Turning Customer Conversations into AI Summaries & Sentiment Insights (CX Product)

AI-powered summaries and sentiment analysis for faster customer insight discovery

Company

Lang.ai

Role

Product Designer

Industry

AI · SaaS · CX

Duration

3 Months

Overview

CX teams needed a better way to understand what was happening across thousands of customer conversations. The initial opportunity looked like a dashboard redesign - but research revealed a deeper problem: users often didn’t know what they were missing.

Instead of relying on Lang.ai's existing dashboards, many customers were using external analytics tools to explore their data. I led the discovery and redesign of the insights experience, working closely with a Data Scientist to turn AI-generated sentiment and conversation patterns into information CX teams could actually trust and act on.

Impact:

↑ Faster understanding of customer trends
↑ More actionable insights from conversation data
✓ Sentiment detection validated and rolled out to real customers

The problem

The project initially started as an opportunity to improve Lang.ai's existing dashboards.

But research challenged that assumption.

Through continuous conversations with CX teams, I discovered that many users weren't relying on our dashboards at all. They were exporting or visualising their data through external analytics tools because the information available in Lang.ai wasn't helping them answer the questions they actually had.

The core problem became:

“I don't know what I don't know.”

Users didn't necessarily need more charts. They needed the product to surface meaningful patterns and explain what was happening in their customer conversations.

At the same time, sentiment detection introduced a technical challenge: early versions frequently classified conversations as negative, creating false positives and undermining trust in the AI.

My role

I owned the product design across discovery, definition and delivery:

Led continuous user research with CX teams
Challenged the initial dashboard redesign direction based on research
Defined the insights experience around real customer needs
Collaborated closely with a Data Scientist to improve sentiment detection
Translated complex AI outputs into understandable product experiences
Iterated with engineering and validated the experience with customers

The work was released progressively, first to smaller customer groups and then more broadly as the experience matured.

The solution

From dashboards to insights

Rather than simply redesigning the existing dashboards, I shifted the experience towards surfacing what users needed to know without requiring them to find it themselves.

01. AI-powered conversation insights

The experience surfaced meaningful patterns across customer conversations, helping CX teams understand emerging topics and trends without manually exploring large datasets.

→ Less time searching for patterns, more time acting on them.

02. CSAT-based sentiment

Research revealed that CX teams naturally evaluated customer experience through CSAT, so we aligned the sentiment experience with the metrics they already trusted.

→ AI outputs became more relevant to how teams actually measured customer satisfaction.

03. Reliable sentiment detection

Early sentiment models produced too many negative classifications.

I worked closely with the Data Scientist to investigate false positives, iterate on the detection logic and validate the resulting experience.

→ The final sentiment detection performed reliably enough to be shipped to real customers.

How I approached it

The key to the project was continuous discovery rather than treating research as a one-off phase.

I used:

User interviews and ongoing customer feedback
Product usage and behavioural data
Session recordings and analytics
Competitive and external-tool analysis
Rapid Figma iterations
Usability testing
Close collaboration with Data Science and Engineering

One of the most important insights came from challenging the original problem rather than immediately redesigning the dashboards.

The research showed that the problem wasn't simply “our dashboards aren't good enough.”

It was:

“We're asking users to look for insights that the product should be helping them discover.”

Outcomes

Features were rolled out progressively to enterprise customers, allowing for validation in real conditions.

Results:

Reduced manual workload for CX agents
Improved clarity and confidence in decision-making
Introduced a new layer of customer understanding through sentiment insights

Key takeaway

Good AI experiences don't start with the model. They start with understanding what users actually need to know.

This project taught me to treat AI quality, user trust and product usefulness as one design problem - working closely with Data Science to make complex intelligence understandable, relevant and reliable.