Customer support · Production

iFIT uses Forethought AI to deflect 33% of chats and save 436 agent hours

The problem

iFIT's customer support organization relied on a chatbot that could not deflect chats, knowledge spread across hundreds of documents in multiple systems agents could not easily access, and a manual topic-discovery process requiring a person to search through more than 10,000 lines of data before any knowledge article could be written.

First attempt

The existing chatbot was not successful at deflecting chats and required every workflow or script change to go through a technical administrator, making it difficult to maintain and ultimately unscalable.

Workflow diagram · grounded in source
1
Member contacts help center
trigger
“iFIT uses Solve as its chat widget”
2
NLU-powered chat resolution
ai_action
“Solve is powered by Natural Language Understanding, not keywords. It comprehends sentence structure, meaning, synonyms, tone, and nuance to extract the perfect answer from millions of documents.”
3
Workflow Builder recommends automations
feedback_loop
“It also recommends which workflows to build based on common questions and issues, helping the support engine stay up to date with relevant information”
4
Triage classifies incoming tickets
ai_action
“Triage uses historical data to proactively predict and classify new and incoming support tickets”
5
Route ticket to right agent
routing
“The iFIT support team then leverages routing capabilities to prioritize and handle tickets without the need to manually triage”
6
Assist surfaces info in Salesforce
integration
“relevant knowledge articles, past cases, macros, and notes are surfaced directly within Salesforce, iFIT's helpdesk. When cases can't be deflected and need the intervention of an agent, Assist provides agents with all of the right resour…”
7
Discover uncovers insight gaps
feedback_loop
“Discover uses generative AI to uncover insights from past support interactions and recommend workflows and content. Discover proactively recommends workflows that can be automated to maximize cost savings. The recommendations are provide…”
Reported outcome

iFIT achieved 33% chat deflection via the Solve widget, 82% prediction accuracy from Triage, 3,689 deflections and 436 agent hours saved through Discover, plus over 20,000 instant chat resolutions and over 39,000 emails deflected.

Reported metrics
Chat deflection rate33%
Triage prediction accuracy82%
deflections via Discover3,689
Agent hours saved436 agent hours
Show all 7 reported metrics
chat deflection rate33%
Triage prediction accuracy82%
deflections via Discover3,689
agent hours saved436 agent hours
instant chat resolutions+20,000
emails deflected+39,000
manual data search volume (pre-Forethought)more than 10,000 lines of data
Reported stack
ForethoughtSolveTriageAssistDiscoverWorkflow BuilderNatural Language UnderstandingSalesforce
Source
https://forethought.ai/case-studies/ifit
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

iFIT achieved 33% chat deflection via the Solve widget, 82% prediction accuracy from Triage, 3,689 deflections and 436 agent hours saved through Discover, plus over 20,000 instant chat resolutions and over 39,000 emai…

What tools did this team use?

Forethought, Solve, Triage, Assist, Discover, Workflow Builder, Natural Language Understanding, Salesforce.

What results were reported?

Chat deflection rate: 33%; Triage prediction accuracy: 82%; deflections via Discover: 3,689; Agent hours saved: 436 agent hours (source-reported, not independently verified).

What failed first in this deployment?

The existing chatbot was not successful at deflecting chats and required every workflow or script change to go through a technical administrator, making it difficult to maintain and ultimately unscalable.

How is this customer support AI workflow structured?

Member contacts help center → NLU-powered chat resolution → Workflow Builder recommends automations → Triage classifies incoming tickets → Route ticket to right agent → Assist surfaces info in Salesforce → Discover uncovers insight gaps.