Customer support · Production

YNAB replaced a basic chatbot with Forethought and saw deflection jump from 25% to 70%

The problem

YNAB's legacy chatbot relied on rigid menus and decision trees, making it hard for users to get correct answers in their own words. Deflection was stuck at 25%, meaning three out of four users required human intervention, and the system could not scale to support YNAB's growth ambitions.

First attempt

The legacy chatbot could not understand natural language context — it could not distinguish whether a user meant their bank account or their YNAB account — and if users chose the wrong menu category there was no easy way to backtrack.

Workflow diagram · grounded in source
1
Customer submits chat or email
trigger
“Since implementing Forethought Solve across chat and email”
2
Agentic AI understands intent
ai_action
“Solve Agent is powered by agentic AI, which means it can understand a customer's intent, reason through complex business policies, and take action to resolve their inquiry fully”
3
AI asks follow-up if needed
ai_action
“It asked follow-up questions when needed, interpreted intent even when the wording was off, and responded in a tone that felt consistent with YNAB's brand. There were no hallucinations, either, just useful, grounded answers, even in edge…”
4
Route to human agent
routing
“If human assistance is needed, the conversation is routed to the next available agent with complete details”
5
Assist Agent supports humans
integration
“Assist Agent to support human agents inside their helpdesk”
6
Discover Agent surfaces gaps
feedback_loop
“Discover Agent is Forethought's insight engine, which analyzes unresolved tickets, identifies gaps in the knowledge base, and recommends articles or workflows to close them”
Reported outcome

After deploying Forethought Solve in October 2024, YNAB's ticket deflection rate rose from 25% to over 70% — a 45% improvement over the legacy chatbot — and monthly chat conversations tripled to about 12,000 without requiring additional headcount.

Reported metrics
Ticket deflection rate (post)70%
Ticket deflection rate (pre)25%
Improvement over legacy chatbot45%
Monthly chat conversation volume increase3x
Show all 5 reported metrics
ticket deflection rate (post)70%
ticket deflection rate (pre)25%
improvement over legacy chatbot45%
monthly chat conversation volume increase3x
monthly chat conversations (absolute)12,000
Reported stack
ForethoughtSolve AgentAssist AgentDiscover Agent
Source
https://forethought.ai/case-studies/ynab-replaced-a-basic-chatbot-with-forethought-and-saw-deflection-jump-from-25-to-70
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

After deploying Forethought Solve in October 2024, YNAB's ticket deflection rate rose from 25% to over 70% — a 45% improvement over the legacy chatbot — and monthly chat conversations tripled to about 12,000 without r…

What tools did this team use?

Forethought, Solve Agent, Assist Agent, Discover Agent.

What results were reported?

Ticket deflection rate (post): 70%; Ticket deflection rate (pre): 25%; Improvement over legacy chatbot: 45%; Monthly chat conversation volume increase: 3x (source-reported, not independently verified).

What failed first in this deployment?

The legacy chatbot could not understand natural language context — it could not distinguish whether a user meant their bank account or their YNAB account — and if users chose the wrong menu category there was no easy…

How is this customer support AI workflow structured?

Customer submits chat or email → Agentic AI understands intent → AI asks follow-up if needed → Route to human agent → Assist Agent supports humans → Discover Agent surfaces gaps.

Related customer support cases