input · the_brief TXT

2024 — Design Research · Harbour.Space × Campus AI

Claims with receipts

Campus AI was building an LLM tool for structured learning and wanted to know who their most motivated users would be. That pointed us at career shifters: mid-career professionals leaving one field for another. Our job was to understand how they pick a new path, how they learn, what keeps them going, and how they know they are making progress.

facts_data DATA
ROLE
Design Researcher (team of 4)
YEAR
2024
METHOD
42-respondent survey, 8 × 90-min interviews, 1,117 highlights coded across 182 tags
CONTEXT
Harbour.Space × Campus AI
campus_ai_view · FUNNEL.scn VIEW
SCREENEDSEATSHIGHLIGHTSTHEMESRECS ─▸─▸─▸─▸ 1,117 CODED MOMENTS HUMAN FACE PROXY FEEDBK GPS ROUTE JOB MARKET

42 SCREENED → 8 INTERVIEWED → 1,117 CODED → 7 THEMES → 4 RECOMMENDATIONS · REAL STUDY NUMBERS

section_break SYS
01 THE STUDY
process · research_questions TXT

Four research questions

RQ1

What must shifters know before the learning starts?

path choice · investigation · what to learn

RQ2

How do they prefer to learn, and how do they judge each method?

structure · tools · value of artifacts

RQ3

How do they maintain motivation over months of self-directed work?

life factors · learning factors

RQ4

Which tools and strategies worked, and which wasted their time?

challenges · pain points · dead ends

process · how_it_was_built TXT

How the study was built

The research questions came first, four of them, drafted as a team: how shifters choose a path, how they prefer to learn, how they stay motivated, and which tools worked or wasted their time. Everything else was built to answer them. The screener enforced quotas, gender balance, ages 20s through 40s, mostly already mid-shift, so the eight seats went to exactly the people the product bet depended on. We piloted the protocol on a representative participant before running it for real, and we cut screener questions we liked because they would not change who got a seat.

process · how_a_session_ran TXT

How a session ran

Each interview was a 90-minute protocol in seven timed movements: rapport and a typical weekday, the shift story, researching the new career, learning methods, challenges, quality of learning, and life after the shift. Stakeholders shadowed with cameras off and asked questions only at the end. After every day of sessions the team debriefed on a fixed template, what surprised us, what mattered to this person, what we would ask differently, and the guide was adjusted between sessions. A topline report shipped within days for fast product decisions; formal coding came after, so speed never substituted for rigor.

data · screener_quotas DATA

Who got a seat

  • 42 SCREENED → 8 SEATS
  • GENDER BALANCE 4M / 4F
  • AGES 20s–40s · HIGHER-ED
  • SHIFTED IN LAST 2 YRS 4–6
  • CONSIDERING A SHIFT 2–4
  • PLATFORM DIVERSITY REQUIRED

RECRUITED VIA RESPONDENT + PERSONAL NETWORKS · CONSENT + NDA SIGNED AHEAD · GIFT-CARD INCENTIVES

sys · analysis_pipeline SYS
section_break SYS
02 THE EVIDENCE
process · the_coding TXT

Method

A 42-respondent screener found the shifters; eight became 90-minute open-ended interviews across four countries. The four of us coded the transcripts in Dovetail, 1,117 highlighted moments across 182 tags, and built the findings from the patterns that held up across interviews. The excerpts below are the real data, first names only, with the coding intact.

campus_ai_view · coded_transcripts VIEW
STAR QUOTES · ALL THEMES REAL STUDY DATA · FIRST-NAME-ONLY · EXCERPTED
STAR

EVIDENCE MAP · EDGE = TAGGED MOMENTS

CODED IN DOVETAIL BY A TEAM OF 4 · QUOTES VERBATIM, FILLER TRIMMED
section_break SYS
03 WHAT WE FOUND
insight_01 TXT

Shifts start with people, not job boards

Nobody described searching for a new life. Shifters keep an open ear: a person, a post, a story of someone who already made the move arrives first, and then the research starts. A learning tool that assumes the journey begins at a search box has missed the first chapter.

11 MOMENTS · 7 OF 8NAMED STUDY INSIGHT
insight_02 TXT

Transferable skills point, personal interest pulls

Every participant weighed the same two factors: what carries over from the old career, and what they actually want to do. Transferable skills made a shift feel feasible and shortened the imagined roadmap. When interest and skills disagreed, interest won, and the extra learning was accepted as the price.

44 MOMENTS · 8 OF 8
insight_03 TXT

Proxy communities do the validating

Before anyone paid for a course or quit anything, they checked the move against strangers who had already made it: Reddit threads, reviews, testimony from people one field ahead. It is the densest theme in our coding, and it is how shifters without a network borrow one.

77 MOMENTS · 8 OF 8 · DENSEST THEME
insight_04 TXT

They want a playbook for the whole shift

Skills were one chapter of the ask. The recurring request was the whole route, phase by phase: what to learn, what to build, who to talk to, when to start applying. Nobody had found the hub that covers breaking out of one career and landing in another, so everyone was assembling it by hand.

17 MOMENTS · 5 OF 8
insight_05 TXT

Progress is something you can show

Progress was measured in artifacts: a project that works, a design you can show, a thing carried from course to practice. Certificates surfaced in 11 of 1,117 highlights, almost always as a door pass for specific fields, proof for an interview rather than proof of learning.

41 MOMENTS · 7 OF 8CERTS: 11 OF 1,117
insight_06 TXT

Human contact: most valued, least supplied

What shifters valued most is what self-serve learning supplies least: a person who knows them, looking at their work. Feedback, immediate clarification, personalized guidance, motivation. Participants with mentors moved with confidence; the ones without described submitting work into a void.

72 MOMENTS · 7 OF 8
insight_07 TXT

Time is the currency of the shift

Every hour of learning came out of the margins of a full life: around jobs, families, savings runway. Participants defended it with calendars, spreadsheets and self-imposed deadlines. A wrong turn does not cost a learner content, it costs months they had already counted.

41 MOMENTS · 8 OF 8
section_break SYS
04 WHAT WE RECOMMENDED
rec_01 TXT

Give the AI a human face

Learners trusted content they could pin to a person: an instructor with a track record, a founder teaching their own product, a face in the comments. So present the AI through visible personas instead of an anonymous engine. We sketched two: Professor AI, the authored voice behind learning guides, and Students AI, the comment layer of other learners. The trust cue that costs nothing: the persona remembers your past sessions and says so.

FROM ▸ CREDIBILITY + INSTRUCTOR-EXPERIENCE TAGS · INSIGHTS 03 + 06

rec_02 TXT

Proxy feedback in the peer gap

The peer gap was not abstract: participants submitted work into a void and studied alone across language barriers. Use the AI where the learner has no one to ask: prompts that check in during learning, immediate correction with explanation when a test goes wrong, and critique flows where submitting an artifact returns feedback in the register of a senior peer. Honestly labeled as a stand-in.

FROM ▸ LACK-OF-PEERS + FEEDBACK-LOOP TAGS · INSIGHT 06

rec_03 TXT

Destination on screen from day one

Learning outcomes are the destination for a GPS: show what the learner will be able to make at the end of each section, up front. And when they take a wrong turn, re-route to the same destination instead of failing them; no red X states, only recalculated routes. Time is why this matters: a wrong turn that costs a month is how shifts die.

FROM ▸ ROADMAP + TIME THEMES · INSIGHTS 04 + 07

rec_04 TXT

Build the roadmap against the live job market

A job is what career shifters are actually paying for, and the job market is the context employability is measured against. So gather the user's career goal, derive waypoints from live posting data, explain how each step serves the goal, and let users nudge the journey with suggested tags. The same mechanism can suggest entirely new paths when performance says the learner is suited to one.

FROM ▸ JOB-DESCRIPTION SELF-ASSESSMENT · INSIGHTS 04 + 05

out · what_it_settled TXT

What it settled

The study still shapes how I design AI products: I optimize for what the person can do afterwards and treat engagement as a side effect, not the goal.

▸ WANT THE FULL WRITE-UP? CONTACT ME