My role
I designed end-to-end interaction and visual design of the product, from initial concept to MVP.
Collaborators:
👩💼 Product Managers
Defining success metrics, and scope. Ensuring what we’re building for is right.
🧑💻 Machine learning and engineering team
Had an early prototype, and I refined and defined the full interaction model, visual design, and overall experience.
👑 Leadership
Gained feedback through showcasing intractable prototypes created using Cursor. Shifted our feedback process by showcasing working prototypes built in Cursor instead of static screens, a game-changing in how clearly stakeholders could engage with the design.
🖼️ Marketing
Collaboration to define the visual language of the product, ensuring that it’s consistent with our branding message
Business goal
Talent teams previously reviewed candidate applications each day manually, one by one, a slow and time-consuming process. Our goal was to help them surface top candidates faster and with more confidence using AI.
Surfacing answers to questions like “Why is this candidate the best fit?”, improving their efficiency and understanding of their candidate pool.
Interaction
Since most recruiters manage candidates in their ATS (Applicant Tracking System), we embedded Tia into our Chrome extension so they can ask questions while browsing candidates.
Research into how AI is being integrated across B2B products have surfaced a common pattern embedding AI natively within the platform, but that approach had limitation for us as a portion of our users review candidates outside of our platform - in external ATSes (Applicant Tracking System).
We already had a Chrome extension in place for interview scheduling. It made sense to integrate Tia directly into it, giving users access to AI assistance while browsing on their ATS.
Brainstorming
Expanding touch points users might need Tia for beyond identifying candidates: Discover, comparing and generate
Mapping out the full hiring journey, we have uncovered opportunities for Tia well beyond candidate search: from drafting offer letters to developing training plan based on their weaknesses.
Working with engineers, we prioritised tasks for MVP, set others aside for later releases based on feedback, and grouped them into 3 query categories:
Discover: Top candidates on the role
Comparing: Between multiple candidates
Generate: Decline and acceptance emails, interview questions or training plans based on candidate data
To prioritise a fast release, the team focused on adoption feedback and deprioritised other ideas in favour of lowest possible effort.
Challenge
Using analytics to surface responses that were simple and digestable
Tia's recommendations are grounded in candidate insights reports, so when a talent team asks "who's the best fit for this role?", Tia pulls directly from that data. The challenge was surfacing the key data at the right moment, helping teams quickly decide whose report was worth a closer look.
With the help of AI tools like Gemini and Claude, my first explorations looked a bit like this: Their scores, summary, strengths and risks, and key competencies that they scored highest in.
As someone who understands the science behind these scores, everything felt important. Each data felt like a potential deciding factor in a hire. But something felt off. It was overwhelming.
Designing for others means constantly reminding yourself that you are not the user. I then stepped back and looked at the bigger picture, turning to product analytics to understand which parts of the report talent teams actually engaged with, and letting that data guide what truly mattered.
Scores were a key deciding factor, and candidate summary was the second section talent teams engaged with most frequently. Users are prioritising speed when browsing, a one paragraph text summarises all key strengths and weaknesses.
From this data, I have ended up with this:
Branding
The visual direction was to feel smart and professional while staying true to our brand colours pink and purple
Our marketing designer was responsible for the Tia logo, and I designed the visual UI for the Tia interface.
Drawing on my design expertise and consolidated team feedback, I ultimately landed on a lighter palette using these colours, as it felt cleaner and allowed more focus to the content.
Stakeholder alignment
With the help of Cursor (AI tool), I have prototyped the full interaction of Tia and showcased it to relevant stakeholders and leadership for validation
This was an intentional de-risking step before any further product changes were made.
Final design
Simple and welcoming for first time users.
Users already know what this product is. The onboarding message serves as an additional step to guide users our most outstanding capabilities, demonstrating value at first visit.
Context is everything
When recruiters are viewing a job in their ATS, we incorporate that context into the chat and evaluate candidates based on the role they’re looking at, creating a more relevant and contextual experience.
Sufficient guidance to deliver value
A blank chat input can be confusing, so a prompt library guides users through the hiring process. I also proposed dynamic prompts tailored to the specific job being viewed - scoped out of MVP to revisit after user feedback.
Delightful waiting experiences
We surface Tia's reasoning in real time, keeping users informed and bringing transparency to how top candidates are evaluated.
Crafting responses in a digestible manner
Core AI tasks were mapped to cards, each with clear follow up actions to keep the workflow moving.
Measuring success
Explicit and implicit feedback were both monitored by the team
For our MVP phase, adoption is the key metric we want to ramp up.
Explicit feedback are collected through “thumbs up” and “thumbs down” in our responses
Implicit feedback were metrics such as:
- Average duration and adoption of chat
- Whether users were asking the same questions again
- Interaction between the tasks cards designed
This project is currently in the pre launch phase, so its success metrics are not yet available.
Learnings and reflection
Feedback becomes increasingly important in designing for an AI product, and products who wins are those that are context aware and has deep knowledge about users.
Understanding what customers might ask are hard to test and assume as there is no linear flow in a non-deterministic chat. Feedback that emerges from real conversations becomes essential as it can reveal how they would naturally talk to AI.