All work

Case studyCX Operations2023-2025

Raised customer satisfaction from 75% to 95% with an AI-assisted contact center

Manual QA reviews a sample. The problems live in the calls nobody hears.

  • 75 to 95%customer satisfaction
  • 100%of calls analyzed instead of a sample
  • 100%of calls analyzed, instead of a sample
Context
In most contact centers QA samples a fraction of calls, so coaching lags weeks behind the conversations it is about.
What I built
An AI-assisted contact-center stack on Five9 AI Assist: automated sentiment analysis on calls, real-time coaching workflows for supervisors, and AI-assisted routing.
Headline result
Customer satisfaction rose from 75% to 95%.

Specification

Role
Design, build, rollout
Timeline
2023 to 2025
Stack
Five9 AI Assist, NLP, performance analytics
Status
Deployed across the team

The problem

Sampled QA means the problems live in the calls nobody hears. Across a team spread over time zones, coaching ran on anecdotes and arrived weeks after the conversations it addressed, and routing misfires burned handle time before an agent ever spoke. Satisfaction sat at 75%.

The team did not need more monitoring. It needed feedback that was complete instead of sampled, and fast enough to act on during the same shift. Any system also had to hold up across a distributed team, where coaching consistency is hardest to keep.

The approach

The sentiment layer analyzes calls automatically instead of sampling them, replacing manual QA reporting with complete, systematic feedback. I chose full coverage over a bigger sample because sampling was the root problem.

Coaching workflows deliver that signal to supervisors in real time, surfacing which calls need intervention while intervention still matters. A dashboard reviewed at month-end would have reproduced the old lag with better graphics.

AI-assisted routing gets customers to the right agent before anyone speaks. Humans stay on the conversation; the AI works on routing, sentiment, and feedback.

The rollout ran through the agents themselves. I trained the team of more than 30 people on working with AI-driven feedback, because a coaching system agents distrust becomes surveillance, and surveillance moves no metric.

  1. 01

    Every call

    Routing places the customer before anyone speaks, and the call is captured in full.

  2. 02

    Sentiment, in full

    Automated analysis covers 100% of conversations instead of a QA sample.

  3. 03

    Same-shift coaching

    Supervisors see which calls need intervention while the shift is still running.

  4. 04

    Evidence, not anecdote

    Performance conversations start from data the agent has already seen.

Fig. 01 The feedback loop. Every call is analyzed rather than sampled, and the signal reaches a supervisor while intervention still changes the outcome.

A coaching system agents distrust becomes surveillance, and surveillance moves no metric.

The outcome

Customer satisfaction rose from 75% to 95%. Coaching became evidence-based instead of anecdotal, QA became systematic instead of sampled, and the platform stayed in daily use across the full team. The durable change was the feedback loop: performance conversations started from complete data the agent had already seen, which changed how coaching landed.

What I would do differently

Freeze a cleaner metrics baseline before rollout, so gains could be attributed between routing, coaching, and sentiment work. Formalize the coaching playbooks earlier instead of letting each supervisor derive their own from the new data.

Some specifics are abstracted for confidentiality. I am glad to go deeper in conversation.

Open line

Thinking about adoption? I have opinions on it.