Customer support · Production

Grammarly achieves 87% deflection and 4.2 CSAT with Forethought

The problem

Grammarly's previous chatbot could not understand conversational context, broke when users rephrased or asked follow-up questions, and required the support team to manually author every answer inside a decision tree that was hard to manage and update.

First attempt

Grammarly's prior chatbot relied on menu-based interactions and canned responses, lacked multi-turn context awareness, and forced the team into constant manual maintenance of a sprawling decision tree.

Workflow diagram · grounded in source
1
Customer contacts via email or chat
trigger
“Grammarly's support team uses email and chat to connect with customers who need help”
2
Agentic AI understands customer intent
ai_action
“It uses agentic AI to understand customer intent, enabling full issue resolution without relying on rigid decision trees”
3
Knowledge base and system integration
integration
“By integrating directly with Grammarly's knowledge base and internal systems, Solve delivers human-like interactions”
4
Ticket deflected or resolved
output
“deflection rates have remained consistently strong at 87%”
5
QA accuracy confirmation
validation
“QA confirms the accuracy of Forethought's resolutions”
6
Unresolved query review loop
feedback_loop
“consistent reviews of unresolved queries helped them fix gaps before they became real problems”
Reported outcome

After deploying Forethought, Grammarly's CSAT tripled to 4.2 out of 5 and deflection climbed from around 60% to a sustained 87%, never dropping below 80%.
API integrations drove an additional 5–10% gain.

Reported metrics
CSAT score4.2 out of 5
CSAT improvement multipliertripled
Deflection rate87%
Deflection rate starting pointaround 60%
Show all 7 reported metrics
CSAT score4.2 out of 5
CSAT improvement multipliertripled
deflection rate87%
deflection rate starting pointaround 60%
deflection rate floor80%
API integration deflection gain5–10%
implementation timea week and a half
Reported stack
ForethoughtSolve
Source
https://forethought.ai/case-studies/grammarly-achieves-87-deflection-and-4-2-csat-early-with-forethought
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

After deploying Forethought, Grammarly's CSAT tripled to 4.2 out of 5 and deflection climbed from around 60% to a sustained 87%, never dropping below 80%.

What tools did this team use?

Forethought, Solve.

What results were reported?

CSAT score: 4.2 out of 5; CSAT improvement multiplier: tripled; Deflection rate: 87%; Deflection rate starting point: around 60% (source-reported, not independently verified).

What failed first in this deployment?

Grammarly's prior chatbot relied on menu-based interactions and canned responses, lacked multi-turn context awareness, and forced the team into constant manual maintenance of a sprawling decision tree.

How is this customer support AI workflow structured?

Customer contacts via email or chat → Agentic AI understands customer intent → Knowledge base and system integration → Ticket deflected or resolved → QA accuracy confirmation → Unresolved query review loop.