AI-powered benefits: driving employee engagement
company:
Ben
roles:
LEAD, interaction & Content design, USer Research
YEAR:
January 2025
Overview
I led the UX product vision that introduced AI-powered benefit explanations, personalised recommendations, and an improved browsing experience. This increased NPS and positioned us as the world's first intelligent benefits platform, helping close major enterprise deals and secure Series B funding.
Results:
Results:
- Employee NPS satisfaction score increased from 12 to 35 in 3 months (191% increase).
- 17% drop in benefit-related inquiries via support.
- The AI-powered decision-support features I designed were first prototyped in response to specific RFP requirements from British Airways and Santander. Both deals were closed ($1M ARR).
Problem
User research with employees and an analysis of Intercom tickets revealed three key barriers:
- Poor understanding of benefits: Employees struggled to understand complex benefits like Private Medical Insurance and Pension, with crucial details about their features buried in lengthy PDFs. This led to confusion about coverage, claims and overall value of their benefits.
- Irrelevant content: Employees were first exposed to benefits they weren’t interested in, leading to decreased engagement with the platform.
- Difficulty finding enrolled benefits: Employees struggled to locate specific benefits and manage their enrollments efficiently, often abandoning tasks.
All this left companies investing in comprehensive benefits but seeing low engagement.
Smart search & AI powered answers
Problem
The platform lacked search functionality, preventing users from quickly finding what they needed. Additionally, support received high volumes of benefit-related tickets, revealing users needed clear, concise explanations of complex policies.
MVP: A single entry point with guardrails
I designed a single entry point where users can either type a question or what they're looking for. The system automatically figures out whether to search or provide an answer.
Since this was an MVP, users could only select from a curated bank of questions. While this constraint reduced flexibility, it was a deliberate tradeoff to balance usefulness with legal and technical constraints.
Since this was an MVP, users could only select from a curated bank of questions. While this constraint reduced flexibility, it was a deliberate tradeoff to balance usefulness with legal and technical constraints.
MVP: Source citations & related questions
To strengthen user trust in the AI’s answers and make verification easier, I decided to include source citations. When there are more than four sources, these are revealed through progressive disclosure.
Related follow-up questions were also surfaced to allow the user to go deeper, if they wish to do so.
Related follow-up questions were also surfaced to allow the user to go deeper, if they wish to do so.
Using AI to improve benefit understanding
Problem
Employees were missing the full value of their benefits because key features were hidden in lengthy PDFs they rarely opened. Case in point: 94% of Ben's employees with Aviva life insurance had no idea it included free annual blood tests and nutrition support.
Making benefit features instantly scannable
When looking at a specific benefit, employees are able to see a section with all of the features included in a benefit.
The section lets employees see at a glance what's available, reducing the need to sift through lengthy, hard-to-digest policy documents.
The section lets employees see at a glance what's available, reducing the need to sift through lengthy, hard-to-digest policy documents.
I added progressive disclosure to limit the amount of features shown in any single view (up to 3), after which a show more button appears that opens a modal.
Enabling HR admins to extract benefit features with AI
When configuring a benefit, managers can use AI to extract features from policy documents. Titles, icons, descriptions, and categories are automatically generated. Results can then be reviewed and manually edited.
I worked with the engineering team to fine tune the AI prompts to ensure the right tone of voice was used and that the length of each field was respected.
I worked with the engineering team to fine tune the AI prompts to ensure the right tone of voice was used and that the length of each field was respected.
The employer control centre allows managers to use AI to rapidly extract benefit features from policy documents.
Increasing time to value through personalisation
Problem
While widget-based dashboards excel in productivity tools like Notion (where users must manage deep hierarchies) they created unnecessary friction for our benefits platform. Because the information architecture was relatively flat, a home page full of 'shortcuts' felt like a solution for a problem we didn't have. I recognised that our goal wasn't just navigation, but meaningful engagement, which required a more intentional, content-forward approach. Below is the old dashboard:
Before: The whitelabelled dashboard displayed shortcuts to Reminders, Available Benefits and the Wallet.
A new personalised home page experience that increases time to value
By asking users to share with us their goals (optional) during onboarding and tracking their behaviour on the product, I was able to redesign a home page experience that surfaces relevant benefits, which increased time to value. The redesign process was accompanied by a UI glow-up to the existing design.
After: The home page now surfaces relevant benefits, discounts, and company-subsidised options, helping employees make more informed decisions.
Process
Through Intercom tickets and conversations with employees, we discovered that users weren't getting full value from their benefits; relevant content wasn't surfaced in a timely way and some simply didn't understand how they worked. The Product Lead and I saw an opportunity to use AI to increase time to value through personalisation, summarising complex benefit policies and extracting features from benefits.
My role included:
My role included:
- Evaluate the AI's output quality by defining what "good" looks like across dimensions such as accuracy, relevance, tone, completeness, and visual formating. This was then used to fine-tune the system prompt.
- Work with the team to structure outputs (lists vs paragraphs, length, level of detail) based on user needs rather than just AI defaults.
- Align founders on the UX vision and key differentiators.
Credits
PRODUCT MANAGEMENT
maria paula loayza
kevin armstrong
kevin armstrong
product design
KEVIN ARMSTRONG
MALU ROMERO
SANTIAGO SOTOCA
MALU ROMERO
SANTIAGO SOTOCA
tech lead
christian bedón
operations
JACQUELINE SOLARI
ESTEFANíA RODRIGUEZ
ESTEFANíA RODRIGUEZ
sponsors
jorge o'hara
hernán carranza
hernán carranza
business lead
alessandra ponce