Tomas Hoyos
Cofounder and CEO of Apt, an AI career mentoring platform
Where to hear and read about Tomas Hoyos1 episode or test
Tomas Hoyos is cofounder and CEO of Apt, an AI career mentor that grew out of a personality and aptitude quiz, and he came to it after leaving private equity and surviving two failed startup pivots. The most useful thing to learn from him is how he turns a long, conversational onboarding quiz into a measured conversion funnel while building a defense against OpenAI and Anthropic.
we're just going to launch one thing every single week, uh, until something happens here.
Building the AI Career Mentor With a Freemium Bet
Tomas Hoyos: the story and the lessons
1How Tomas Hoyos went from private equity to Apt
Hoyos left private equity because he wanted to build, not engineer returns. His fund bought slow-growing firms with large physical assets, 30 to 40 years old, and loaded them with debt. He says the work taught him how to value, buy, run and exit a business.
He thought about technology on nights and weekends, and leaving meant stepping into deep uncertainty. He felt finance asks how to make the most money, not what to build to improve lives.
His first startup, Voro, lost about 90% of its users in two weeks. Voro was a health social network that surfaced doctors recommended by friends, plus quality data such as risk-adjusted readmission, mortality and complication rates for surgeons. It reached several million users, all organically, then monthly actives fell about 90% when COVID hit in March 2020.
Asked how to pivot, investor Mike Maples of Floodgate answered with stories: Odeo became Twitter, and a head-camera show became Twitch.
Hoyos shipped one new product a week until something worked. After the collapse, the team split into small groups and launched a product every week. One, a Zoom live-event tool for creators, was built in 48 hours and took $100,000 in gross sales in its first week.
It grew to a few million in volume, then faded as the world reopened. The team sold assets and returned capital to investors.
The failed pivots led him to his cofounder, James. James was an angel investor in the live-event company, and the two hit it off at once. They had long wanted to start something together.
James brought the seed idea from his father's recruiting firm, where as a boy he cold-called clients. His father gave candidates personality and aptitude tests, and James proposed a modern version because existing tests really aren't that great. Hoyos saw that AI could make the test a conversation.
2What Apt is and why Hoyos is building it
Hoyos argues AI will unsettle careers while few tools help people through it. He frames a career as about 80,000 hours that people plan alone, in single player mode. He expects hundreds of millions, maybe billions, to ask what they are good at and what to do next.
He cites Apt's own figures: 7 in 10 people say they struggle in their career and only 4% say they love what they do. These come without baselines or sources.
Apt started with a narrow wedge: one quiz that scores many tests. A single 15 to 20 minute assessment produces scores on Big Five, Holland Code, MBTI, DISC, Enneagram and more, plus several hundred psychometric traits. About two million people have taken it.
The hard part was choosing questions so one sitting could feed every test. The team built its own algorithm and scoring and kept only scientifically valid, reliable measures, then added AI on top. The slogan was one test to rule them all.
Apt expands from self-knowledge into paths, job matches and an always-on mentor. Users get scores, then career paths that fit, each explained with a typical day, pay, growth and exit options. Apt then helps with applying, negotiating the offer and thriving in the role.
The aim is a proactive mentor that spots a below-market salary, drafts a 6 to 8 week raise plan, or points to a growing local company paying 8% more with a cold email to the founder.
Hoyos chose the word mentor over coach on purpose. Career coaching is a roughly $7 billion market, but people skip it because it is stigmatized, costs thousands and needs scheduled hours.
A mentor in your pocket, available 24/7 for less than a cup of coffee a day, removes that friction. Hoyos says Apt is very intentional about the word.
3How Apt's conversational quiz drives conversion
Switching to an AI chat quiz lifted conversion by more than 60%. Apt replaced multiple choice with a back-and-forth where the mentor reacts to each answer and asks a follow-up. It feels more enjoyable than a school-style test, and probing questions make users believe the product understands them.
Hoyos says it is 'so engaging and just drives up conversions so much that it's definitely worth it.' The figure comes with no baseline or test details.
Hoyos found a longer quiz converted better than a short one. Apt's 15 to 20 minute quiz beat shorter versions in A/B tests. People who invest time are more curious about the payoff and more likely to pay.
Hoyos admits it is hard to say exactly why. Sunk cost is the likely reason, and a deeper question set also makes the results more believable.
He instruments every step of the quiz funnel. Apt tracks landing page visits, quiz starts, drop-off at each question, paywall views and paid conversions, and A/B tests the landing copy that gets people to start.
Quizzes are highly quantifiable, so each step can be tuned.
He accepts messy chat data and analyzes it afterward. Free-form answers lose the clean segments of multiple choice. Apt runs LLM analysis on the conversations to tag and bucket users and to read core motivations they never stated.
It adds a step but opens many more directions to personalize the product. Hoyos says the fact that it's probabilistic is like a feature, not a bug.
4Hoyos on staying defensible against ChatGPT and Claude
He defends Apt with proprietary data, a deep workflow and network effects. Hoyos takes the three-part framework from Bill Gurley. Apt scores people on hundreds of variables, including image rotation, hole-punched paper unfolding and inductive and deductive reasoning tests, which general chatbots do not do.
The workflow has many steps: narrowing by city and pay, finding connections, applying, interviewing and negotiating the offer. The two million assessments are the proprietary dataset.
Apt matches people to jobs from its own database, not web search. The database holds over 10 million open jobs. Apt splits each profile into about 500 vectors, matches them to jobs by cosine similarity, then adds filters.
A chatbot runs a slow web search instead. Hoyos does not expect the labs to care enough to build a dedicated version soon, and calls caring more squishy but huge.
He bets on an opinionated interface and a trusted brand. In high-stakes areas like career, Hoyos says people want a dedicated product, not a blank chat page or one mega super app.
The interface should deliver dense information efficiently and engagingly. He notes the worry 'what if Google does this' became 'what if Meta does this' and is now about the labs, and the old rules still apply.
5Hoyos on monetization, freemium and the marketplace
His biggest regret was postponing the business model at Voro. Because Voro was a social network, the team kicked the can on monetization. At Apt they set the business model up front and charged from day one.
Charging also forces a user acquisition mechanism aligned with paying customers, which he says matters as vibe-coded apps flood the app store.
He treats paid acquisition as product development. An adviser who bootstrapped a mobile games business to nine figures told him to treat early ad spend as tuition. Spend about $100 on Google, run users through the quiz and learn who they are and what they want.
Then iterate on the product with tight feedback loops.
Hoyos still bets on freemium, even though it hurt conversion so far. Apt's freemium tests lowered paid conversion against a hard paywall. He wants free reach to create the category and, as he puts it, suck the oxygen out of the room for rivals.
The paid tier would cost about 20 or 30 dollars a month. Free job seekers add liquidity for a marketplace that earns placement fees from employers. No data yet shows freemium paying off.
He plans a marketplace that learns from every hire. Phase two is matching seekers and employers using deep profiles. Seekers compete with 5,000 to 7,000 others per posting, and employers can get 7,000 applications in an hour.
Because the mentor relationship lasts, Apt can see how matches turn out, and employer feedback improves the algorithm with every placement. No evidence is given yet that the matching predicts good hires.
6What Hoyos believes about work and energy
Hoyos says what energizes you is the best guide to a career. He tells people to look for places that feel like a time warp and things that feel easy. He wanted to be an athlete as a kid, but what he loved was coaching and leading.
He cites Graham Weaver's line that energy is the language of the soul. His billboard for humanity would ask what energizes you.
He expects AI to change the shape of work, not end it. Hoyos foresees a Cambrian explosion of opportunity and many more entrepreneurs with AI teammates. He points to Stripe's business formation stats and the surge of apps in the app store.
A 1900s coal miner would not call a food video creator's job work, and the same may be true of future jobs.
Building the AI Career Mentor With a Freemium Bet
Apt turned a personality and aptitude quiz into an AI career mentor, using a long conversational onboarding quiz as its main conversion funnel. The episode covers why longer quizzes converted better, how an AI chat quiz lifted conversion more than 60%, and how an app-layer startup can stay defensible against OpenAI and Anthropic.
Hoyos left private equity because he wanted to build, not engineer returns. His fund bought slow-growing firms with large physical assets, 30 to 40 years old, and loaded them with debt. He says the work taught him how to value, buy, run and exit a business.
He thought about technology on nights and weekends, and leaving meant stepping into deep uncertainty. He felt finance asks how to make the most money, not what to build to improve lives.
His first startup, Voro, lost about 90% of its users in two weeks. Voro was a health social network that surfaced doctors recommended by friends, plus quality data such as risk-adjusted readmission, mortality and complication rates for surgeons. It reached several million users, all organically, then monthly actives fell about 90% when COVID hit in March 2020.
Asked how to pivot, investor Mike Maples of Floodgate answered with stories: Odeo became Twitter, and a head-camera show became Twitch.
Hoyos shipped one new product a week until something worked. After the collapse, the team split into small groups and launched a product every week. One, a Zoom live-event tool for creators, was built in 48 hours and took $100,000 in gross sales in its first week.
It grew to a few million in volume, then faded as the world reopened. The team sold assets and returned capital to investors.
The failed pivots led him to his cofounder, James. James was an angel investor in the live-event company, and the two hit it off at once. They had long wanted to start something together.
James brought the seed idea from his father's recruiting firm, where as a boy he cold-called clients. His father gave candidates personality and aptitude tests, and James proposed a modern version because existing tests really aren't that great. Hoyos saw that AI could make the test a conversation.
Hoyos argues AI will unsettle careers while few tools help people through it. He frames a career as about 80,000 hours that people plan alone, in single player mode. He expects hundreds of millions, maybe billions, to ask what they are good at and what to do next.
He cites Apt's own figures: 7 in 10 people say they struggle in their career and only 4% say they love what they do. These come without baselines or sources.
Apt started with a narrow wedge: one quiz that scores many tests. A single 15 to 20 minute assessment produces scores on Big Five, Holland Code, MBTI, DISC, Enneagram and more, plus several hundred psychometric traits. About two million people have taken it.
The hard part was choosing questions so one sitting could feed every test. The team built its own algorithm and scoring and kept only scientifically valid, reliable measures, then added AI on top. The slogan was one test to rule them all.
Apt expands from self-knowledge into paths, job matches and an always-on mentor. Users get scores, then career paths that fit, each explained with a typical day, pay, growth and exit options. Apt then helps with applying, negotiating the offer and thriving in the role.
The aim is a proactive mentor that spots a below-market salary, drafts a 6 to 8 week raise plan, or points to a growing local company paying 8% more with a cold email to the founder.
Hoyos chose the word mentor over coach on purpose. Career coaching is a roughly $7 billion market, but people skip it because it is stigmatized, costs thousands and needs scheduled hours.
A mentor in your pocket, available 24/7 for less than a cup of coffee a day, removes that friction. Hoyos says Apt is very intentional about the word.
Switching to an AI chat quiz lifted conversion by more than 60%. Apt replaced multiple choice with a back-and-forth where the mentor reacts to each answer and asks a follow-up. It feels more enjoyable than a school-style test, and probing questions make users believe the product understands them.
Hoyos says it is 'so engaging and just drives up conversions so much that it's definitely worth it.' The figure comes with no baseline or test details.
Hoyos found a longer quiz converted better than a short one. Apt's 15 to 20 minute quiz beat shorter versions in A/B tests. People who invest time are more curious about the payoff and more likely to pay.
Hoyos admits it is hard to say exactly why. Sunk cost is the likely reason, and a deeper question set also makes the results more believable.
He instruments every step of the quiz funnel. Apt tracks landing page visits, quiz starts, drop-off at each question, paywall views and paid conversions, and A/B tests the landing copy that gets people to start.
Quizzes are highly quantifiable, so each step can be tuned.
He accepts messy chat data and analyzes it afterward. Free-form answers lose the clean segments of multiple choice. Apt runs LLM analysis on the conversations to tag and bucket users and to read core motivations they never stated.
It adds a step but opens many more directions to personalize the product. Hoyos says the fact that it's probabilistic is like a feature, not a bug.
He defends Apt with proprietary data, a deep workflow and network effects. Hoyos takes the three-part framework from Bill Gurley. Apt scores people on hundreds of variables, including image rotation, hole-punched paper unfolding and inductive and deductive reasoning tests, which general chatbots do not do.
The workflow has many steps: narrowing by city and pay, finding connections, applying, interviewing and negotiating the offer. The two million assessments are the proprietary dataset.
Apt matches people to jobs from its own database, not web search. The database holds over 10 million open jobs. Apt splits each profile into about 500 vectors, matches them to jobs by cosine similarity, then adds filters.
A chatbot runs a slow web search instead. Hoyos does not expect the labs to care enough to build a dedicated version soon, and calls caring more squishy but huge.
He bets on an opinionated interface and a trusted brand. In high-stakes areas like career, Hoyos says people want a dedicated product, not a blank chat page or one mega super app.
The interface should deliver dense information efficiently and engagingly. He notes the worry 'what if Google does this' became 'what if Meta does this' and is now about the labs, and the old rules still apply.
His biggest regret was postponing the business model at Voro. Because Voro was a social network, the team kicked the can on monetization. At Apt they set the business model up front and charged from day one.
Charging also forces a user acquisition mechanism aligned with paying customers, which he says matters as vibe-coded apps flood the app store.
He treats paid acquisition as product development. An adviser who bootstrapped a mobile games business to nine figures told him to treat early ad spend as tuition. Spend about $100 on Google, run users through the quiz and learn who they are and what they want.
Then iterate on the product with tight feedback loops.
Hoyos still bets on freemium, even though it hurt conversion so far. Apt's freemium tests lowered paid conversion against a hard paywall. He wants free reach to create the category and, as he puts it, suck the oxygen out of the room for rivals.
The paid tier would cost about 20 or 30 dollars a month. Free job seekers add liquidity for a marketplace that earns placement fees from employers. No data yet shows freemium paying off.
He plans a marketplace that learns from every hire. Phase two is matching seekers and employers using deep profiles. Seekers compete with 5,000 to 7,000 others per posting, and employers can get 7,000 applications in an hour.
Because the mentor relationship lasts, Apt can see how matches turn out, and employer feedback improves the algorithm with every placement. No evidence is given yet that the matching predicts good hires.
Hoyos says what energizes you is the best guide to a career. He tells people to look for places that feel like a time warp and things that feel easy. He wanted to be an athlete as a kid, but what he loved was coaching and leading.
He cites Graham Weaver's line that energy is the language of the soul. His billboard for humanity would ask what energizes you.
He expects AI to change the shape of work, not end it. Hoyos foresees a Cambrian explosion of opportunity and many more entrepreneurs with AI teammates. He points to Stripe's business formation stats and the surge of apps in the app store.
A 1900s coal miner would not call a food video creator's job work, and the same may be true of future jobs.