‍

Amir Vaheb describes how personalization projects drive continuous learning, and why transparency and relevance matter more than perfect models.

Who are you and what do you do?

I’m Amir Vaheb, a Senior AI Scientist at Data Design. My work focuses on understanding how users interact with products and transforming those insights into business value. I help organizations make better, data-driven decisions by designing, building, and validating AI and machine learning solutions that deliver real impact. Every project starts by clarifying the business problem and translating it into the right data questions, always keeping relevance, scalability, and measurable outcomes at the center.

‍

Which project has stayed with you in particular, and why?

One project that has stayed with me was a personalization initiative for a digital service platform. It was memorable because it was the first time I took a machine-learning-based solution all the way into production. Seeing the real-world impact of our work, improved engagement and tangible business results, was very rewarding. It was also a true team effort that required close collaboration between data, product, and business teams.

‍

What was the project about?

The client operated a large video-on-demand platform with thousands of titles, but user engagement was low because everyone saw the same generic recommendations. Users disengaged when shown content that didn't match their interests or timing: valuable content simply wasn't being discovered.

The challenge was to guide users toward the right content in a relevant, scalable way without increasing operational complexity or relying on manual rules. The logic is the same as Netflix: better recommendations engage users longer, and more watch time translates directly to revenue.

‍

What kind of solution did you develop?

We developed a flexible recommendation system that combined multiple behavioral and contextual signals, user behavior patterns and available content characteristics. The system was designed to work well for both existing users and new visitors, as well as newly added content.

Instead of relying on one complex model, we integrated several complementary algorithms from simpler heuristics to more advanced models and controlled their interaction carefully. This hybrid approach made the system easier to explain, maintain, and continuously improve.

For results, we tracked both offline metrics like model quality scores and online metrics like engagement. We achieved a 20% increase in watch time per user and a 50% improvement in click-through rate.

‍

How was the solution received?

The solution was very well received. It integrated smoothly into existing workflows, so teams didn’t have to change how they worked overnight. Product and business stakeholders appreciated that they could understand why recommendations were made and how they supported business goals. This transparency built trust early on, allowing for gradual rollout, feedback collection, and continuous refinement. Over time, the system evolved from a data science experiment into a trusted part of daily decision-making.

‍

What challenges or key insights emerged?

There were several challenges, both technical and organizational. Since this was the company’s first data-science-driven initiative, we had to build new capabilities while delivering results. One major insight was that the “perfect” recommendation system doesn’t exist, the key is balancing relevance, business goals, and user trust.

We learned that focusing on transparency and stakeholder involvement mattered more than squeezing out the last bit of model accuracy. Gradual rollouts and early communication built far more confidence than technical optimization alone. Another important realization was the need to translate data insights into a language that product and marketing teams understood connecting models to business value.

‍

Could the same solution be applied elsewhere?

Absolutely. The personalization principles we used have since been applied in other industries. For example, I later used similar methods in a mobile gaming project, where recommendations guided players toward relevant in-game content and offers. The same ideas also work in e-commerce, financial applications, and beyond. Personalization helps organizations in almost any industry increase engagement, improve satisfaction, and drive measurable outcomes.

‍

What was the best part of the project?

The most rewarding part was seeing personalization become part of the product mindset,  not just a data initiative. Discussions shifted from “how accurate is the model?” to “is this relevant for the user right now?”

That cultural shift toward data-driven, user-centric decision-making was the real success. It influenced how teams across the organization thought about product development, user experience, and long-term strategy. When the first version was live and I saw people in meetings naturally discussing metrics and user data, that was genuinely rewarding.

‍