The Story of Building Lyft's Marketing Automation Platform

Lyft's Mission and Growth Background

Lyft's mission is to "improve people's lives with the world's best transportation," providing over 50 million carbon-neutral rides per month. But all of this is just the beginning. Lyft's growth stems from improving the user acquisition process, raising brand awareness through region-specific ad campaigns and encouraging consideration of various transportation options to attract new users. However, coordinating these campaigns at scale was time-consuming, leading Lyft to take on the challenge of automation.

The Importance of Growth Acquisition

Lyft's user acquisition is led by a data-driven cross-functional team, focused on scale, measurability, and predictability. User acquisition sits at the top of the onboarding funnel and occurs through various channels (search, display ads, social media, etc.). Since each channel requires different techniques and strategies, Lyft identifies and executes the optimal approach for each.

"Our goal is to make Lyft the best choice for consumers."

The Need for Automation

Lyft had to make thousands of decisions daily. For example:

• Setting ad bid prices and budgets
• Disabling inefficient ad content
• Changing referral rewards by market
• Identifying high-value user segments
• Sharing strategies across campaigns

All of these tasks placed a huge burden on marketers and could lead to inefficient decisions. "Let's reduce repetitive work through automation so marketers can focus on innovation and experimentation!" With this goal, Lyft developed Symphony, its marketing automation platform.

Symphony: Lyft's Marketing Automation Platform

Symphony is a system that predicts user value based on business objectives, allocates budgets, and acquires new users accordingly. Symphony's architecture consists of three main components:

1. Lifetime Value (LTV) Predictor
   • Predicts users' potential value to measure each channel's efficiency and optimize budgets.
2. Budget Allocator
   • Distributes budgets based on marketing performance data, using Thompson Sampling to achieve optimal results.
3. Bidding System (Bidders)
   • Sets ad bid prices and deploys them to internal and external channels.

LTV Predictor: Understanding User Value

LTV (Lifetime Value) represents the long-term value a user can bring to Lyft. Initially, it's difficult to know a user's retention rate, usage frequency, and transaction value precisely, so Lyft predicts LTV based on historical data. As users engage more with the service, prediction accuracy gradually improves.

"Accurately predicting a user's potential value is essential for setting medium- to long-term strategic goals."

Budget Allocator: Optimal Budget Distribution

The budget allocator uses LTV prediction data to assign appropriate budgets to each campaign. Lyft uses Markov Chain Monte Carlo (Thompson Sampling) to optimize budget distribution. This technique introduces a degree of randomness to explore possibilities not previously considered and help reach the global optimum.

"A moderate amount of exploration leads to optimal results in the long run."

Bidding System: Executing Ad Strategy

The bidding system executes all the changes needed to display ads at appropriate prices.

• Tuners: Set ad bidding strategies and allocate capital considering channel-specific characteristics.
• Actors: Deliver actual bid prices to internal and external channels.

Lyft uses customized bidding strategies that reflect each channel's characteristics, incorporating recency weighting and seasonality to account for market volatility.

Conclusion: Human-Machine Collaboration

Lyft uses a "Human-in-the-loop" approach to integrate human feedback into machine learning systems. While automation reduces repetitive tasks, marketers can focus on high-value activities such as:

• Gaining deeper understanding of users and their interests
• Designing new ad formats, messages, and channels
• Formulating hypotheses for large-scale experiments

"Automation gives marketers more time and energy. Now they can focus on creativity and strategy."

Future Development of Symphony

Lyft plans to continuously improve Symphony with features such as:

• Always-on experimentation
• Integration of seasonal factors like weather and time of day
• Better market context delivery
• Intelligent segmentation and personalization

"Thanks to Symphony, we're getting better returns on investment and saving marketers' time."

Lyft is still in the early stages of marketing automation, but looks forward to a bright future based on machine learning and experimentation.

Lyft Is Hiring!

Lyft is looking for engineers, data engineers, scientists, product managers, and marketers to join the growth team. Check out life at Lyft and job openings at Lyft Careers.