Real-Time Cab Fare Comparison Engine

The Auto-Rickshaw Insight That Started Everything

I was on my way to a client meeting in Bangalore, stuck in one of those peculiarly still traffic jams where nobody is moving and nobody seems bothered by it. The auto driver and I got talking — as you do.

He told me he'd stopped taking Ola bookings in the evenings because "the surge pricing is too unpredictable — passengers get angry when the app shows one price and they're charged another." He'd switched to a mix of Rapido and street hails. He didn't know — and had no way to know — that across town, a completely different surge situation was unfolding on Uber.

That conversation stuck. Here was someone whose income depended entirely on platform-specific surge dynamics — and he had no visibility into the comparative landscape. The same invisibility problem existed for the 2 million daily cab users in Tier 1 Indian cities.

That insight became the premise for a cab fare comparison engine: not just which app is cheapest right now, but a real-time, route-specific view across platforms — with surge prediction baked in.

The Market Gap

What Users Were Doing Without a Comparison Tool

Before building, we surveyed 400 frequent cab users in Bangalore, Mumbai, and Delhi. The current "comparison" behavior was manual and painful.

Behavior	% of Users	Time Spent	Success Rate
Open 2–3 apps manually and compare	68%	3–7 minutes	Moderate (surge changes during this time)
Stick to one platform always	21%	0 minutes	Low (missing better deals)
Ask colleague/friend what they're using	8%	1–2 minutes	Low
Use comparison tool (existing options)	3%	—	Poor (none were real-time)

The existing 3% using comparison tools were using browser extensions or third-party apps that scraped static data — not real-time pricing. A user who opened a comparison at 7:55pm was seeing data from 7:40pm, by which time surge multipliers had often changed significantly.

The comparison problem wasn't information access — it was information velocity. Cab fares change every 2–4 minutes during peak hours. Any comparison tool that wasn't updating at that frequency was creating false confidence in stale data.

Who Needed This Most

Five Types of Cab Users

User Type	% of Market	Price Sensitivity	Key Need	Comparison Value
Daily Commuter	34%	High	Predictability + lowest morning fare	Highest — repeat value
Weekend Leisure	22%	Medium	Convenience over price	Medium
Corporate Traveler	18%	Low (expensed)	Reliability + receipt	Low
Late Night User	15%	High	Safety + availability	High — surge sensitivity
Airport Shuttle	11%	Medium-High	Fixed-price predictability	High — single high-value trip

Daily Commuters (34% of users) were the core segment — high price sensitivity, repeat use, predictable routes. If the tool saved a commuter ₹40/day on a daily commute, that was ₹880/month in real savings. That's the value proposition that builds daily habits.

The Core Insight

Comparison Isn't Enough — Prediction Is the Differentiator

Every competitor showed current fares. That was necessary but not sufficient. The higher-value feature — and the technically harder one — was surge prediction: telling users not just what prices are now, but whether waiting 8 minutes would drop the surge multiplier.

The best comparison tool in the world fails if the comparison it shows becomes wrong before the user acts on it. Real-time was table stakes. Prediction was the moat.

Surge patterns are predictable with enough historical data. An 8:15am Monday ride from Koramangala to MG Road follows a consistent surge curve. We had the data. The question was whether users would trust and act on a prediction.

What We Built

The Engine Design

Real-Time Fare Fetching

Platform API integrations + scraping layer for platforms without public APIs. Fare refresh every 90 seconds. Route-specific (not city-wide) pricing — because surge varies by pickup zone within the same city at the same time.

Surge Prediction Model

LSTM model trained on 14 months of route-specific surge data. Features: time of day, day of week, weather (rain increases surge 40–80%), local events, historical patterns for that specific origin-destination pair. Output: 15-minute surge forecast.

Wait vs Book Recommendation

'Wait 8 minutes — surge likely to drop from 1.8× to 1.2×, saving ~₹65' or 'Book now — surge rising over next 15 minutes.' Probabilistic recommendation with confidence level shown. Users see the bet they're making.

Commute Route Memory

Repeat commuters: save up to 3 frequent routes. One-tap comparison for saved route. Daily commute digest: 'Your 8:30am Koramangala → MG Road: best fare right now is Rapido at ₹128.' Habit-forming format.

Feature	Adoption Rate	Impact on Return Rate
Saved Commute Routes	61% of daily users	+3.2× D30 retention vs non-savers
Surge Prediction (used)	44% of sessions	+28% conversion to app open
Wait/Book Recommendation	38% followed it	88% found it accurate per exit survey
Real-Time Refresh	Base feature	Foundation of trust metrics

Technical Architecture

The Data Layer That Made It Work

Data Ingestion

→Ola/Uber/Rapido APIs polled every 90 seconds per active route query
→Route-level pricing stored: city + zone + time + fare + surge multiplier
→Historical surge data: 14 months, 6 cities, 840 route pairs

Prediction Pipeline

→LSTM model: sequence input (last 6 hours of surge for that route) → 15-minute surge forecast
→Weather API integration: rain detection adds 45–80% to surge probability
→Model accuracy: 78% within ±10% of actual surge at 15-minute horizon

User Experience Layer

→Saved routes: one-tap comparison on app open
→Push notification: 'Your commute route looks good right now — low surge until 8:45am'
→Fare alert: notify when fare crosses user-set threshold for specific route

Traction

Growth Without Paid Acquisition

Month 1 · Launch

Bangalore + Delhi Only

→ D7 retention: 41% (far above typical fintech; habit forming quickly)

✓Launched with 6 cities' route data; Bangalore and Delhi prioritized
✓No paid acquisition — organic only (word of mouth, LinkedIn post, tech communities)
✓First 2,000 users within 10 days

→ D7 retention: 41% (far above typical fintech; habit forming quickly)

Month 2–3 · Surge Prediction Launch

The Differentiating Feature

→ DAU growing organically 18% week-over-week through sharing

✓Surge prediction launched; initially beta-gated to 20% of users
✓In-app accuracy feedback: users reported 88% found it useful or very useful
✓Wait/Book feature launched and became most-used non-comparison feature

→ DAU growing organically 18% week-over-week through sharing

Month 4–6 · Scale

4 New Cities + Commute Routes

→ 80K MAU. ₹3.2L avg monthly user savings across active users. D30 retention: 38%.

✓Mumbai, Hyderabad, Chennai, Pune added
✓Saved commute routes launched — adopted by 61% of daily users within 4 weeks
✓80,000 monthly active users; primarily organic

→ 80K MAU. ₹3.2L avg monthly user savings across active users. D30 retention: 38%.

Results

Six-Month Numbers

80K

Monthly Active Users

organic growth only

38%

D30 Retention

habit-grade engagement

88%

Surge Prediction Accuracy

per user exit survey

₹3.2L

Monthly User Savings

across active user base

The metric I'm most proud of: 80,000 users who came back every month without a single rupee of paid acquisition. Every user came because someone told them about it, or because they searched for "cheapest cab in Bangalore" and we showed up. That's product-market fit in its clearest form.

Insights

What Built the Habit

Saved routes are retention mechanisms, not convenience features

Users who saved commute routes had 3.2× better D30 retention than users who didn't. The route save wasn't a convenience — it was a commitment to return. The product held something they'd invested in setting up.

Prediction accuracy above 80% is the threshold for behavioral trust

Below 80%, users tried the surge prediction and filed it as 'interesting but unreliable.' Above 80%, they started making booking decisions based on it — waiting when told to wait, booking when told to book. Trust threshold: 80%.

Real-time refresh cadence directly correlates with user trust

When we experimented with 5-minute refresh (to reduce API costs), user complaints about 'wrong prices' increased significantly. When we moved back to 90-second refresh, complaints fell. Accuracy and recency are inseparable in pricing contexts.

Failures

Where We Got It Wrong

✕ First prediction model used city-level surge, not route-level

Why it broke

Initial LSTM trained on city-wide average surge data. Bangalore as a whole doesn't surge — Koramangala surges while Whitefield might not. City-level predictions were often directionally wrong.

What fixed it

Rebuilt model with 840 route-pair specific models. Each popular origin-destination pair gets its own surge model trained on that route's specific historical data. Prediction accuracy: 61% → 78%.

✕ Weekend surge patterns were different; model didn't know

Why it broke

The initial model was trained and performed well on weekday commute patterns. Weekend surge is driven by entertainment venues, late nights, weather events — completely different pattern. Weekend predictions were poor.

What fixed it

Separate weekday/weekend model training. Weekend model includes proximity to entertainment districts, historical event data, late-night surge decay curves. Weekend prediction accuracy improved from 54% to 74%.

Future

The Roadmap

Auto-Book Integration

When a user sets a fare alert and the threshold is hit, auto-initiate the booking in the preferred app — eliminating the step between notification and action where surge can change again.

Tier 2 City Expansion

Extending to Tier 2 cities where Rapido and local platforms have gained significant share. The comparison value is higher in markets where platform fragmentation is greater.

Monthly Savings Report

Personal savings report: 'This month you saved ₹2,340 by using our comparison tool.' Real, attributable savings that reinforce the habit and create referral motivation.

The auto driver who started the conversation saved ₹600/week by switching platforms based on surge patterns. He didn't use our app — but the 80,000 people who did were solving the same problem he described in that traffic jam.

How We Built a Zero-Cost Growth Engine Using Organic + Product Hooks

Growth

Designing a Comparison Platform Like Monito for the Indian Market

Product

Creating a Scalable Notification System for 1M+ Users (Trigger-Based)

9 min read

Product

How We Built a Multi-Calculator Finance App That Drives Daily Retention

8 min read

Product

Building a CRM That Handles 180K+ Monthly Transactions at 99.6% Uptime

12 min read

All Case Studies

Browse all 30 breakdowns

Work With Me

Let's build something together

Connect on LinkedIn

Follow for more insights

Where to go next

Marketplace & Comparison Platform Consultant
The consulting work this sits inside, with the rest of the evidence behind it.
Install-to-purchase conversion benchmark, travel
Published figures, traced to their sources — and where the sources disagree.
App store conversion rate benchmark, medical & healthcare
Published figures, traced to their sources — and where the sources disagree.
Product Growth Score
Thirteen questions that locate which part of the funnel is actually constraining growth.

Building a Real-Time Cab Fare Comparison Engine (Uber vs Ola vs InDrive)

Results at a glance

The Auto-Rickshaw Insight That Started Everything

What Users Were Doing Without a Comparison Tool

Five Types of Cab Users

Comparison Isn't Enough — Prediction Is the Differentiator

The Engine Design

The Data Layer That Made It Work

Growth Without Paid Acquisition

Six-Month Numbers

What Built the Habit

Saved routes are retention mechanisms, not convenience features

Prediction accuracy above 80% is the threshold for behavioral trust

Real-time refresh cadence directly correlates with user trust

Where We Got It Wrong

The Roadmap

You might also like

Where to go next

The Auto-Rickshaw Insight That Started Everything

What Users Were Doing Without a Comparison Tool

Five Types of Cab Users

Comparison Isn't Enough — Prediction Is the Differentiator

The Engine Design

The Data Layer That Made It Work

Growth Without Paid Acquisition

Six-Month Numbers

What Built the Habit

Saved routes are retention mechanisms, not convenience features

Prediction accuracy above 80% is the threshold for behavioral trust

Real-time refresh cadence directly correlates with user trust

Where We Got It Wrong

The Roadmap