How an AI writing assistant transformed live chat quality overnight

The Situation

A cryptocurrency exchange platform running a high-concurrency live chat support team had a problem that was quietly undermining every performance metric, the quality of agent communication was inconsistent, slow and dependent on constant supervisory intervention before a single message reached the customer.

Every response was being reworked. Drafting a single message was taking up to 10 minutes. Escalations were climbing. And the supervisory team was so consumed by reviewing outgoing messages that coaching, compliance and product training had taken a back seat.

The operation needed a fix that worked in real time, inside the conversation, not around it.

The Problem
Four issues were compounding each other daily:

Communication quality
Recurring grammatical errors and poor sentence formation were pulling QA scores down and creating a poor experience for customers.

Missing empathy and tone
Agents frequently missed required soft skill elements appropriate empathy, apologies, and sentiment matching, during live interactions.

Supervisory bottleneck
Every outgoing message required manual review before sending, directly inflating Average Handling Time and suppressing First Time Resolution rates.

Tool dependency
Agents relied on multiple external tools for drafting support, with constant copy-pasting that disrupted formatting and added unnecessary time to every response.

The Solution - Penpal

Penpal was deployed directly into the live chat workflow, embedding four real-time capabilities into every agent interaction:

Drafting
Context-aware response suggestions generated in real time, aligned to the customer query and chat history.

Rephrasing
Instant rewrite of agent-drafted text for clarity, professionalism and sentiment alignment.

Tone Correction
Real-time calibration ensuring every response carried the right level of empathy and brand voice.

Grammar
Automatic correction of errors before the message reached the customer.

The maker-checker process was eliminated entirely. Agents moved from juggling multiple external tools to a single, native interface that handled communication quality automatically, leaving them free to focus on the customer, the query and the resolution.

What Changed

QA Score
~ 6% of improvement in overall quality within the first deployment cycle

Reduction in drafting time 5 to 7 min → 1 to 2 min
Ultimately improving the AHT score overall by 80% ~

Escalation Rate
Customer escalations dropped by more than 65%.

Agent Confidence
Every agent reported measurably higher confidence in their responses.

Attrition
Full team stability maintained throughout and after go-live.

Training focus shift
With communication quality handled by Penpal, the QA and training teams redirected their attention entirely to product knowledge and compliance.

The maker-checker process was eliminated entirely. Agents moved from juggling multiple external tools to a single, native interface that handled communication quality automatically, leaving them free to focus on the customer, the query and the resolution.