meaningful conversations; Chat reply accuracy: more accurate chat replies (source-reported, not independently verified).

What failed first in this deployment?

Earlier AI platforms offered only IVR-like scripted flows because language variability was too complex for the models of that era; human agents were used to patch gaps in AI understanding rather than to add genuine va…

How is this customer support AI workflow structured?

Customer chat initiated → Business data ingested → LLM generates contextual response → Free-form and playbook handling → Human agent intervention → Chat response delivered.

Related customer support cases