Designing Career Intelligence
Making Sense of the Job Market
I helped CareerBuilder turn dormant workforce data into a career intelligence product that recorded 250,000 monthly visitors and was later integrated into the core platform as 'Career Insights'.
The problem
CareerBuilder was, for years, a name you’d hear when someone was actively looking for a job. Type in a keyword, scroll through listings, click apply, move on. It was a useful tool, but with no real reason to come back. The business was built on transactions, not relationships.
I joined as Senior UX Designer to lead the experience for CareerCoach, a “big bet” project with a clear mission: turn the company’s massive pile of dormant workforce data into career intelligence, something that could actually help people manage their careers. If we pulled it off, CareerBuilder could shift from being a job board into a career partner, a resource job seekers would keep coming back to.
We had four months to prove the concept. We operated as a small, distributed team across London, Paris, Munich, and Atlanta, working separately from the main business. It was a deliberate move: test radical ideas without affecting the core product, and move fast enough to beat competitors who were circling the same opportunity.
Discovery
The starting hypothesis was simple: people need help understanding their career path. But we needed to check if that was true, and if so, what form that help should take.
Over a couple of months, I spoke with around 50 job seekers across the UK, USA, Germany, and France. The conversations were more revealing than I’d expected. It wasn’t just that people didn’t know what skills to develop. They didn’t even know how to assess where they stood in the first place. They had no way to quantify their market value, no clear picture of what roles they were actually qualified for, and no idea which skills would make the biggest difference.
There was a confidence gap. People would look at job descriptions and think, “Maybe I could do that,” but had no data to back it up. Others felt stuck in roles they’d outgrown, but couldn’t see a clear next step. The anxiety wasn’t just about finding a job. It was about making the right career move without a map.
We took those insights back to the team and turned them into a product hypothesis. To ground it, we ran a two-day workshop in Paris where we built our own set of personas, a subset of CareerBuilder’s core groups, tailored to the career-progression problem rather than the job-search one.
The project was also a deliberate chance to practise Lean UX properly. We’d attended a workshop run by Jeff Gothelf at Numa in Paris, and this felt like the right place to put the methodology to work: prioritise learning over delivery, and let the research steer.
The turning point
From day one, we committed to learning over building. In many companies, success is measured by how many features you ship. In our little lab, success was measured by how quickly we could test an idea and decide whether to keep going or change direction.
One of our early assumptions was that a mentoring marketplace would be the core of the product. The thinking was that once people saw where they needed to improve, they’d want to connect with someone who could help. We spent time sketching the concept, mapping user flows, even building early prototypes.
Then we tested it. The feedback was consistent. Users weren’t ready to ask a stranger for career mentoring. They wanted to understand the data first, to see the proof that they needed to develop certain skills before they’d consider paying for guidance. The mentoring angle felt premature.
We made the call right there: kill the mentoring feature, and put the whole effort behind the career intelligence that had been the mission from the start. It was the right decision, and it freed up engineering effort to focus on what users actually valued.
Killing mentoring freed us to focus entirely on the career intelligence itself. What followed was a period of rapid testing, and the lessons from it became the principles behind the final product.
What testing taught us
The first lesson came quickly. Users didn’t trust a single “match score” out of context. Show them a number like “70% fit” and they’d assume the algorithm was guessing. But present them with raw workforce data, and they’d feel overwhelmed.
The breakthrough came when we started showing the answer first: the clear, actionable insight. Then we’d layer in the evidence behind it. We called this the “Verified Insights” model.
For example, instead of just saying “You’re a 70% match for this role,” we’d show the headline and then provide a breakdown using benchmarks from official sources like the Bureau of Labor Statistics and O*NET: the specific skills the role requires, how your CV compares, and where the gaps are. Once users saw legitimate, government-backed data behind the recommendation, they stopped treating it as a guess and started trusting the guidance.
Then something strange happened. We’d built a “view full details” screen so people could interrogate the data themselves, and they stopped looking at it. At first that felt like a failure. But the interviews explained it: users didn’t want to work through the data, they wanted to know it was there. The option itself was what built the trust. We’d hear “How did you figure that out?” and, when the answer didn’t match their expectations, “That’s not what we got, what are your workings?” The open door to the data gave them confidence they didn’t have when it was closed, even if they never walked through it.
The final lesson was about comparison. Users had a strong compulsion to benchmark themselves against their peers, especially on skills. They saw the gap as a route to upskilling rather than a judgement.
By the end of testing, three principles were clear: show the answer first, keep the door to the data open, and make every insight actionable.
Shaping the solution
Those principles became the structure of the product: a three-stage diagnostic journey that felt like a conversation with a career consultant.
-
Stage 1: Where are you now? Upload your CV and the parsing engine did the heavy lifting, extracting your skills, experience, and qualifications without you filling in a single form. The system then compared your profile against the market to show your current standing. For instance, “Your profile puts you in the top 10% of candidates for senior marketing roles in your region.” The answer came first, with the evidence behind it.
-
Stage 2: Where do you want to go? You’d select a target role, and we’d pull in labour market data and CareerBuilder’s internal job seeker data to show the gap. “60% of people in this role have Python certification. Based on your CV, you’re a 40% match.” Or “This role in New York pays about 12% more on average than in your current city.” This was the peer comparison in action: users could see exactly where they stood against the market.
-
Stage 3: Here’s how you get there. We’d give you a prioritised list of skills to develop, each one linked to active job postings on CareerBuilder where that skill was required. Every recommendation was actionable: you weren’t just preparing for a future role, you could see opportunities that fit your growing profile right now.
The journey connected career planning with relevant CareerBuilder job listings. Its final stage gave people somewhere practical to apply what they had learnt while supporting the business goal of connecting them with open roles.
Building and testing
Making the three-stage journey work meant constant iteration: sketch, prototype, test, repeat.
For quick validation, pen-and-paper and Balsamiq wireframes got ideas in front of users within minutes. When an interaction depended on real data, we built functional HTML and JavaScript prototypes that felt close to the finished thing.
We ran constant A/B tests, testing everything from button copy to chart types. One test compared different ways of showing skill gaps: complex radar charts versus simple bar charts. The bar charts won by a clear margin. Users understood them faster and felt more confident interpreting the results.
We also built micro-experiments to answer specific questions. CareerBuilder had three competing skills parsing engines, each matching words in a CV to skills in its own way. Instead of asking the data team to guess which was best, we built a small interface where users uploaded their CV, saw it parsed by each engine, and ranked how well the tagged skills represented them. That qualitative input guided our choice of engine and fed useful feedback back to the teams who built them.
The outcome
The MVP version of CareerCoach launched under its own distinct brand, positioned as a standalone proof of concept. We placed it prominently in CareerBuilder’s global navigation, giving it high visibility while keeping it separate from the main platform. The goal was simple: prove the concept worked, fast and with minimal investment.
The early traffic supported our core hypothesis that people were interested in using workforce data to understand their career options.
After the initial launch, the effort shifted from validation to scaling. CareerBuilder launched a programme to expand the proposition and integrate CareerCoach into the main ecosystem.
The integrated version, rebranded as “Career Insights”, used CareerBuilder’s full visual design language and became part of the core product suite. Career intelligence was no longer a separate sandbox experiment; it was available through the main CareerBuilder platform.
Not long after CareerBuilder launched Career Insights, LinkedIn introduced a similar career intelligence feature. Its arrival reflected wider market interest in this kind of guidance. By then, CareerBuilder had already tested the concept and integrated it into its core platform.
The skunkworks approach had paid off. By operating outside the main brand, we could test bold ideas quickly, learn fast, and bring successful experiments back into the core business without the risk of damaging the existing platform.
What I’d carry forward
Some of our choices were hard to justify to stakeholders because nobody was asking for them. The “view full details” option was the clearest example: few people clicked it, so on paper it looked like wasted effort. Making the case meant pulling clips from user interviews and showing how the option itself helped people trust the recommendation.
With team members in four countries, we gave individuals ownership of their work while keeping everyone aligned through regular check-ins and permanent live video feeds in each location. It wasn’t about documenting every decision in meticulous detail. It was about making sure the right people were talking at the right times, and that context flowed naturally across the distributed team.
The four-month timeline meant trade-offs. We used off-the-shelf templates instead of building a custom visual system from scratch, an approach that was uncomfortable at times, but it let us focus on the user experience and data architecture. We knew we could refine the branding later if the concept held up.
What stayed with me most was how research helped the team decide where the product should focus. We weren’t improving an existing feature. We were testing whether workforce data could provide useful career guidance, then using the evidence to focus CareerCoach on career intelligence.