# Generative AI customer support and smarter webchat

> Canonical: https://deliverbetter.co.uk/our-work/ai-customer-support/
> Published: 24 Sep 2026. Last updated: 1 Oct 2026.
> Author: David Bain, Deliver better.

## In short

The old webchat matched keywords and hit dead ends. A generative assistant, fed from the knowledge we actually had, now answers most webchat questions without a person.

CASE STUDY 

 
 
111 Generative AI customer support and smarter webchat.
 

 
 How moving from a keyword webchat to a generative assistant, fed from one consolidated knowledge base, let a UK B2B services business answer most webchat questions without a person.

 

 

 

 
 
 
 

 

 

 

 
 
 
 

 

 

 

 
 
 
 
222 The situation
 

 
 Customers waited too long for simple answers, and the webchat wasn’t much help. It matched keywords: if your question happened to line up with one, you got an answer; if it didn’t, you went round in loops or gave up and called. The answers were only ever as good as the keywords behind them, and there were a lot of gaps.

 

 

 

 
 
 
 
222 What was getting in the way
 

 
 Customers waited too long for simple answers, and the old webchat matched keywords and hit dead ends — only ever as good as the keywords behind it, with plenty of gaps.

 

 

 

 
 
 
 
222 What changed, and why
 

 
 The bigger move wasn’t the model, it was the content. We consolidated the scattered knowledge into one reliable store for the assistant to draw on, moving from keyword matching to a generative assistant that handles the simpler, self-serve questions directly and hands over to an agent when a question genuinely needs one.

 

 

 

 
 
 
 
222 The results
 

 
 The assistant now answers the large majority of webchat conversations without an agent, up from none when it was keyword-based, and the answers are more accurate because they come from real, consolidated content. Agents spend their time on the questions that actually need a person.

 

 

 

 
 
 
 

 

 

 

 
 
 
 Related

 

 
 
222 Related services and insights.
 

 
 [

Service
333 Deliver better AI
See the service →

](/what-we-do/ai)[

Service
333 Deliver better Digital
See the service →

](/what-we-do/digital)[

Insight
333 Your human in the loop may have stopped looking
Read the insight →

](/insights/human-in-the-loop)[

Insight
333 Before you buy the AI tool, ask who it gets wrong
Read the insight →

](/insights/questions-before-buying-ai)

 

 

 

 
 
 
 
222 Thinking about AI for customer service?
 

 
 Tell us what customers ask most. We’ll work out where AI would help and where a person still should.

 

 
 
 [
 
 Start a conversation
 
 ](/contact)
 

 
 [
 
 Try the free tools
 
 ](/free-tools)

## The honest lesson

A generative assistant is only as good as the content behind it. Pull scattered knowledge into one reliable source and the model does the rest; get that wrong and a better model just gives you wrong answers faster.
