Role:
Product Coach, Product Strategist, & Product Designer
Date:
Feb 2019
Problem
A traditional airline company wanted to pivot towards a digital-first identity. Their vision was to become a robust digital platform that operated flights, rather than just an airline. This transition involved acquiring tech startups and launching internal tech initiatives aimed at solving travel problems. However, an unintended consequence emerged: the internal initiatives and the acquired startups began to compete, leading to inefficiencies, duplication of efforts, and a fragmented user experience. This posed a core challenge of harmonizing these efforts while ensuring scalability and effectiveness, particularly in addressing complex user decision-making processes within digital environments, such as booking, recommendations, and personalization.
How to unify internal initiatives & external partnerships to deliver an intuitive superapp experience?
Process
The approach to solving this problem involved analyzing decision-making behavior and translating it into actionable product strategies. Key aspects included:
Understanding Decision Frameworks: The study outlined three factors critical to intelligent decision-making: available information, time to decide, and the anticipated reward. Each recommendation system needed to balance these elements to provide the most relevant options.
Exploring Filtering Models:
Popularity Filtering: Basic model relying on high-frequency items but lacked personalization.
Collaborative Filtering: User-centric recommendations based on similarities but struggled with scalability in large datasets.
Content Filtering: Item-based similarities that ensured stability but needed a robust similarity framework.
Association Rule Mining: Focused on patterns in user sessions rather than just items, offering session-based relevance.
Hybrid Models: Combined techniques for layered recommendations, addressing the weaknesses of individual models.
Incorporating Machine Learning: Machine learning (ML) and data mining techniques were central to handling complex datasets, filtering noise, and enabling systems to predict user needs in real-time. The process also involved implicit (behavioral) and explicit (user-provided) data capture mechanisms to refine the personalization process.
Leveraging Reinforcement Learning: By integrating reinforcement learning principles, the recommendation engine could dynamically learn from user interactions, optimize its reward functions (e.g., purchases, shares), and adapt over time to achieve higher accuracies and personalized results.
Product
The designed recommendation engine employed multiple iterative layers to ensure precision and user engagement:
Data Capture & Interaction: The system collected implicit data (e.g., location, calendar) and explicit inputs (user preferences captured interactively). Gamification elements were introduced to make data collection engaging, like interactive bubbles for capturing user interests.
Recommendation Layers:
First Layer (Reco.1): Initial recommendations based on basic user inputs and implicit data. The initial user inputs helped solve the cold-start data problem. This in turn helped initial product adoption as well.
Second Layer (Reco.2): Dynamic recommendations using collaborative and content filtering models, analyzing similar users and items.
Third Layer (Reco.3): Advanced recommendations derived from ML models analyzing historical and real-time data, generating deeper insights and actionable suggestions.
Product
Real-Time Adaptation: The system incorporated real-time analysis, moving beyond reactive to proactive recommendations. It leveraged emotional insights, conversational interfaces, and virtual agents to simulate personalized, context-aware recommendations.
User-Centric UI/UX: Screens showcased onboarding processes, personalized suggestions, and seamless navigation between implicit and explicit inputs. Examples included travel suggestions based on user schedules and tailored offers influenced by browsing history.
Inclusion of Serendipity: By introducing a "surprise element," the engine encouraged serendipitous discovery, a critical feature for building trust and delight in users.
Result
The intelligent recommendation engine had the potential to significantly enhance user experiences while driving business outcomes. By guiding users through decision stages—awareness, research, and decision-making—the engine streamlined the journey, reducing abandonment rates and increasing conversion rates. Key performance metrics included:
Conversion Rate Improvement: Anticipated to increase by up to 35% by offering relevant, timely suggestions.
Customer Satisfaction Scores (CSAT): Higher scores through trust-building and personalization.
Revenue Growth: Proven models like Amazon's recommendation engine show that similar systems can contribute to over 30% of revenue.
User Engagement: Enhanced engagement via gamification and real-time responses, reducing churn and increasing app retention rates.
Reflection
Reflecting on past projects, I see two perspectives: the successes that make me happy and the insights I've gained now that could have enhanced my approach in the past.
Some things that I'm happy about:
This is one of the most iconic manufacturing juggernaut brand in the world. I'm grateful that I got to work on building data platform strategy and long-term product vision. Personally, this project underscored the importance of persona development, human-centered design and iterative development with quick feedback. Post-launch, integrating advanced analytics like prescriptive analytics could further enhance decision-making capabilities for users.
LLM Disclaimer: In 2018—the pre-LLM era—"chatbot" and "conversational UI" were the dominant terms used to describe human-machine natural language interactions. I've used "LLM" in the title since the core principles of interface design remain largely unchanged. Whether we're invoking a bot, LLM, or AI agent during a task, the fundamental interaction patterns stay consistent.
Creative Confidentiality: In the spirit of professional discretion and digital camouflage, some client identities have been subtly transformed into their alter-ego personas. This ninja-like name-swapping applies exclusively to projects completed as an external design mercenary. For all other showcased works, brand and product names remain true to their original, registered identities.
Authenticity Stamp: Every pixel, wireframe, approach, and design concept you'll discover here is 100% crafted by the hands (and occasionally bulletproof caffeinated brain) of Naren Katakam. No design outsourcing, no smoke and mirrors—just pure, unadulterated creative craftsmanship.