Customer support · Production

Lessons learned from building over 50 chatbots on 5 continents

The problem

Most chatbots fail due to poor implementation: bots trap users in 'not understood' loops for out-of-scope questions, generate false positives that mislead users, and respond unhelpfully rather than routing them toward a solution.

First attempt

Common failure modes include bots asking for information users already provided, rephrasing loops that can't resolve out-of-scope questions, and dead-end 'Sorry, I don't know' responses that add frustration rather than helping.

Workflow diagram · grounded in source
1
Define scope and train NLP
trigger
“The team agrees on an information architecture, defines the bot's knowledge and responses, trains the natural language processing (NLP) model”
2
Add 99-intents for out-of-scope
routing
“create what I refer to as "99-intents". These are intents that capture categories and intents that aren't in scope. Their main purpose is to attract out of scope questions and define the correct next steps for users”
3
Launch at ~70% recognition
output
“When you first launch your bot, your NLP model will recognize roughly 70% of incoming, in-scope questions”
4
Review and retrain NLP model
feedback_loop
“You'll need to review the incoming questions, update the NLP model with new expressions, set it live and repeat until you reach 90% recognition”
5
Safety net routing for unresolved queries
routing
“the bot can direct the user to an FAQ page, provide a link to a video that walks the user through resetting the router, give the user the customer support phone number, or offer to connect the user to a live agent”
Reported outcome

(not stated)

Reported metrics
Chatbots built across industriesmore than 50
NLP recognition rate at launch70%
target NLP recognition rate90%
maximum NLP recognition rate95%
Reported stack
NLPCampfire AI
Source
https://sinch.com/blog/lessons-learned-building-over-50-chatbots-5-continents/
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

(not stated)

What tools did this team use?

NLP, Campfire AI.

What results were reported?

Chatbots built across industries: more than 50; NLP recognition rate at launch: 70%; target NLP recognition rate: 90%; maximum NLP recognition rate: 95% (source-reported, not independently verified).

What failed first in this deployment?

Common failure modes include bots asking for information users already provided, rephrasing loops that can't resolve out-of-scope questions, and dead-end 'Sorry, I don't know' responses that add frustration rather tha…

How is this customer support AI workflow structured?

Define scope and train NLP → Add 99-intents for out-of-scope → Launch at ~70% recognition → Review and retrain NLP model → Safety net routing for unresolved queries.

Related customer support cases