What we do on the data science team

by Aviel Makmal

Data science stands at the heart of Agoda’s operational strategy, transforming complex data into actionable insights. It’s a key player in our pursuit of technological advancement and business efficiency. In this blog, our data science manager, Aviel Makmal, shares insights into his role as a data science manager.

Roles and Responsibilities

My journey at Agoda began five years ago as a data scientist. With time, my role evolved into leading the data science fraud teams. Today, my responsibilities involve managing the anomaly detection team, customer-facing LLM, internal LLM initiatives, and spearheading pilot projects in areas like Marketing and payment optimization.

I consider my role as a data science manager to be my true calling as it aligns perfectly with my interests — being data-driven, solving complex problems, ideation, and communication. Seeing the real-world impact of our work in the company is rewarding.

My days are dynamic, filled with strategy meetings, interviews (we are hiring!), and aligning with my team to ensure our goals are met. I also dedicate time to hands-on tasks, from data analysis to prototyping models for upcoming projects.

One of our key projects was developing an internal RAG (Retrieval Augmented Generation) Chatbot. This tool has been a game-changer in providing technical support across Agoda, allowing our IT teams to focus on more complex challenges while delivering prompt responses to their issues. Though the project is internal facing, the impact is quite significant.

Navigating Challenges

With the introduction of GenAI, LLM, and RAG, we ventured into relatively uncharted territories. Our challenge was to figure out the best way to scale in the right direction, measure the quality of LLM-generated answers, and how to improve the overall accuracy of the bot. As there wasn’t any streamlined gold standard approach yet, we had to iterate and go back to the drawing board quite often till we found the right way. Since then, we have scaled much faster, understood our pitfalls, and iterated between different solutions.

It’s important to remain flexible when initiatives don’t go as planned. Sometimes, you can fall in love with your idea and your way of thinking, but the reality/data shows different results. It is good to let go of ideas if the data doesn’t support them and consult with others.

Tools and Technologies

My day-to-day stack is Python and pyspark/SQL and their corresponding libraries. However, the introduction of Co-Pilot and ChatGPT has been transformative, helping me brainstorm ideas and write code more efficiently. These tools have been helpful when working on different projects as they enhance productivity. I encourage every data scientist or developer to use these technologies daily.

Collaboration and Team Dynamics

Collaboration is ingrained in Agoda’s culture. Our scrum teams, including product owners, developers, and data scientists, co-own the products we work on. This collaborative effort ensures we align our objectives, understand the trade-offs, and collectively drive our projects to success.

We also collaborate with our data legal team to ensure that we uphold the highest standards of data ethics and privacy. Their proactive and responsive approach enables us to use data responsibly while advancing our data science initiatives.

In data science, research is key, but it can sometimes be a lengthy process with no real learning or outcomes. We keep Agoda’s value of moving fast and experimenting while still being scientific about it in mind. We start with simple baselines or noise of our data (running the randomized experiments) to see if our baseline is really impactful. From there, we iterate to a fully optimized solution.

Professional Growth

Managing a diverse team requires recognizing each individual’s unique strengths and identifying areas for improvement. Regular one-on-ones and performance reviews are part of this process. My professional growth over the past year, guided by my manager, reflects this commitment to continuous development.

Also, building the fraud DS team remains a highlight of my career at Agoda. Fraud has different aspects, from payment fraud (stolen CC, account takeover, new product initiatives, Property onboardings), and each requires a unique solution. We took on a challenging problem and provided innovative solutions that significantly contributed to the company’s success. We even scaled to different parallel horizons, such as marketing abuse and payment optimizations.

Career Advice for Data Scientists

To be a successful data scientist, you must be scientific and business-oriented. This means understanding scientifically what you are trying to achieve and if your method is impactful. Being business-oriented means understanding the problem you are trying to solve and their main KPI; your stakeholders care less about precision/recall but more about the true outcome and the tradeoffs.