Back to selected work

Growing MX Games from zero to 500K daily users in 18 months

How an intern doing sports content ended up running the games pipeline, catching a fraud ring through the data, and learning to give users what they actually wanted.

Listen to this story 5:30

Role: Program Executive, promoted to Associate Product Manager
Where: Amazon MX Player, Mumbai
When: 2020 – 2021
Team: About 10 people, 3 reporting to me

500Kdaily users in 18 months
30–45minutes a day, average time spent
35%of revenue recovered from fraud
145games tested and launched

Team chat, 29 Aug 2021: "Waiting for midnight" and "The way you and your team are beating nails each hour, I don't think we need to wait till midnight" — 29–30 August 2021: our team chat, "Mission 500k", as we crossed 500K daily users. Names and numbers blurred.

Team chat: "500 ke chakkar mein 600 na aajaye" (we might hit 600 while chasing 500) and "This is a good problem to have!" — 29–30 August 2021: our team chat, "Mission 500k", as we crossed 500K daily users. Names and numbers blurred.

The bet

MX Player was one of India's biggest streaming apps, with over 200 million monthly users. In early 2020 it bet on hyper-casual games: short, simple games that could raise time spent, improve retention and pull new people into the streaming side. Then COVID hit, and mobile gaming in India exploded.

I had joined in 2019 as an intern, programming sports content for the streaming app, and had taken over all of sports streaming on my own. When MX Games launched, the product team asked for me. I worked directly with the Head of Product and the Head of Games, and was promoted from Program Executive to Associate Product Manager on performance alone.

Auro standing outside the MX Player office in Mumbai — Outside the MX Player office in Mumbai.

“I saw him grow from being an individual contributor to managing a team of 2 junior associates under him in a short span of time.”

“He also took up additional responsibility proactively that contributed to taking the team forward and was rewarded with a fast track promotion to Associate Product Manager.”

Satyajit SwainHead of Games, MX Player. Managed Auro directly.

“I have witnessed Arabinda's transformation from an intern to a product manager at MX Player. He is extremely hardworking and has an out-of-the-box approach to solving complex problems.”

Rishabh KumarDirector of Product Management, Data, MX Player

From recommendations on my LinkedIn profile.

Hard part one: getting good games

Building quality games in-house was slow and expensive. So we bought games from small studios, either outright or on revenue share. I ran that pipeline end to end: finding studios, testing every game by hand, picking the best and getting them live. Over my time there, 145 hyper-casual games went live this way. As the catalogue grew, I took on three associates to keep up, and automated the repetitive work: setting up tournaments, reloading prize budgets and other daily operations. That cut our CRM costs by 67%.

Air Lift — A slice of the catalogue. Every game was tested by hand before it went live.

Aqua Bricks — A slice of the catalogue. Every game was tested by hand before it went live.

Hard part two: a fraud ring on the leaderboards

MX Games ran on rewards. The top ranks on each game's leaderboard won small cash prizes or MX coins, which unlocked ad-free viewing and new series. Then cash-outs started surging, and they didn't match the rise in plays or time spent.

Cricket 2020: play and win ₹1,00,000 — Every game ran on cash-prize tournaments. That was the hook, and the reason cheating paid.

Guns & Bottles: play and win ₹1,00,000 — Every game ran on cash-prize tournaments. That was the hook, and the reason cheating paid.

I went through the data every day. Some accounts had one or two plays and impossible scores. In one game, Airlift, first place had 10,000 points while second place had about 500. We ruled out bugs, saw bot-like patterns, then went looking where cheaters talk. On YouTube and Telegram we found tutorials showing how to hack MX Games with Android settings and screen-size tricks.

A typical Air Lift leaderboard during the fraud, top four scores

First place was 20 times second place, after just one or two plays.

Air Lift splash screen — Air Lift, the game where first place scored 10,000 and second scored about 500.

Air Lift rules screen: slide the ball to clear obstacles and keep the balloon safe — Air Lift, the game where first place scored 10,000 and second scored about 500.

We couldn't change the games' code, and the 5 to 10 hackable games out of roughly 100 were also our biggest for plays and ad revenue. So we fixed it in two steps:

Stop the bleeding. I owned the prize budgets, and fraud had sent spend ballooning. We cut the prize pool roughly in half. The hard part was doing that without losing honest players, so we kept tournaments running and watched engagement daily while the fix was built. Engagement dipped at first. Working with the performance marketing team, I ran carefully timed notifications and targeted campaigns, and it recovered and stabilised within two weeks.
Fix it properly. We designed a policy around the highest score a real player could reach in each game, and our backend engineers worked out those limits and built the checks. Anyone above the limit was disqualified automatically and warned. A second offence meant a ban.

Score submitted
Above the game’s realistic maximum?
NoCounts on the leaderboard

YesDisqualified and warnedSecond time: banned

The rule that replaced chasing cheaters one by one.

That recovered 35% of revenue that had been leaking to fraud.

What actually drove growth

Daily active users, launch to August 2021

Approximate milestones.

Listening to users, not our assumptions

We assumed players wanted polished, graphics-heavy games. Surveys and data said otherwise. Most players were in tier 2 and tier 3 towns across the north, opening a game in a short break from work. They wanted something over in three or four minutes. We shifted almost all our acquisition toward quick hyper-casual games.

Rewards worth coming back for

On top of cash prizes and coins, we partnered with Zomato for coupons, giving players a reason to return daily.

Using the streaming app's reach

MX Player already had 200 million users. I ran our notification journeys on CleverTap, sending targeted messages to specific groups at the right times, with strong thumbnails and copy. That brought in new players and kept existing ones coming back.

Games built around live sport

Two games were my own ideas, built for the IPL season. One was a score-prediction game where players picked a final score on a grid, played for virtual coins rather than cash. The other asked a question every over, updated automatically in real time.

Where retention landed

About 25% of new players came back the next day, 10–13% were still playing after a week, and 5–7% after a month.

Share of new players still playing, by days since install

What didn't work

Single-player Ludo was a huge hit. So we tried a multiplayer version, where two friends in different cities could play each other. The infrastructure was far heavier than we expected, and it pulled the team away from the hyper-casual games that were driving growth. We stepped back from it.

MX Ludo banner — MX Ludo. The single-player game was a hit; the multiplayer version is what we stepped back from.

The lesson I took: when one thing is clearly working, double down until it runs on autopilot. Only then spread out.

What I'd take from it

The data speaks before anyone else does.

Nobody reported the fraud. It showed up as cash-outs outrunning plays, in numbers I read every morning. By the time a problem is a complaint, it has already cost you.

Build for the user you have, not the one you imagine.

We pictured gamers who wanted polish. Our players were in small towns across UP and Bihar, playing for four minutes between tasks. The simple games won.

Fix the system, not the symptom.

Chasing cheaters one by one never ends. One rule, a realistic score ceiling for each game, ended it without touching a line of game code and recovered 35% of revenue.

Every incentive is a game someone will try to win.

Cash prizes brought players back every day, and they attracted the people gaming us. Design the reward and the guardrail together.

Protect what's working before chasing what's next.

Single-player Ludo was a hit. Multiplayer pulled the team off the engine that was driving growth. Double down until it runs on autopilot, then expand.

These are the instincts I bring to building with AI today: read the data, design for real users, and fix the system.

Back to selected work