How inDrive Drives More Efficient Rider Acquisition at Scale
By the numbers
Overview
Company
Industry
Headquarters
Campaign Type
Snapshot
With over 400 million downloads across 48 countries, inDrive is the world's second-most downloaded ride-hailing app, with a mission to make mobility fair and accessible in underserved markets. Their unique bidding model lets riders and drivers negotiate fares, keeping rates affordable for both sides. Today, the company is building a super app with services expanding to grocery delivery, fintech, and freight.
Fueling that ambition demands efficiency. inDrive's model emphasizes cost efficiency for both riders and drivers, placing a premium on sustainable acquisition strategies. In other words, they needed to acquire riders who would come back week after week, not sporadically.Their existing tCPA campaigns focused on early conversion signals, which are well-suited for rapid optimization, while longer-term engagement required additional layers of insight.
The inDrive team knew that optimizing for predicted long-term value would solve this, but operationalizing that strategy at global scale would require significant long-term internal investment. Voyantis offered the predictive modeling, signal engineering, and infrastructure inDrive needed, without the build.
Impact at a glance
Validated predictive growth strategy across markets,
driving 50% lower average CPA for first time riders and retained riders alongside 84% higher average ROAS.
Deployed signals across 14 geographies
with region-specific delivery plans tailored to user behaviors, built to scale across inDrive's full 48-market footprint.
Built to evolve as the market and business shift
with programmatic model retraining.
The challenge
Google's bidding algorithms rely on early conversion events and values to optimize campaigns quickly and allocate ad spend effectively. The faster advertisers feed Google quality signals about their audience, the faster it learns and the less budget is wasted.
The challenge for inDrive was that the most meaningful insight - whether a rider would come back again and again in early months - didn't surface until weeks after acquisition.
To provide Google with that insight earlier, the team explored predictive modeling internally but quickly realized that a predictive model alone wasn’t enough. Turning those predictions into signals Google could learn from required infrastructure that would take months to build in-house and ongoing resources to maintain it.
The solve
inDrive needed to predict rider intent early, turn those predictions into signals Google could use, and scale across markets. Voyantis delivered all three.
Predict rider intent in hours, not weeks
inDrive identified early repeat usage as a key indicator of long-term rider value. Voyantis built a propensity model that generates predictions within an hour of a user's first ride, with updated scores at 6 hours, 12 hours, and daily through day seven.
Voyantis’s model draws on thousands of features, ranking each based on its conversion impact: early engagement patterns like completed orders and driver assignments, contextual signals like device type and timezone, and behavioral indicators like price sensitivity and activity breadth. The model weighs each feature relative to the others, capturing nuance that simpler approaches might miss.
Rank users so Google knows exactly how to bid
Under tCPA, every rider looks the same to Google - a single cost target, regardless of whether someone would ride once or weekly. That's a problem when your goal is retention, not just acquisition.
Voyantis's model scores users on a continuum, creating well-defined buckets from lowest to highest predicted intent. At 92% accuracy, those buckets give Google the clarity to bid aggressively on high-potential riders and pull back on the rest, so inDrive pays for users who'll become regular riders.
Operationalize predictions without the infrastructure lift
Google's algorithms learn from real conversion events, not probability scores. Predictions need specific formatting, precise timing, and constant recalibration. Every new market means new pipelines. Every new product means retraining. Building that in-house would have required a dedicated team.
Voyantis brought that infrastructure out of the box. The platform integrates predictive insights in ways aligned with Google’s optimization frameworks from day one. As inDrive's business evolves to new geographies and services, the system retrains automatically without adding to their engineering backlog.
Execute at global scale
A weekend rider in São Paulo behaves differently than a weekday commuter in Almaty. inDrive operates across 48 countries, each with its own user patterns, competitive dynamics, and price sensitivities.
Voyantis is built for complexity. Models retrain continuously, adapting to regional differences and evolving product lines without manual intervention.
Operate as an extension of the team
Voyantis operates as an extension of inDrive's global growth and user acquisition team, regularly aligning on strategy, reviewing performance, and planning what's next.
Behind the scenes, the technical infrastructure matches the hands-on support, supported by enterprise-grade reliability, monitoring, and operational safeguards. As inDrive's needs evolve, Voyantis delivers ongoing education, documentation, and a long-term delivery plan tied to the company's roadmap.
Voyantis’ impact
50%
lower CPA for frequent riders
84%
increase in avg ROAS (measured on net revenue)
92%
model accuracy
Efficiency across markets
Two geographic regions became the proving ground with head-to-head tests comparing Voyantis-powered tROAS campaigns against inDrive’s existing tCPA approach.
Country A
- 120% increase in ROAS (measured on net revenue)
- 65% decrease in CPA for riders completing 4+ trips in their first month
- 72% decrease in CPA for first-time riders
Country B
- 48% increase in ROAS (measured on net revenue)
- 36% decrease in CPA for riders completing 4+ trips in their first month
- 32% decrease in CPA for first-time riders
Embedded growth infrastructure
Voyantis signals are now live across 14 geographies, providing the predictive infrastructure to acquire high-value users at scale without pulling engineering off the product.
