Personalization at Lyft | Ride Hiding Experience Ultra Pro

During the weekend, I reviewed various engineering blogs and videos that highlighted the onboarding and checkout processes of Lyft . This experience led me to the realization that the ride-hailing industry in India is not as mature as Lyft's, which provides its users with an exceptional and tailored experience. While inDrive is making strides in understanding Indian users' needs, Uber and Ola have a long way to go.

In my opinion, the ride-hailing industry presents a particularly fascinating field for data science. Although I was only able to work on it for a few months as a contractor, I found the experience to be incredibly engaging. If you're interested in learning more about this topic, feel free to check out the first edition of my newsletter.

Personalisation Scope in Ride-Hailing

In a previous Ride Hailing Newsletter, I covered various use cases of operations data science, including the Driver-Rider Matching Algorithm, Dynamic Pricing, and Demand Forecasting. However, in this edition, I will delve into the techniques used for personalization, ease of checkout, and upselling/cross-selling that are aimed at enhancing the user experience on a platform such as Lyft .

OneTap Ride and Pricing Based on Time to Arrive: For frequent travelers, or "super users," Lyft has a one-tap module for whom they can predict their route, giving a faster checkout experience. And for those who value time, Lyft ranks options based on the user type. Students, for example, prefer cheaper options and don't mind waiting a few extra minutes. Plus, Lyft remembers your previous option selections based on route and time for a quick checkout experience.

Visual Highlighting: The real kicker is visual highlighting. Even if the price and cab are the same, Lyft shows them in different ways just to make you feel special. I mean, who doesn't want to feel special, right? It would be interesting to analyze how user behavior changes with different visual highlights. Great strategy.

Upselling: And if that's not enough, after you checkout, Lyft offers upsells for a faster or higher-segment ride, an Upgrade option with a timer that gives you a sense of urgency. You might hit that upgrade button when you see that original fare plus a bit more messaging.

Cross-selling: If someone booked a bike ride in the next booking, give a discount on a cab ride, or if you have something like Uber Eats, give a food coupon.

To gain a clear understanding of the differences between Upsell and Cross-sell, refer to the accompanying image. This will help individuals who are unfamiliar with these concepts to grasp them more easily.

How to Achieve the Above with Data Science?

1. OneTap Ride: Predictive model on user behavior, location, and time of booking a cab ride, we can accurately predict routine booking of a frequent traveler and give them OneTap booking experience to Home or Work.
2. Price based on Time to Arrive: This is very interesting and can be very supportive to demand forecasting models. Lyft is playing around with Dynamic pricing with transparency on demand and rider availability in a particular area. We need accurate ETA forecasting models for this to work.
3. The ranking of entities can be customized to user preferences by preselecting entities and ranking them based on the user's level or cohort level. This approach provides a more personalized experience. For example, a student cohort may prioritize the cheapest ride, while businessmen may prefer luxury cabs. To effectively model the ranking of entities, it is crucial to consider user segmentation, past selection behavior, geographic features, etc.
4. Visual Highlight is a classical problem: One can solve this using context Multi-Arm Bandits giving a more personalized feel to the user.
5. Upsell and Cross-sell are not something special as it is well solved in the e-commerce domain. It is part of a selective model with pricing & margin play.

Ranking of Options at Lyft

Lyft has so many options, and entity ranking point 3 discussed above is important for a faster, more personalized checkout experience for the user.

AIM: Develop a propensity model to predict a rider's likelihood of converting to each mode and customize rankings accordingly. The models take into account rich information, including temporal features like location and time info, supply/demand signals, ride histories, and user preferences

In the background, to achieve this, Lyft uses:

• Algorithm: LightGBM (each mode is considered as a distinct class, with weights determined by analyzing mode-specific financial metrics)
• Model objective: lambda rank or multi-class classification, dependent on different use cases
• Hyperparameters: a lot of tuning on the typical hyperparameters like maximum depth and the learning rate is required to achieve desirable results.

Although it may appear to be a straightforward problem, recommending and ranking items has become increasingly difficult for businesses. This is because they now require both user personalization and higher monetization metrics. To address this challenge, many platforms have begun utilizing advertising personalization techniques. These techniques involve ranking entities based on a combination of the user-preference score and the monetization metric. By leveraging this approach, businesses can provide users with personalized recommendations while also maximizing their revenue.

As businesses increasingly prioritize profitability, data scientists are tasked with finding solutions that balance monetization metrics and user preferences.

There is a lot more to this problem. Ranking and Relevance problems are always fun to design and think about. I would love to hear your thoughts on this problem and share your views in the comments.

Reference

The Recommendation System at Lyft by Jinshu Niu

I'm honing my superhero problem-solving skills, and if you're in need of a data scientist to save the day with some advisory or consultancy work, you know where to find me! Shoot me an email at [email protected], and let's make some magic happen.

HIT THE????ICON ON MY PROFILE TO GET MY DATA SCIENCE UPDATES AND NEWSLETTER NOTIFICATIONS

Subscribe to get Email Notifications of every Newsletter being Published:?HERE