The Suggestion System at Lyft Introduction: Scope of the Suggestion System What Does the Suggestion System Do? Upcoming Improvements Are You as Excited as We Are?

-

Suggestion plays a crucial role in Lyft’s understanding of its riders and allows for customizing app experiences to raised fulfill their needs. At times, recommendations are also leveraged to administer the marketplace, ensuring there’s a healthy balance between ride demand and driver supply. This enables ride requests to be fulfilled with more desirable dispatch outcomes equivalent to matching riders with the perfect driver nearby.

This blog post focuses on the scope and the goals of the suggestion system, and explores a few of the most up-to-date changes the Rider team has made to raised serve Lyft’s riders.

The suggestion system covers user experiences throughout the ride journey.

Screenshots are illustrative. May not capture the present experience.

At app open, riders can skip the whole request flow and request a ride with a faucet at a button. This is on the market for prime frequency users which have extensive travel history with Lyft. It is a convenient addition to the house screen and saves time and energy for users who’re in a rush.

After the user sets their destination, the app presents a ranked list of product offerings based on the user’s travel preferences and the present marketplace conditions. Some visual highlights are also exhibited to help make clear the tradeoffs across different options. As an example, Lyft highlights the “fastest” and “your usual” ride types to enable riders to make faster and easier decisions when evaluating their options.

Marketplace conditions are dynamic, and mode availability, prices and ETAs (i.e. estimated time of arrival) can change in a short time. In select sessions, these changes will be captured and made helpful to riders in our post request cross-sell experience, where an interstitial prompt is introduced detailing upgrade options with a greater ETA or price.

This blog mostly focuses on the mode selector to clarify how rankings have evolved up to now years and briefly touches on the post request cross-sells.

Years ago, when mode options were limited, the mode list was organized simply through rating by price or ETA. Moreover, in certain regions, the mode list was static as a way to observe rider behaviors as a baseline.

For the past two years, Lyft has invested in improving customization of mode recommendations. Since then, the suggestion system has proven to drive more desirable outcomes in comparison with a static system, by solving problems in three major areas:

  • The challenge of overchoice
  • The cold start problem for brand spanking new modes
  • Dynamic business and user experience goals

The challenge of overchoice

As shown within the mock below, these are a subset of offerings provided to users across Lyft available regions. It’s unusual to see all of them appearing at the identical time, but in larger markets like San Francisco or Recent York, users can easily find themselves in front of at the least 8–10 options across rideshare, bikes, scooters and rentals. As Lyft’s portfolio grows, a typical rider can have a tough time discovering and understanding the big variety of products that Lyft has to supply, which can end in riders by chance booking the fallacious mode.

Screenshots are illustrative. May not capture the present experience.

To resolve this, the suggestion system leverages a set of machine learning models to predict a rider’s propensity of converting into each mode and customizes the rankings based on it. Wealthy information has been considered in constructing these models, including temporal features like location and time info, supply / demand signals, ride histories and user preferences.

  • Algorithm: LightGBM (each mode is taken into account as a definite class, with weights determined by analyzing mode specific financial metrics)
  • Model objective: lambda rank or multi-class classification, depending on different use cases
  • Hyperparameters: plenty of tuning on the everyday hyperparameters like maximum depth and learning rate is required to realize desirable results. Specifically, Lyft’s in-house distributed hyperparameter optimization pipeline is used for the vast majority of its business critical models.
Screenshots are illustrative. May not capture the present experience.

The modes above the fold and the preselection are powered by the set of model results. When users initially land on the mode selector, the variety of modes shown is proscribed to three–4 above the fold (see mock above). That is to scale back cognitive overload for users, and have them deal with modes that may best represent their transportation needs. In fact, users would still have the flexibility to swipe up and explore other options below the fold, that are ranked based on categories and pricing.

Along with rating, preselection helps reduce steps in our ride request flow. In 2019, the user’s last mode taken was preselected. This captured user stickiness for some use cases, but introduced natural exploration bias. That said, in 2020, Lyft moved towards a more user centric approach — preselecting a user’s most ceaselessly used mode. This was more robust to outliers but could possibly be jumpy, especially for infrequent users. In 2022, Lyft aimed to resolve this by moving towards a model-based approach, preselecting the mode with the very best predicted propensity rating. This solution is by design real-time, dynamic and has proven to be more accurate and effective.

The cold start problem for brand spanking new modes

Rating of the choices highly impacts a rider’s mode decisions. Due to this fact, recent offerings with limited user touch points should still be really useful properly to realize awareness and habituation. When Wait & Save and Shared products were first introduced, they got very limited visibility above the fold since no or limited data was at our disposal to tell rankings. Within the suggestion system today, a post-processor was introduced as an extra layer to regulate the machine learning model results, mitigating the natural bias created by the shortage of sufficient training data. This manner, Lyft’s suggestion system can sufficiently and effectively make the strategic decisions to advertise certain options without making substantial system changes.

Dynamic business and user experience goals

Rides and conversions — reflected within the increasing accuracy rates in model predictions — usually are not at all times the perfect goals to optimize for within the suggestion system. To create a seamless request experience and ensure a more balanced marketplace, other metrics like ride reliability and provide / demand balance have to be actively monitored. This will be achieved in a myriad of how. One example is the selective cross-sell experience following a ride request. The triggering of this experience just isn’t quite common, but will be an efficient lever for elevating the matching efficiency and smoothing demand. Nonetheless, this doesn’t mean that the user experience becomes secondary — rider propensity, price and wait time trade-offs remain critical guardrails for determining eligibility for these mode cross-sells.

In summary, the suggestion system uses a hybrid approach towards personalizing rankings and mode suggestions. Machine learning propensity models serve about 90% of the use cases and other layers of adjustments are applied to optimize for added business and user goals.

Reimagine the request flow

In late 2022, Lyft introduced the “one-tap” modules, a seamless experience that highlights the perfect options (each destinations and modes) for riders upon opening the app, speeding up the request experience. The coverage of “one-tap” shall be expanded to incorporate more relevant use cases and make it easily accessible to a wide selection of users.

Introduce reinforcement learning

Lyft’s in-house contextual bandit system will enable more dynamic user-system interactions to be considered and react in a more real-time fashion (go to this blog to learn more about its first application in pricing). Long run user engagement behaviors will be taken under consideration too to resolve any long run effects uncaptured by the present manual training of models.

Enthusiastic about applying science at scale and learning more about how data-driven decisions are made at Lyft? Try more exciting blogs in Lyft Data Science. Should you find this blog helpful and wish to debate more, please don’t hesitate to succeed in out via email.

ASK DUKE

What are your thoughts on this topic?
Let us know in the comments below.

1 COMMENT

0 0 votes
Article Rating
guest
1 Comment
Oldest
Newest Most Voted
Inline Feedbacks
View all comments

Share this article

Recent posts

1
0
Would love your thoughts, please comment.x
()
x