Frederik Sally is trying to teach machines to read the parts of a conversation that never make it into words: the hesitation before an answer, the confidence behind a claim, the nervous energy under a smile. If he and his team succeed, one of the first careers their technology could make obsolete is his own mother's. She has spent her life running a communication consultancy and writing books on how people talk to each other. Frederik tells her as much, and she jokes that he might just put her out of a job. He isn't sure the two things are connected. He suspects it's more coincidence than inheritance. But it's hard to ignore that he also has a grandfather still launching ventures at almost 90 and a biological father who ran one of Denmark's first commercial radio stations. Something in the family clearly likes building things other people said couldn't be built. That instinct led him and two co-founders, Paula Petcu and Line Clemmensen, to found Interhuman AI, a Copenhagen startup building what Frederik calls a "social intelligence layer" for AI. The idea didn't start where it is now. It began as a therapist app, failed to find buyers, and pivoted hard once the founders noticed that companies were far more interested in the underlying technology than in the product itself. That pivot has since raised a €2 million pre-seed round, and Interhuman AI is now betting on sales training and user research as the use cases that prove the technology works before the bigger, harder problems (cultural nuance, healthcare, therapy) are ready to tackle. Whatever stage you're building at, this conversation will land somewhere familiar: being early to a market that isn't ready yet, discovering your actual business by watching what customers ask for instead of what you set out to sell, and using story and branding to buy yourself time while the technology catches up. 🔑 In this episode, we talk about: What Interhuman AI's "social intelligence layer" actually reads (tone, hesitation, confidence, body language) and why today's LLMs miss all of it Growing up in a family of entrepreneurs: a grandfather still starting companies near 90, a mother who built a career teaching communication, and a biological father who ran early commercial radio in Denmark The pivot from a failed therapist app to licensing the underlying tech to sales-training and user-research platforms Raising a €2 million pre-seed round, and why deep tech is a harder sell to European investors than American ones How they collect training data: Creative Commons sources, synthetic data, and recording their own lives, including a rose party, with Meta smart glasses Competing with hyperscalers like OpenAI and Anthropic and specialists like Hume AI, and why depth beats general intelligence for this problem Turning down an investor who wanted the technology used for defense and lie detection Building the company's ethical guardrails around the EU AI Act A pilot with the Region of Copenhagen helping parents practice hard conversations with kids who have additional needs Why Frederik thinks a single homemade video did more to open investor doors than any slide deck The 93%-nonverbal statistic he's been told by his own behavioral scientists to stop quoting 🧠 Key Takeaways Sell what the market is asking for, not what you set out to build. Interhuman AI's therapist app flopped, but the underlying technology kept generating inbound interest. That signal, not the original plan, became the business. Being early means testing constantly, not waiting to be proven right. With no playbook to follow, the team treats every week as a chance to pivot, informed by pilots and customer feedback rather than a fixed roadmap. Sometimes speed beats accuracy. Their new "real-time experimental" model trades some precision for instant feedback, on the theory that a rough first read (much like meeting a stranger) can still be useful while a fuller picture builds over the conversation. Draw your ethical line before you need the money. The founders turned down a defense-focused investor before they had funding secured, so there was never a moment where the answer might have been different. Branding can open doors the technology hasn't earned yet. A short, well-made video of signals overlaid on a conversation did more to get early investor meetings than research slides, and the team kept using it for almost two years. People remember the trust a conversation created, not what was said in it. Two years after meeting at a conference, neither Frederik nor Helga can recall their actual conversation, only that it built enough trust to stay in touch, which is itself the thesis Interhuman AI is built on. Specialization can beat scale for narrow, deep problems. Frederik doesn't think general models will out-compete focused ones on things like cultural nuance and personal baseline, because that depth isn't what the big labs are optimizing for. A mismatched skill set among co-founders can be the real fundraising asset. Investors backed the combination of Frederik's storytelling, a technical AI researcher, and a data scientist, not any one of those alone. 🗣️ Sound Bites "She tells me that if we succeed in teaching this AI the things we are trying to do, it could potentially take over what she's been doing her whole life.""You're nodding right now, for example. That is a signal that is important in this conversation. You're understanding what I'm saying, you're acknowledging, you're listening. All these signals we're trying to teach a model to pick up on.""I can remember you, and I can remember maybe two more, out of the maybe fifty people I talk with at these conferences. So that's definitely something.""We turned down an investor that wanted us to go the defense route. Because to the bone, we want to do AI for good.""Once they saw this very short video clip, they immediately understood it. From that, we just learned that it means something, what you show." 📍 Chapter Highlights 00:03 – Kicking off, and checking the clock before diving in 00:31 – Welcoming Frederik, founder of Interhuman AI: teaching machines to read what's beneath the words 03:07 – Frederik's childhood and a family of entrepreneurs 04:03 – A grandfather still launching ventures near 90, and a mother who teaches communication for a living 07:03 – The original idea: a therapist app 09:23 – The pivot: companies wanted the underlying tech, not the app 10:26 – The €2 million pre-seed round, and why deep tech is a harder sell in Europe 11:51 – The biggest current challenge: data, research, and a field moving week to week 13:55 – Collecting data with Creative Commons sources, synthetic data, and Meta glasses at a rose party 15:28 – Positioning against hyperscalers and specialists like Hume AI 20:29 – Choosing sales training as the first commercial use case 23:11 – Explaining Interhuman AI to someone hearing it for the first time 24:34 – Why it's not a lie detector, and turning down a defense-focused investor 25:41 – The risk of young people leaning on chatbots for emotional support 27:52 – The 93%-nonverbal statistic, and why he's been told to retire it 29:32 – Do companies know how much of their message gets lost beyond the words? 34:48 – Staying motivated and pivoting weekly when the market isn't ready yet 36:18 – Trading accuracy for speed: building an instant "real-time experimental" model 39:56 – Callback to the Tech BBQ in Copenhagen, two years on 41:45 – "We don't remember the words. We remember the trust." 42:56 – Cultural differences, personal baselines, and reading the same signal differently 46:43 – The real-world stakes of getting it wrong: AI interviewers, healthcare, therapy 47:13 – Where Frederik draws the ethical line, and building around the EU AI Act 51:14 – Piloting with the Region of Copenhagen to help parents of kids with additional needs 57:26 – Advice to his younger self, and the sales lessons learned the hard way 1:00:51 – Why demos, not decks, win meetings 1:01:39 – The two-year-old video that opened investor doors 1:06:28 – The co-founder mix that actually got them funded, and growing to 12 people 1:11:58 – Where to find Frederik and Interhuman AI, and closing thoughts 🔗 Links & Resources Interhuman AI: Interhuman AI Frederik Sally on LinkedIn: linkedin.com/in/frederiksally Interhuman AI's €2M pre-seed coverage: EU-Startups, Sifted 🏢 Companies / Organisations Mentioned Interhuman AI: Copenhagen startup building a social intelligence layer that reads nonverbal social signals for AI systems Hume AI: competitor building emotion-focused AI models, referenced as working from a different (basic-emotions) framework Region of Copenhagen: Danish regional government piloting an Interhuman AI app to help parents of kids with additional needs Tech BBQ: the Copenhagen event where Helga and Frederik first met, two years before this recording SERIOUS.BUSINESS: host Helga's company 🧡 Why You'll Love This Episode This one has a twist most founder stories...
