AI’s Invisible Leaks — Why Startups Need Causal Thinking and Stakeholder Clarity
As a founder, I’m often looking for the signals that separate the winners from the pack. Two recent talks at the first flagship House of AI event in London — one on the power of causal inference in AI, the other on resonating with stakeholders — provided both practical and philosophical messages, specifically that most AI-driven businesses are leaking value, and they don’t even know it.
TL;DR
Invisible Losses: Predictive AI models often miss the real impact of decisions, leading to hidden business losses (think: $800M Zillow mistake).
Clarity Beats Complexity: Stakeholders want actionable clarity, rather than technical jargon, because trust is built on credibility, reliability, and intimacy. It6’s not built on buzzwords.
Causal AI = Better Decisions: Understanding why things happen (causal inference) is more valuable than just predicting what will happen.
Actionable Takeaway: For founders, combine causal reasoning with clear, trust-building communication to avoid costly missteps and drive real value.
My Founder’s Lens: Don’t let speed, hype, or complexity blind you, instead ask “why,” focus on outcomes, and remember that the future is in your hands.
Anton Dudarenko speaking at the flagship House of AI event in London
The Hidden Cost of Predictive AI
Aleksander Molak’s session was meant as a wake-up call. He outlined how even the best teams, such as those at a major European bank, can completely miss the real effect of their actions when they rely solely on predictive models. In his example, every analyst said a marketing campaign would have zero effect. Yet, the true answer was that for every one-dollar spent, the bank would have made two-dollars. But, not a single analyst caught it.
The Zillow case is even more dramatic because their machine learning models, built on the assumption that the future would mirror the past, led to an $800 million loss in a single year. You may be wondering why this happened… Because the real world changed, and the models didn’t have a crystal ball, any more than you or I do.
Key insight: Predictive AI is great, at least until the world changes. If you’re not asking “why” things happen (causal inference), you’re flying blind.
Why “Why” Matters
Causal inference may seem like it’s an academic concept but it has relevant real-world application, because it’s about comparing alternatives, understanding the structure of problems, and choosing the right tools for the job. If you’re only predicting what will happen, you’ll miss the impact of your own interventions, like giving discounts to customers who would have stayed anyway, or running campaigns that quietly bleed cash.
I found myself reflecting on a very recent dinner conversation with a friend who has recently become a day trader. He told me that AI hasn’t helped him yet, but when I pressed him on where he thought it may help, he felt that it would be about the speed of executing transactions. However, speed without understanding can just be a faster way to make mistakes. AI won’t predict black swan events any better than a human if it doesn’t understand causality.
The Trust Equation: Clarity Over Complexity
Anton Dudarenko’s talk focussed on how we communicate AI’s value. Early in his career, he thought complexity was impressive, at least until a client interrupted his 35-slide deck and asked, “What should we do differently tomorrow?” This is the question every founder needs to have a ready-made and convincing answer for.
Anton has built a framework that appears to be simple yet powerful. He indicates that stakeholders want clarity, conviction, and direction. Trust is built on credibility, reliability, and intimacy, and buzzwords only get in the way. He talked through a financial services case study, whereby a giant in the sector gained 1.5% market share (a massive win at that scale) by focusing on actionable insights, rather than just models.
The contrast between WPP and Publicis drives it home: WPP’s AI narrative focused on savings and efficiency, losing them market share. Whereas, Publicis talked about value creation and talent empowerment, and won.
Founders: Here’s What to Do Next
Adopt Causal Thinking: Before you launch that next AI-driven feature or campaign, ask yourself, “What’s the real effect of this action? What if the world changes?” Use causal models in addition to predictive models.
Communicate for Action: Ditch the jargon and make your insights actionable. Then tie your insights into business outcomes. If you can’t explain your AI in plain English, you probably don’t understand it well enough yourself.
Build Trust: Use the trust equation by focussing on your credibility, reliability, and spending quality time with your stakeholders. Keep self-orientation (ego, hype) low.
Balance Speed with Depth: Don’t let the race to move fast blind you to underlying risks. Sometimes, slowing down to ask “why” can save you millions.
Aleksander Molak speaking at the flagship House of AI event in London
A Philosophical Note
There’s a Zen story Anton shared….
A trickster tries to outsmart a wise man with a butterfly in his hand, asking if it’s alive or dead. The wise man replies, “Everything is in your hands.” That’s the founder’s reality with AI. The tools are extremely powerful, but the outcomes depend on how thoughtfully — and transparently — we decide to use them.
The audience leaned in to both talks and they were inquisitive with their questions. The chat continued long after the event had finished, with all speakers surrounded, which indicated a genuine interest in the subject matter and some messages that struck a chord.
Final thought: As a second-time founder, I’ve learned that the biggest risk isn’t necessarily moving too slow, it’s about moving too fast without understanding the consequences. ‘More haste, less speed’ goes the famous proverb. Rather than more data and faster models, the future of AI in startups is about asking better questions, making better decisions, and building trust, using one clear and causal insight at a time.
Speaker Profiles
Aleksander Molak is a leading expert in causal inference and AI, renowned for making complex statistical and machine learning concepts accessible for business impact. He is the author of "Causal Inference & Discovery in Python," a guest tutor in causal machine learning at the University of Oxford, and the host of the Causal Bandits podcast. As founder of CausalPython IO, Aleksander delivers corporate and startup training on applying causal AI to real-world decision-making. His work bridges the gap between academic rigour and practical application, helping organisations avoid hidden losses by focusing on the why behind data-driven decisions.
Anton Dudarenko is a strategy consultant and brand-builder with a passion for analytics, stakeholder communication, and AI-driven value creation. As Founding Director of Lift-Off Consulting and former Consulting Director of Analytics at Kantar, Anton specialises in transforming complex insights into actionable strategies that resonate with business leaders. He advocates for clarity, conviction, and direction in stakeholder engagement, emphasizing trust-building over technical jargon. Anton’s methodologies have helped major financial services firms achieve significant market gains by aligning AI initiatives with real business outcomes and stakeholder needs