Building AI? This Booking.com Leader Says Start Small and Solve One Problem Well

You started at Booking.com as a machine learning specialist focusing on Trust & Safety, and now you lead a team that combines machine learning (ML) and software engineering. Looking back, what’s one lesson from your hands-on work that still shapes how you lead today?

When I look back at my journey, if I had to choose one core lesson that defines my transition into management, it is the importance of Ownership in Navigating Ambiguity.

As an ML Scientist, you inherently live with uncertainty; you never know if a model will hit its target KPIs, and timelines are inherently less predictable than in traditional software engineering. Learning to navigate this constant ambiguity taught me a vital skill: to own the process of discovery and risk from end-to-end.

As a manager this translates into creating a culture of psychological safety and resilient planning. I lead by empowering my teams to own the process of solving the unknown, rather than fearing the lack of a guaranteed outcome. I prioritize transparent communication of risk and probability over fixed certainty, ensuring we make deliberate decisions while keeping stakeholders aligned, even when the project path is evolving. This skill of managing uncertainty is perhaps the most important segue from the scientific world to the managerial one.

Managing teams that work on both cutting-edge AI and critical safety features must be complex. How do you encourage your team of engineers to innovate while keeping trust and reliability at the core of everything you build?

Balancing innovation with trust and reliability is indeed one of the biggest challenges in our industry, and it’s something I constantly keep in mind as a manager. Our primary goal is always to help in protecting our customers and do the right thing, not just chase the most technically exciting ideas. Yet, staying effective means we must continuously evolve.

To encourage innovation while maintaining trust, we practice what I call “risk-gated innovation.” We clearly separate experimentation from production. For example, we run frequent internal hackathons as safe sandboxes for learning and creativity. These ideas become Proofs of Concept that undergo rigorous risk assessment and validation before any integration into our live, safety-critical systems. This approach allows us to innovate boldly without compromising reliability.

Reliability itself is non-negotiable. Every innovative ML solution must rest on a foundation of robust software engineering. We measure success not only by model accuracy but by the stability, observability, and maintainability of the entire production pipeline. This ensures our safety-critical services are built to last.

Finally, I’m a strong advocate for sharing quick wins and knowledge. We hold regular knowledge-sharing sessions and maintain active Slack channels to spread new insights, from conference learnings to blog discoveries. This culture of continuous learning helps the team solve challenges faster and inspires fresh ideas.

By fostering this mindset, we ensure that innovation directly supports our core mission: make it easier for everyone to experience the world safely.

Your keynote at the 8th Regional E-commerce Conference, “How We Keep Our Customers Safe Using AI,” explores how Booking.com safeguards not just online but also real-world experiences. What inspired this focus, and what key message do you hope the audience will take away?

The inspiration for this focus comes directly from Booking.com’s mission to enable everyone to experience the world safely. Working in Trust & Safety, this mission guides everything we do. As a manager of ML and Engineering teams, I’m privileged to see how AI and technology drive this mission forward at scale.

Interestingly, many people assume safety means just cyber concerns like fraud detection or secure payments. While those are crucial, our Trust & Safety efforts go even beyond that. We focus on helping to protect the physical and psychological well-being of travelers and partners throughout their entire journey, from booking to check-out.

The key message I want the audience to take away is two-fold. First, businesses must move beyond merely meeting regulatory compliance or ticking security checkboxes. We need to shift from “what we have to do” toward “what we can do” to truly elevate customer experience. Second, new technologies and AI are the essential enablers of this shift. It bridges the gap between our online platform and the real world, allowing us to predict, prevent, and respond to complex human risks at scale.

The travel industry is constantly evolving, with new digital risks and expectations from customers. How do you see AI reshaping the future of online travel and e-commerce in the next few years?

As an AI enthusiast, I see AI transforming the e-commerce industry by making customer experiences more personalised, efficient, and seamless. But as we build these systems, it’s important to stay focused on what really matters: solving real problems for people, and doing this safely, responsibly, and at scale.

One area that’s already changing fast is search and discovery. I’ve realised that more and more people including myself now prefer talking to LLMs rather than Googling something. Whether it’s planning a trip, looking for a place to eat, or comparing which robot cleaner to buy, we want answers that are tailored, conversational, and actually helpful. AI is shifting discovery from basic filtering to intelligent, proactive experiences that guide users through their entire journey, not just one transaction. An exciting recent example of this shift is that Booking.com became one of the first partner apps launched in ChatGPT. This makes it even easier for travelers to explore our wide range of accommodations using conversational prompts (like “Booking.com, find me hotels in Miami beach for Nov 3rd–5th with a pool”), with ChatGPT app generating relevant options and seamless access. (If you’re curious to read more about this feature, you can check the full update here.)

Then there’s operational resilience, another huge area of impact. In the industry, AI is taking over complex, repetitive tasks like customer service triage, dynamic pricing, or compliance checks. That means faster platforms, smoother operations, and less manual effort, freeing up teams to focus on high-value work that requires creativity and human judgment.

I genuinely believe AI can boost businesses, simplify lives, and unlock totally new experiences, as long as we use it responsibly. The teams that will lead in this space are the ones that move fast, but also stay grounded in ethics, customer trust, and doing the right thing.

Machine learning is a key part of AI, which is central to your work as both a specialist and an engineering manager. Are there any AI tools, frameworks, or technologies you rely on most — personally or with your team — that other e-commerce professionals should know about?

Needless to say that today we all rely on AI applications which are becoming a more and more crucial part of our daily work, both as engineers and leaders.

Beyond the now widespread reliance on open source tools, in my view, currently the biggest challenge and focus for engineers and leaders lies in MLOps and making models reliably deploy, scale, and stay healthy. In our team in Booking.com, we tackle this by leveraging robust MLOps platforms such as AWS SageMaker that standardise the pipeline from feature creation to monitoring. This approach is critical for e-commerce because it eliminates the risk of model drift causing revenue loss and ensures that reliable models can be deployed with one-click functionality, proving far more crucial than initial model accuracy.

Many startups and companies in the Western Balkans are starting to experiment with AI. Based on your experience at a global platform, what would be your top advice for teams here looking to build AI solutions responsibly and effectively — without losing sight of the customer experience?

In my view, building AI solutions should always go hand in hand with customer centricity. The most successful AI applications I’ve seen or been part of start with a clear understanding of real user needs, not just the latest technology trend.

For companies in the Western Balkans, I’d recommend starting small with focused use cases that genuinely improve the customer journey. Don’t aim to “AI everything” from day one, instead identify a specific friction point you can solve well. It’s also important to bake responsibility into the process early on by considering bias, fairness, and data privacy as integral parts of your development strategy, not as an afterthought.

Having an interactive approach is crucial: continuously collect user feedback and keep humans in the loop during the early stages until you excel, especially for decisions that directly impact customers or are critical in nature. This helps maintain trust and ensures ongoing oversight. Success should be measured not just by technical accuracy, but by how helpful and reliable the solution is for users.

The Western Balkans has a unique opportunity to build AI solutions with fresh perspectives and fewer legacy constraints. If teams stay grounded in ethics and customer value, there’s a real chance to lead with smart, responsible innovation, which is truly exciting.