Conversations Insights
From a high-risk manual service to a successful product: A data dashboard design that helped secure a key contract renewal.
Role
Solo Product Designer
Team
1 Product owner, 4 Engineers, Designer (me)
Client
One of the leading Airline Carriers in North America
Impact
Drove adoption across 3 contact centre teams and reinforcing client trust that proved key to contract renewal.
Context: A High-Stakes Project
I want to walk you through a project I led at RozieAI, a B2B startup in the Customer Experience (CX) space that builds software for clients in the airline and insurance sectors.
The product is Conversation Insights, an AI-powered analytics tool that helps contact center teams at one of North Americas leading airline carriers, understand their customer conversations in depth.
Problem: A Manual Process That Couldn't Scale
Before we built this product, our process for delivering conversation insights to the client was completely manual and wasn't just scalable. It caused:
01
Internal Inefficiency
Our Product Owners had to spend hours working with data scientists to package up findings into documents, like excel sheets. This prevented our product owners from focusing on important tasks within the organization.
02
A critical client bottleneck
Our client couldn't get the insights they needed on demand. This was a major problem, as they rely on this data for weekly stakeholder presentations and key operational decisions.
Opportunity: A Contract Renewal and a Product Opportunity
With a critical client contract renewal on the horizon, we saw a clear opportunity to move from a manual service to a self-serve product.
By creating a tool that put the power of our AI directly into our clients' hands, we could solve both problems at once. We could eliminate our internal inefficiency while delivering significant new value to our client, strengthening our partnership right when it mattered most.
Most importantly, we also saw an opportunity in expanding this offering to other clients to drive revenue.
My Responsibilities
This was one of the projects where I was responsible for leading the entire dashboard design. This meant I was:
Partnering with our product owner on user research and feedback analysis.
Owning the end-to-end design of the dashboard.
Conducting usability testing with client teams for prominent features like filters.
Iterating at a rapid rate with constant feedback from POs, Design Lead, and Users.
Collaborating closely with engineering to support development and ensure a high-quality launch.
Design Challenge: Balancing Two User Needs with Business Velocity
To understand the problem better, I conducted a two-part discovery.
First, I partnered with our product owners to understand the types of data and insights they were manually sharing.
Second, I conducted interviews with client teams to understand their core needs and to spot any gaps that existed.
This immediately revealed a core conflict between our key user personas, compounded by a critical business constraint.
Senior Leadership
These are the higher-level stakeholders who are responsible for overall contact centre performance. They needed an immediate pulse check - a quick, at-a-glance summary.
Quality Optimization Specialists
These are the power users whose job is to identify root causes, spot trends, and find specific coaching opportunities. They needed a deep dive - the power to investigate granular, call-by-call data.
RozieAI Business
They needed to ship at high velocity to drive adoption ahead of a key contract renewal.
For Senior Leadership
Only the insights for a quick scan
For QO Specialists & PM
Detailed analysis of insights + AWS connect attributes
For RozieAI Business
Ship Fast within 2 weeks!
Solution?
So, how might we design a product that feels intuitive enough for a 10-second scan, yet robust enough for a 10-minute investigation, all while working within an accelerated timeline?
Solution Highlights
In order to address the conflicting needs I designed a single-page dashboard with two views:
Overview: Top Part
Overview: Bottom Part
Table View
Overview is the default view a user sees upon signing into the dashboard. This is designed to be highly visual and scannable for digesting insights as quickly as possible.
Table view enables deep-dive analysis of individual call records and supports detailed exploration across over 60 data columns.
The tabbed structure provides users much flexibility, supporting both quick monitoring and detailed investigation without requiring complex navigation.
Filter System
One of the key challenges was designing an effective filtering system that could handle over 60 columns with ease of use, while trying to maintain the velocity at which the business wanted to ship the product.
In order to overcome this challenge, I conducted a rapid competitive benchmarking for filtering patterns in data-intensive applications. I did this to identify best practices that matched with our use case and found two potential patterns:
Column Filters (in Jira):
Always visible, offering quick access but adding significant visual clutter to the interface.Query Builder (in Airtable):
Hidden by default, maintaining a clean UI but requiring an explicit user action to reveal.
Given these two patterns, I chose Airtable style query builder as a reference point because of the following factors:
User Experience:
It provided a far superior experience by keeping the interface clean and uncluttered, revealing complexity only when the user needed it. Additionally, a query builder allows for any complexity of filter criteria to be applied, making it more powerful to accommodate any complex use case for our users.Technical Feasibility and Business Alignment:
After collaborating with my engineering lead, we confirmed it was a single, reusable component that was significantly faster to customize than 60 individual column filters. This allowed us to hit our velocity goals and ship on time.
Manage Columns
Manage Column
Following a successful launch, we hit an immediate, high-volume stress test: the client's contact centre were operating in a peak call season. This influx of data revealed a critical latency issue.
A slow interface is an unusable interface, so I took the lead on finding a design-led solution. After collaborating with engineering to identify the bottlenecks, it was clear we needed to make some smart, pragmatic tradeoffs. My strategy was to introduce intelligent constraints that would improve speed without sacrificing the tool's core power, which led to our two-pronged solution:
Constraining the Default Date Range:
Action
I initiated rapid user validation sessions to understand the most common timeframes for data analysis.
Insight
The research clearly indicated that a 90-day look-back window was sufficient for the vast majority of everyday use cases.
Outcome
By implementing a 90-day default limit, we drastically reduced initial query times and improved overall system responsiveness, directly addressing the latency issue while preserving the most critical user workflow.Introducing a "Manage Columns" Feature:
Action
To tackle the complexity of the 60+ column table, I again validated with users that a curated, smaller set of columns would serve most primary tasks.
Insight
A default view of the 15 most critical columns was identified as the optimal balance between information density and clarity.
Outcome
I designed the "Manage Columns" feature with this sensible default. This dramatically reduced the data payload, particularly for saved views, while still offering the full customization required by power users.
Getting Into the UI Details
KPI Cards
KPI Cards Anatomy
All the components in the overview section were driven by one core principle: clarity at a glance because our users need to absorb key information as quickly as possible.
For KPI cards, the value is intentionally the largest and boldest element to make it the immediate focal point. The KPI Label is secondary, providing context without competing for attention. For complex values, the numbers remain the focus while the unit specifiers ("min", "sec") are de-emphasized to provide context without clutter.
Call Volume Chart
This was one of the biggest asks from our client teams because understanding call volume patterns would help them identify call spikes for a time period.
I chose a simple line graph as it's the clearest way to show data over time. While the default state provides a clean overview of the selected time-period, hovering reveals a tooltip with precise data. As a deliberate choice to reduce visual clutter, the x-axis only displays the start and end dates, keeping the focus on the overall shape of the data.
This design offers both a high-level view and granular detail on demand.
Primary Topics, Customer Intents, Root Causes
I grouped Primary Topics, Customer Intents, and Root Causes into a single tabbed component because they are all key indicators that help users understand the nature of the calls. Since they share a similar data structure (a text label and a call count), this approach creates a consistent and organized experience.
Within each list item, both the label and the count are equally important. To give them equal visual priority, I designed a two-column layout, separating them to create distinct scannable paths. This allows a user to either easily read the list of topics or quickly scan and compare the call volumes.
Call Resolution
The 'Call Resolution' component is designed to give users a quick, visual summary of call outcomes. While it follows the same scannable, two-column layout as other lists to maintain consistency, I introduced a simple horizontal bar chart for each category. This addition is intentional.
The length of the bar provides an immediate visual cue, allowing users to compare the scale of each outcome much faster than by reading the numbers alone. It instantly highlights the most frequent category and helps users spot patterns at a glance.
Journey Moments
This component provides the crucial context of when in the customer journey calls are happening. To make this information easy to digest, I designed the layout as a two-column grid. I intentionally placed related but opposing stages on the same row, such as 'Pre Flight' and 'Post Flight'. This creates logical pairings that allow users to quickly contrast call volumes and spot imbalances between different points in the journey.
To further enhance scannability and speed up recognition, I added a unique icon for each journey moment. These icons are simple visual signifiers that reinforce the meaning of each label.
Sentiment Analysis
Now, this was a tricky component to design.
The goal of this component is to show the impact the CC agents have on customer sentiment from the start to the end of a call. I designed the default state as a clean and high-level summary, using the principle of progressive disclosure to invite users to click and explore the data flow. This approach prevents overwhelming users with complex information upfront.
Once a user clicks, the component reveals the full story. Visual flow lines, contextual 'Part of Total' numbers, and color-coded borders all work together to show exactly how an initial sentiment was impacted by the interaction. This transforms a simple summary into a powerful and scannable analysis of effect.
Impact
Adoption:
Adopted by 4 internal teams within the client organization; demonstrated to 3 external clients, all of whom expressed interest in rollout.
Efficiency gain:
100% elimination of manual insight delivery which enabled contact center teams to access insights faster.
Qualitative validation:
Broader adoption across departments and strong stakeholder feedback reinforced the product’s value as an indispensable daily tool for clients.
Business outcome tie-in:
Success led to requests for new feature support requests (under NDA) and secured a renewed client contract which validated the product’s long-term business impact and and an estimated 4-5% growth in revenue.
Thanks for Reading!
Thanks for following along! While I've walked you through the key strategic decisions, there's a lot more to the story.
I'd love to chat more about the detailed design iterations and the process of balancing business goals and engineering constraints. If you're interested in hearing more, I'm always up for a conversation!