Customer support · Production

Extendly boosts AI chat deflection from 5% to over 30% with Botpress

The problem

Extendly's in-house chatbot resolved only 5% of conversations without human help, making 24/7 support increasingly costly: every new agency client meant more agents, more overhead, and more complexity.

First attempt

The legacy in-house AI system could not keep up with demand and was difficult to integrate with new platforms or LLM models, requiring a full development team and increasing total cost of ownership without delivering additional value.

Workflow diagram · grounded in source
1
Support request arrives
trigger
“every support request starts with an AI agent built on Botpress”
2
AI handles routine inquiries
ai_action
“These agents handle subscription questions, technical issues, onboarding help, and more — all without taking time from an employee”
3
Complex tickets routed to humans
routing
“gave Extendly's team the ability to focus on more complex tickets and high-touch customer relationships”
Reported outcome

After switching to Botpress, chat deflection jumped from 5% to over 30% — a 500% improvement — and the volume of chats handled by human agents fell by 30%, allowing agents to focus on complex issues.

Reported metrics
Chat deflection rate (before)5%
Chat deflection rate (after)over 30%
Chat deflection improvement500%
Reduction in chats handled by human agents30%
Show all 6 reported metrics
chat deflection rate (before)5%
chat deflection rate (after)over 30%
chat deflection improvement500%
reduction in chats handled by human agents30%
agencies servedover 1,000
potential end-usersup to 400,000
Reported stack
Botpress
Source
https://botpress.com/customers/extendly
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

After switching to Botpress, chat deflection jumped from 5% to over 30% — a 500% improvement — and the volume of chats handled by human agents fell by 30%, allowing agents to focus on complex issues.

What tools did this team use?

Botpress.

What results were reported?

Chat deflection rate (before): 5%; Chat deflection rate (after): over 30%; Chat deflection improvement: 500%; Reduction in chats handled by human agents: 30% (source-reported, not independently verified).

What failed first in this deployment?

The legacy in-house AI system could not keep up with demand and was difficult to integrate with new platforms or LLM models, requiring a full development team and increasing total cost of ownership without delivering…

How is this customer support AI workflow structured?

Support request arrives → AI handles routine inquiries → Complex tickets routed to humans.