Finding users where they already were: a resume feature that drove 62% adoption and 3× retention

I led the design of a resume feature that helped early-career job seekers surface missing experiences, tighten loose writing, and build a stronger resume through guided AI.

The AI resume feature — hero

Impact

Adoption

62%

Nearly 2× higher than our previous best.

Retention lift

3×

~50% of users stayed by Day 10, ~30% by Day 30 (vs <10% before).

First action

Top action

New users tried it soon after signup.

My role
Product Design, User Interviews, Research Synthesis, Cross-Functional Alignment
Team
CEO, Head of Growth, Head of Customer Success, 2 Engineers
Timeline
Feb 2025 – Mar 2025
Company
nSpire AI uses AI to help college students and early-career professionals secure their dream jobs.

An AI resume workflow that makes job-fit gaps visible and easy to act on.

Core flow

1 · Set the target

Users upload a resume and job description, or just name the role, and get a match score with prioritized gaps. It answers the first question every applicant has: am I even qualified?

2 · Build the evidence

A guided AI chat helps users recall experiences they'd left off, then turns them into resume-ready stories. It closes the “what do I include?” gap.

3 · Generate the resume

The strengthened content comes together into a polished, role-targeted draft, so users never get stuck on “how do I say this well?”

The starting point

The problem: our core value moment depended on an infrequent trigger

At the time, nSpire AI was centered around interview practice, built for early-career job seekers.

PRODUCT VALUE Practice interviews REQUIRED TRIGGER An interview invite REQUIRES USER REALITY Rarely get interviews RARELY MET SO WHAT IT COST US Weak activation & retention

I reframed it as a timing and frequency problem. We needed an earlier, more frequent job-search moment where users could feel value.

The product bet

Resume prep was the earlier, more repeatable entry point

The instinct on the team was to deepen interview prep, our strongest feature. I argued for the opposite: instead of waiting for the interview, bet on the moment users were already in, getting ready to apply.

I. Resume prep More job seekers reach it, sooner,and return often II. Interview Fewer job seekers, and farless often

The core insight

Users needed more than wording; they needed to know what counted

Choosing resume prep told me where to help. The interviews I ran told me what to fix.

What users said What it meant for the product
"I'm not sure if I'm qualified. Show the match
"I don't know what to include. Uncover what they'd forgotten
"I don't know how to say it well. Sharpen the evidence

Product direction

Show users where they stand, surface missing evidence, and help them turn relevant experiences into a stronger resume.

Design principles

The rules I designed against

01

Make the next step obvious

02

Balance guidance and control

03

Make progress visible

04

Don't fragment the resume

Design iteration · Decision 01

Structuring an AI chat that guides without taking over

An open chat leaves users lost. A rigid flow leaves them trapped. I weighed three structures against this tension.

Options explored

Idea 01

AI surfaces one gap at a time. Users address it or skip.

Prototype: AI surfaces one gap at a time in chat

Users had to trust the AI's pick without seeing what else was on the table.

Idea 02

Clickable gap list. Selecting a gap opens its own scoped chat.

Prototype: clickable gap list, each gap opening its own scoped chat

Real conversation rarely stays inside the structure you draw for it.

Direction chosen

Idea 03

Gap panel on the left for visibility. One continuous chat on the right. AI starts on the highest-priority gap; users can redirect freely.

The panel kept the full picture visible. The continuous chat let the conversation follow how users actually think.

The shipped design: gap panel on the left, one continuous chat on the right

Design iteration · Decision 02

Making AI prioritization legible and motivating

Users liked seeing their gaps and matches, but to act on them they needed more than a list.

Prototype of the gaps and matches list users saw after the AI review

Options explored

Familiar patterns each answered one piece, not the whole picture.

Star rating

Star rating pattern prototype

Encoded match strength. But a rating in isolation couldn't show what improving one star would do to overall fit.

Progress bar

Progress bar pattern prototype

Encoded effort spent. But a full bar on a low-priority gap looked the same as one on a high-priority gap.

Ordered list

Ordered list pattern prototype

Encoded rank, not magnitude. Users couldn't tell if #2 was nearly as important as #1, or far less.

What shipped

Points toward a total score carried weight, priority, status, and progress at once.

The total score design that shipped

Design iteration · Decision 03

Scoping what the AI should generate

One bullet at a time, or the whole resume? The scope decided what the AI could see, and what users had to carry.

Options explored

Idea 01

AI generates one bullet at a time.

Prototype: AI generates one bullet at a time inside chat

Idea 02

AI generates the whole resume in one pass.

Comparison

Bullet-by-bullet Full resume Shipped
What the AI controls One bullet at a time. Has the user's resume as context, but no say in what gets cut, combined, or emphasized across the whole. The whole resume. Can balance content, avoid duplication, and weight what matters for the role.
What users still have to do Decide what to keep, cut, and combine on their own, the editorial work early-career users need help with most. Light polishing if they want it, like adjusting a phrase to sound more like them.
Time to ship Fast. Narrow scope, simpler prompts. Heavier. Required a system that reasoned across the resume.

The tradeoff

In V1, generating the whole resume meant users couldn't iterate on a single bullet alone. I accepted that cost. Bullet quality, and how bullets work together, depends on a holistic view of the resume.

The reception

What users said

I find it more useful than other tools I have tried (such as Job Scan), especially in providing specific, non-generic feedback.
Undergraduate Student in Computer Science
The chat-based optimization felt accurate and useful. It made my resume more specific and ATS-friendly.
Undergraduate Student in Data Science
The detailed scoring plus suggestions helped me know exactly where to improve.
Graduate Student in Computer Science