Audience Platform: Building the Growth Rails Junglee Ran On

The problem

Junglee's seven game teams ran their marketing and lifecycle stack on a patchwork of tools rather than one platform. CleverTap (adopted shortly before I joined, replacing an earlier tool) handled segmentation and communication, but nothing in that stack did gamification — it had to be built in-house from nothing. CleverTap also came with real limits of its own: it capped A/B testing at the industry-standard 4 variants, and, like most third-party tools, it worked on data exported to it rather than the games' own databases directly.

Approach

After user research and competitive analysis, I built the roadmap to consolidate that fragmented stack onto one in-house platform — not a like-for-like clone of any vendor, but a fully extensible architecture designed to support any custom marketing interface, so new capability could be added without hitting the same ceiling again. I owned the lifecycle end-to-end: requirements gathering, development including vendor migration, deployment, and adoption across marketing, product marketing, and product teams on all seven games.

What shipped

• Consolidated segmentation, communication, and gamification — split across CleverTap and no tool at all — onto one platform, saving ₹2 Crores annually in vendor costHard
• An A/B testing framework supporting an unlimited number of variants — against an industry standard of 4 — running ~50 concurrent tests at any time across 2,000+ campaigns a monthHard
• Initial SaaS-style platform revamp, optimizing UI layouts in the most-used components to reduce mistakes and drive adoptionProjected — 30% adoption increase

What made Audience different from a vendor tool

The advantage over CleverTap wasn't just cost — it was direct access to the games' own internal databases, which made segmentation meaningfully easier than working through a vendor's export/import layer. I later extended this further by building MongoDB-backed (Audience DB) query-based segmentation on top of the platform's existing attribute-based segmentation. The two complement each other: attributes capture lifetime aggregatable or replaceable data per event, while queries pull the exact cut or shape a campaign manager actually needs — a distinction that matters once you're running thousands of campaigns a month across seven very different games. That query layer, plus personalized segmentation across multiple databases, saved analysts roughly an hour a day that had previously gone into manual data-pullingHard — 1 hour/day/analyst.

Audience was also structured differently at the campaign level. A single campaign could contain sub-campaigns — for example, running push, SMS, and in-app messaging under one campaign with separate comm IDs per channel, or running multiple segments under the same campaign — which made it far easier for the data team to query and pull results, since everything rolled up under one campaign identity instead of being scattered across disconnected sends. Audience could also handle reward distribution directly — including instant cash payouts in a real-money gaming context — a capability outside what a tool like CleverTap could do at all.

The database access advantage showed up directly in results, too. Building events for cross-platform data integration — connecting datastores that had previously been siloed — let campaigns target users with more precision, and boosted a card game's revenue share from 17% to 31% within HowzatHard. On the operations side, adding block cloning to the funnel-builder cut funnel and mission creation time by 65% across multiple teamsHard — the kind of unglamorous tooling work that compounds across 2,000+ campaigns a month.

The risk and reliability work

Owning a platform that 120+ people used to reach 100mn+ players meant the interesting product decisions weren't just growth features — several of the most consequential ones were about preventing the platform from hurting the business it was meant to grow.

Maker Checker Flow. Multiple incidents forced this: an SMS pipeline overload, and separately a campaign that went out to 1.7 crore users on the wrong dates, which could have cost roughly ₹17 lakhs had it not been caught. A first fix — calculating user-base size before a campaign saved, with a 5-lakh-plus threshold — got rolled back for being too slow at query time. The system that stuck used per-game user-count and cost thresholds, requiring managerial approval on high-impact campaigns before they could fireProjected — ₹40+ lakhs/quarter.

Role-Based Access Control. A fraud incident — an employee moved ₹32+ lakhs into a game wallet using a preview option not meant for that — forced a full access-control revamp: roles expanded from 3 (Viewer, Creator, Admin) to 5, every existing user's role was reassigned, and login moved from username/password to SSORisk mitigated — ₹30+ lakhs.

Communication Categories. Users were receiving up to 14 communications a day across marketing, product marketing, and product teams — a spam problem made worse by Howzat's super-app structure hosting multiple games at once. I designed a tagging system that throttled communications per user per category per dayHard — ₹5 lakhs/month + NPS increase.

Product Sanity Alerts. Roughly 50 automated alerts for proactive issue detection, catching underperforming or misconfigured campaigns before they burned budget or leaked revenueHard — 32% improvement in Campaign Delivery Success.

The gamification progression

Timeline Missions (see the full case study) wasn't the first gamification primitive I built on Audience — it was the last in a sequence, each one building on the last: Experiments in Quests — the platform's unlimited-variant experimentation framework, already running for campaigns, funnels, and segments (feeding into Zookeeper for offers and UI), extended to cover quests — then Mission Nudges — context-aware, automated messaging timed to specific mission stages rather than generic remindersHard — 40% MACU engagement, 20% MACU completion — then Hidden Missions, a control-group primitive that assigns real missions to real users without ever surfacing them, giving a true behavioral counterfactual instead of a pre/post comparison. Timeline Missions reused both: Mission Nudges' events drove its day-gating logic, and Hidden Missions became the control group for measuring its impact.

Platform reliability and team

I guided a team of 4 developers and mentored 2 associate product managers, serving 120+ active platform users across seven game teams with almost no documentation or formal training in the platform's early days. I reduced user-reported issues by 90% through targeted training, cross-team mediation, and advocating for the user's point of view directly in development — becoming, informally, the go-to person for Audience as new teams (a new Poker team, entirely reliant on the platform) came onlineHard. That was recognized with a Top Rookie award, cited for agility, precise requirement detailing, and consistent follow-through on issues raised across the platform.